Neural network based in-loop filter architecture for video coding

Abstract

A video encoder and a video decoder are configured to perform neural network (NN)-based filtering. The video encoder and video decoder can receive images of video data, reconstruct images of the video data, and perform an NN-based filtering process on one or more blocks of the reconstructed images of the video data using NN-based filters. The NN-based filters include a pair of backbone blocks, each of which includes a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

Description

[0001] This application claims priority to U.S. Patent Application No. 18 / 984,088, filed December 17, 2024, and U.S. Provisional Patent Application No. 63 / 612,279, filed December 19, 2023, the entire contents of each of which are incorporated herein by reference. U.S. Patent Application No. 18 / 984,088, filed December 17, 2024, claims the benefit of U.S. Provisional Patent Application No. 63 / 612,279, filed December 19, 2023. Technical Field

[0002] This disclosure relates to video encoding and video decoding. Background Technology

[0003] Digital video capabilities can be incorporated into a wide variety of devices, including digital televisions, digital live broadcast systems, wireless broadcasting systems, personal digital assistants (PDAs), laptops or desktop computers, tablets, e-book readers, digital cameras, digital recording devices, digital media players, video game devices, video game consoles, cellular or satellite wireless phones (so-called "smartphones"), video conferencing equipment, video streaming devices, and more. Digital video devices implement video decoding technologies, such as those defined by MPEG-2, MPEG-4, ITU-T H.263, ITU-T H.264 / MPEG-4 (Part 10, Advanced Video Decoding (AVC)), ITU-T H.265 / High Efficiency Video Decoding (HEVC), ITU-T H.266 / Variety Video Decoding (VVC) and extensions to these standards, as well as proprietary video codecs / formats such as AOMedia Video1 (AV1) developed by the Open Media Alliance. By implementing such video decoding technologies, video devices can more efficiently send, receive, encode, decode, and / or store digital video information.

[0004] Video decoding techniques include spatial (intra-picture) prediction and / or temporal (inter-picture) prediction to reduce or remove redundancy inherent in video sequences. For block-based video decoding, video slices (e.g., video pictures or portions of video pictures) can be divided into video blocks, which may also be referred to as decoding tree units (CTUs), decoding units (CUs), and / or decoding nodes. Video blocks in a slice after intra-frame decoding (I) of a picture are encoded using spatial prediction relative to reference samples in adjacent blocks within the same picture. Video blocks in a slice after inter-frame decoding (P or B) of a picture can use spatial prediction relative to reference samples in adjacent blocks within the same picture or temporal prediction relative to reference samples in other reference pictures. A picture may be referred to as a frame, and a reference picture may be referred to as a reference frame. Summary of the Invention

[0005] Generally speaking, this disclosure describes techniques for video encoding and decoding. Specifically, this disclosure describes methods, techniques, and structures that can reduce the computational complexity and memory bandwidth requirements of neural network (NN)-based video decoding tools. In some examples, the NN-based decoding tool can be a video decoding tool based on convolutional neural networks (CNNs) (such as CNN-based filters).

[0006] In some examples, this disclosure describes a CNN-based in-loop filtering architecture for video decoding. The techniques disclosed herein can improve decoding performance under constraints of complexity and memory requirements. Decoding performance improvements can be achieved for architectures with low and very low complexity by introducing multi-scale elements into the CNN architecture. Complexity reductions can be achieved for the architecture by using separable convolutions, or one-dimensional convolutions, instead of multi-dimensional convolutions.

[0007] In one example, this disclosure describes a neural network (NN)-based filter architecture in which the backbone blocks are configured as pairs of backbone blocks, each pair of backbone blocks comprising a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction. The complexity of the NN-based filter can be reduced with minimal performance degradation by using feature channel reduction. In another example, the NN-based filter architecture of this disclosure includes a fusion block having two separable convolutions, each of which applies spatial downsampling. This spatial downsampling reduces the complexity of the NN-based filter with minimal performance degradation.

[0008] In one example, this disclosure describes a method for decoding video data, the method comprising: receiving an image of video data; reconstructing the image of video data; and performing a neural network-based filtering process on one or more blocks of the reconstructed image of video data using a neural network-based filter, wherein the neural network-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0009] In another example, this disclosure describes an apparatus configured to decode video data, the apparatus comprising: a memory and one or more processors in communication with the memory, the one or more processors being configured to receive a picture of the video data, reconstruct the picture of the video data, and perform an NN-based filtering process on one or more blocks of the reconstructed picture of the video data using an NN-based filter, wherein the NN-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0010] In another example, this disclosure describes an apparatus configured to decode video data, the apparatus comprising: components for receiving an image of the video data; components for reconstructing the image of the video data; and components for performing an NN-based filtering process on one or more blocks of the reconstructed image of the video data using an NN-based filter, wherein the NN-based filter comprises a pair of backbone blocks, each of the backbone blocks comprising a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0011] In another example, this disclosure describes a non-transitory computer-readable storage medium storing instructions that, when executed, cause one or more processors of a device configured to decode video data to decode the video data to receive a picture of the video data, reconstruct the picture of the video data, and perform a NN-based filtering process on one or more blocks of the reconstructed picture of the video data using a NN-based filter, wherein the NN-based filter includes a pair of backbone blocks, each of the backbone blocks including a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0012] Details of one or more examples are set forth in the accompanying drawings and the following description. Other features, objects, and advantages will be apparent from the description, drawings, and claims. Attached Figure Description

[0013] Figure 1 This is a block diagram illustrating an example video encoding and decoding system that can perform the techniques of this disclosure.

[0014] Figure 2 This is a block diagram illustrating an example of a hybrid video decoding framework.

[0015] Figure 3 This is a conceptual diagram illustrating a hierarchical prediction structure using a group of pictures (GOP) of size 16.

[0016] Figure 4This is a block diagram illustrating an example of a four-layer convolutional neural network (CNN) based filter.

[0017] Figure 5 This is a block diagram illustrating an example CNN-based filter with padded input samples and supplementary data.

[0018] Figure 6 This is a block diagram illustrating another example of a CNN-based filter with padded input samples and supplementary data.

[0019] Figure 7 This is an example Figure 6 A block diagram of an example attention residual block for a neural network filter.

[0020] Figure 8 This is an example Figure 6 A block diagram of an example of a spatial attention layer.

[0021] Figure 9 This is a block diagram illustrating an example of a simplified CNN-based filter architecture using a residual block structure, with padded input samples and supplementary data.

[0022] Figure 10 This is an example Figure 9 Here is a block diagram of an example residual block structure.

[0023] Figure 11 This is a block diagram illustrating an example of a simplified CNN-based filter architecture that uses a filter block structure and has padding input samples and supplementary data.

[0024] Figure 12 This is an example Figure 11 Here is a block diagram of an example filter block structure.

[0025] Figure 13 This is a block diagram illustrating various examples of multidimensional convolution solutions.

[0026] Figure 14 This is a block diagram illustrating another example of a multidimensional convolution solution.

[0027] Figure 15 This is a block diagram illustrating an example of a multi-scale feature extraction backbone network with two-component convolutions.

[0028] Figure 16A This is a block diagram illustrating an example of a unified filter with a joint model (joint luminance and chrominance).

[0029] Figure 16B This is an example Figure 16A A block diagram of an example backbone block.

[0030] Figure 17This is a block diagram illustrating an example of a unified filter with separate luminance / chrominance models (luminance).

[0031] Figure 18 This is a block diagram illustrating an example of a unified filter with separate luminance / chrominance models (chrominance).

[0032] Figure 19A This is a block diagram illustrating an example of a low-complexity in-loop filter architecture.

[0033] Figure 19B This is an example Figure 19A A block diagram of an example backbone block.

[0034] Figure 19C This is an example used for Figure 19A A flowchart illustrating an example CP decomposition of a 3×3 convolution.

[0035] Figure 20A This is a block diagram illustrating an example of a low-complexity in-loop filter architecture with luminance / chrominance separation.

[0036] Figure 20B Based on one or more aspects of this disclosure Figure 20A Example trunk block.

[0037] Figure 21A This is a block diagram illustrating the first example of a trunk block pair.

[0038] Figure 21B This is a block diagram illustrating the second example of a trunk block pair.

[0039] Figure 21C This is an example used for Figure 21A A block diagram showing different arrangements of 1×1 convolution.

[0040] Figure 22A This is a block diagram illustrating another example of a backbone block pair with an activation layer, representing a first example arrangement.

[0041] Figure 22B This is a block diagram illustrating another example of a backbone block pair with an activation layer, representing a second example arrangement.

[0042] Figure 22C This is a block diagram illustrating another example of a backbone block pair with a third example arrangement having multiple activation layers.

[0043] Figure 23A This is a block diagram illustrating a first example of a concrete implementation of the fusion block.

[0044] Figure 23B This is a block diagram illustrating a second example of a concrete implementation of the fusion block.

[0045] Figure 23CThis is a block diagram illustrating a third example of a concrete implementation of the fusion block.

[0046] Figure 23D This is a block diagram illustrating the fourth example of a concrete implementation of the fusion block.

[0047] Figure 23E This is a block diagram illustrating the fifth example of a concrete implementation of the fusion block.

[0048] Figure 24 This is a block diagram illustrating an example video encoder that can perform the techniques of this disclosure.

[0049] Figure 25 This is a block diagram illustrating an example video decoder that can perform the techniques of this disclosure.

[0050] Figure 26 This is a flowchart illustrating an example method for encoding the current block according to the technology of this disclosure.

[0051] Figure 27 This is a flowchart illustrating an example method for decoding the current block according to the technology of this disclosure.

[0052] Figure 28 This is a flowchart illustrating an example method for decoding the current block according to the technology of this disclosure. Detailed Implementation

[0053] Video coding is typically a lossy process. For example, during video coding, blocks of video data are encoded using quantization and transform. Generally, quantization of values ​​involves reducing several least significant bits of those values, which is usually irreversible. This bit reduction is generally performed in a way that avoids detectable loss. However, sometimes such loss can lead to detectable artifacts in the video data, such as block artifacts.

[0054] Filtering can be applied to decoded and / or reconstructed video data to enhance it, thus improving the output video data. For example, filtering can compensate for block artifacts or other losses in video data. Research has shown that neural network (NN)-based filtering techniques are highly capable of improving decoded and / or reconstructed video data. NN-based filtering techniques can be highly complex and require significant processing power to execute effectively.

[0055] This disclosure describes decoding performance improvement techniques applicable to neural network-based filtering. The application of these techniques can reduce the processing required by one or more processors to perform neural network-based filtering. In this way, the techniques of this disclosure can improve the performance of video decoding devices. Similarly, these techniques enable more devices to perform neural network-based filtering, thereby substantially improving the field of video decoding.

[0056] In one example, this disclosure describes a neural network (NN)-based filter architecture in which backbone blocks are configured as pairs of backbone blocks, each pair of backbone blocks comprising a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction. The complexity of the NN-based filter can be reduced with minimal performance degradation by using feature channel reduction. In another example, the NN-based filter architecture of this disclosure includes a fusion block having two separable convolutions, each of which applies spatial downsampling. This spatial downsampling reduces the complexity of the NN-based filter with minimal performance degradation.

[0057] In one example, a video encoder and a video decoder may be configured to receive an image of video data, reconstruct the image of video data, and perform a neural network-based filtering process on one or more blocks of the reconstructed image of video data using a neural network-based filter, wherein the neural network-based filter includes a pair of backbone blocks, each of the backbone blocks including a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0058] Figure 1 This is a block diagram illustrating an example video encoding and decoding system 100 capable of performing the techniques of this disclosure. The techniques of this disclosure generally relate to decoding (encoding and / or decoding) video data. Generally speaking, video data includes any data used for processing video. Thus, video data may include unencoded raw video, encoded video, decoded (e.g., reconstructed) video, and video metadata, such as signaling data.

[0059] like Figure 1 As shown, in this example, system 100 includes a source device 102 that provides encoded video data to be decoded and displayed by a destination device 116. Specifically, source device 102 provides video data to destination device 116 via computer-readable medium 110. Source device 102 and destination device 116 can be or may include any of a wide range of devices, such as desktop computers, laptop computers, mobile devices, tablet computers, set-top boxes, mobile phones (such as smartphones), televisions, cameras, display devices, digital media players, video game consoles, video streaming devices, broadcast receiver devices, etc. In some cases, source device 102 and destination device 116 may be configured for wireless communication and are therefore referred to as wireless communication devices.

[0060] exist Figure 1In the example, source device 102 includes a video source 104, memory 106, video encoder 200, and output interface 108. Destination device 116 includes an input interface 122, video decoder 300, memory 120, and display device 118. According to this disclosure, the video encoder 200 of source device 102 and the video decoder 300 of destination device 116 can be configured to apply techniques for filtering video data. Therefore, source device 102 represents an example of a video encoding device, while destination device 116 represents an example of a video decoding device. In other examples, the source device and destination device may include other components or arrangements. For example, source device 102 may receive video data from an external video source such as an external camera. Similarly, destination device 116 may interface with an external display device instead of including an integrated display device.

[0061] like Figure 1 The system 100 shown is merely an example. Generally, any digital video encoding and / or decoding device can perform techniques for filtering video data. Source device 102 and destination device 116 are merely examples of such decoding devices, where source device 102 generates decoded video data for transmission to destination device 116. This disclosure refers to a “decoding” device as a device that performs the decoding (e.g., encoding and / or decoding) of data. Thus, video encoder 200 and video decoder 300 represent examples of decoding devices, specifically, a video encoder and a video decoder, respectively. In some examples, source device 102 and destination device 116 may operate in a substantially symmetrical manner, such that each of source device 102 and destination device 116 includes video encoding and decoding components. Therefore, system 100 may support one-way or two-way video transmission between source device 102 and destination device 116, for example, for video streaming, video playback, video broadcasting, or video telephony.

[0062] Generally, video source 104 represents the source of video data (i.e., unencoded raw video data) and provides a sequential series of pictures (also referred to as "frames") of video data to video encoder 200, which encodes the data for the pictures. Video source 104 of source device 102 may include video capture devices such as cameras, video archives containing previously captured raw video, and / or video feed interfaces for receiving video from video content providers. Alternatively, video source 104 may generate computer graphics-based data as source video, or a combination of live video, archived video, and computer-generated video. In each case, video encoder 200 encodes the captured, pre-captured, or computer-generated video data. Video encoder 200 may rearrange the pictures from the received order (sometimes referred to as "display order") to a decoding order for decoding. Video encoder 200 may generate a bitstream comprising the encoded video data. Then, the source device 102 can output the encoded video data to the computer-readable medium 110 via the output interface 108 for reception and / or retrieval by, for example, the input interface 122 of the destination device 116.

[0063] The memory 106 of source device 102 and the memory 120 of destination device 116 represent general-purpose memory. In some examples, memories 106 and 120 may store raw video data, such as raw video from video source 104 and raw decoded video data from video decoder 300. Additionally or alternatively, memories 106 and 120 may store software instructions executable by, for example, video encoder 200 and video decoder 300. Although memories 106 and 120 are shown separately from video encoder 200 and video decoder 300 in this example, it should be understood that video encoder 200 and video decoder 300 may also include internal memory for functionally similar or equivalent purposes. Furthermore, memories 106 and 120 may store encoded video data, such as output from video encoder 200 and input to video decoder 300. In some examples, portions of memories 106 and 120 may be allocated as one or more video buffers, for example, to store raw decoded and / or encoded video data.

[0064] Computer-readable medium 110 may represent any type of medium or device capable of transmitting encoded video data from source device 102 to destination device 116. In one example, computer-readable medium 110 represents a communication medium that enables source device 102 to directly transmit encoded video data to destination device 116 in real time, for example, via a radio frequency network or a computer-based network. According to a communication standard such as a wireless communication protocol, output interface 108 may modulate the transmitted signal including the encoded video data, and input interface 122 may demodulate the received transmitted signal. The communication medium may include any wireless or wired communication medium, such as radio frequency (RF) spectrum or one or more physical transmission lines. The communication medium may form part of a packet-based network such as a local area network, a wide area network, or a global network (such as the Internet). The communication medium may include a router, switch, base station, or any other equipment that may be useful for facilitating communication from source device 102 to destination device 116.

[0065] In some examples, source device 102 can output encoded data from output interface 108 to storage device 112. Similarly, destination device 116 can access encoded data from storage device 112 via input interface 122. Storage device 112 may include any data storage medium of various distributed or locally accessed data storage media, such as hard disk drives, Blu-ray discs, DVDs, CD-ROMs, flash memory, volatile or non-volatile memory, or any other suitable digital storage medium for storing encoded video data.

[0066] In some examples, source device 102 can output encoded video data to file server 114 or another intermediate storage device that can store the encoded video data generated by source device 102. Destination device 116 can access the stored video data from file server 114 via streaming or download.

[0067] File server 114 can be any type of server device capable of storing encoded video data and sending the encoded video data to destination device 116. File server 114 may represent a web server (e.g., for a website), a server configured to provide file transfer protocol services (such as File Transfer Protocol (FTP) or FLUTE-based file delivery protocol), a Content Delivery Network (CDN) device, a Hypertext Transfer Protocol (HTTP) server, a Multimedia Broadcast Multicast Service (MBMS) or Enhanced MBMS (eMBMS) server, and / or a Network Attached Storage (NAS) device. File server 114 may additionally or alternatively implement one or more HTTP streaming protocols, such as HTTP-based Dynamic Adaptive Streaming (DASH), HTTP Live Streaming (HLS), Real-Time Streaming Protocol (RTSP), HTTP Dynamic Streaming, etc.

[0068] Destination device 116 can access encoded video data from file server 114 via any standard data connection, including an internet connection. This may include a wireless channel (e.g., Wi-Fi connection), a wired connection (e.g., digital subscriber line (DSL), cable modem, etc.), or a combination of both, suitable for accessing encoded video data stored on file server 114. Input interface 122 can be configured to operate according to any or more of the various protocols discussed above for retrieving or receiving media data from file server 114, or other such protocols for retrieving media data.

[0069] Output interface 108 and input interface 122 can represent a wireless transmitter / receiver, a modem, a wired networking component (e.g., an Ethernet card), a wireless communication component operating according to any of the various IEEE 802.11 standards, or other physical components. In examples where output interface 108 and input interface 122 include wireless components, output interface 108 and input interface 122 can be configured to transmit data, such as encoded video data, according to cellular communication standards such as 4G, 4G-LTE (Long Term Evolution), Advanced LTE, 5G, etc. In some examples where output interface 108 includes a wireless transmitter, output interface 108 and input interface 122 can be configured to comply with other wireless standards such as the IEEE 802.11 specification, the IEEE 802.15 specification (e.g., ZigBee), etc. ™ ),Bluetooth ™Standards are used to transmit data, such as encoded video data. In some examples, source device 102 and / or destination device 116 may include corresponding system-on-chip (SoC) devices. For example, source device 102 may include an SoC device for performing functions belonging to video encoder 200 and / or output interface 108, and destination device 116 may include an SoC device for performing functions belonging to video decoder 300 and / or input interface 122.

[0070] The technology disclosed herein can be applied to video decoding to support any multimedia application in a variety of multimedia applications, such as over-the-air television broadcasting, cable television transmission, satellite television transmission, internet streaming video transmission (such as HTTP-based Dynamic Adaptive Streaming (DASH)), digital video encoded onto data storage media, decoding of digital video stored on data storage media, or other applications.

[0071] The input interface 122 of the destination device 116 receives an encoded video bitstream from a computer-readable medium 110 (e.g., a communication medium, storage device 112, file server 114, etc.). The encoded video bitstream may include signaling information defined by the video encoder 200 and also used by the video decoder 300, such as syntax elements having values ​​describing the characteristics and / or processing of video blocks or other decoded units (e.g., slices, pictures, picture groups, sequences, etc.). The display device 118 displays a decoded picture of the decoded video data to the user. The display device 118 may represent any of a variety of display devices, such as a liquid crystal display (LCD), a plasma display, an organic light-emitting diode (OLED) display, or another type of display device.

[0072] Despite Figure 1Not shown, but in some examples, both the video encoder 200 and the video decoder 300 may be integrated with the audio encoder and / or audio decoder (e.g., audio codec), and may include appropriate MUX-DEMUX units or other hardware and / or software to process multiplexed streams that include both audio and video in a common data stream. Example audio codecs may include AAC, AC-3, AC-4, ALAC, ALS, AMBE, AMR, AMR-WB (G.722.2), AMR-WB+, aptX (various versions), ATRAC, BroadVoice (BV16, BV32), CELT, Enhanced AC-3 (E-AC-3), EVS, FLAC, G.711, G.722, G.722.1, G.722.2 (AMR-WB), G.723.1, G.726, G.728, G.729, G.729.1, GSM-FR, HE-AAC, iLBC, iSAC, LA Lyra, Monkey's Audio, MP1, MP2 (MPEG-1, 2 Audio Layer II), MP3, Musepack, Nellymoser Asao, OptimFROG, Opus, Sac, Satin, SBC, SILK, Siren 7, Speex, SVOPC, True Audio (TTA), TwinVQ, USAC, Vorbis (Ogg), WavPack and Windows Media Aud.

[0073] Both the video encoder 200 and the video decoder 300 can be implemented as any of a variety of suitable encoder and / or decoder circuits comprising a processing system, such as one or more microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), discrete logic components, software, hardware, firmware, or any combination thereof. When the technology is partially implemented in software, the device may store instructions for the software in a suitable non-transitory computer-readable medium and execute the instructions in hardware using one or more processors to perform the technology of this disclosure. Each of the video encoder 200 and the video decoder 300 may be included in one or more encoders or decoders, either of which may be integrated as part of a combined encoder / decoder (CODEC) in the respective device. Devices including the video encoder 200 and / or the video decoder 300 may implement the video encoder 200 and / or the video decoder 300 in processing circuitry such as integrated circuits and / or microprocessors. Such devices may be wireless communication devices (such as cellular phones) or any other type of device described herein.

[0074] Video encoder 200 and video decoder 300 may operate according to video decoding standards such as ITU-T H.265 (also known as High Efficiency Video Decoding (HEVC)) or extensions thereof such as Multi-View and / or Scalable Video Decoding Extensions. Alternatively, video encoder 200 and video decoder 300 may operate according to other proprietary or industry standards such as ITU-T H.266 (also known as Multi-Functional Video Decoding (VVC)). In other examples, video encoder 200 and video decoder 300 may operate according to proprietary video codecs / formats such as AOMedia Video 1 (AV1), extensions to AV1, and / or subsequent versions of AV1 (e.g., AV2)). In other examples, video encoder 200 and video decoder 300 may operate according to other proprietary formats or industry standards. However, the techniques disclosed herein are not limited to any particular decoding standard or format. In general, video encoder 200 and video decoder 300 may be configured to perform the techniques of this disclosure in conjunction with any video decoding technique that uses a neural network to filter video data.

[0075] Generally, the video encoder 200 and video decoder 300 perform block-based decoding of images. The term "block" typically refers to a structure that includes data to be processed (e.g., encoded, decoded, or otherwise used during encoding and / or decoding). For example, a block may include a two-dimensional matrix of samples of luminance and / or chrominance data. Generally, the video encoder 200 and video decoder 300 decode video data represented in YUV (e.g., Y, Cb, Cr) format. That is, instead of decoding the red, green, and blue (RGB) data used for images, the video encoder 200 and video decoder 300 decode both the luminance and chrominance components, where the chrominance components may include both red and blue hue chrominance components. In some examples, the video encoder 200 converts the received RGB format data to a YUV representation before encoding, and the video decoder 300 converts that YUV representation to RGB format. Alternatively, preprocessing and postprocessing units (not shown) may perform these conversions.

[0076] This disclosure generally relates to the decoding (e.g., encoding and decoding) of images to include processes of encoding or decoding data of the image. Similarly, this disclosure may relate to the decoding of blocks of images to include processes of encoding or decoding data used for the blocks (e.g., prediction and / or residual decoding). Encoded video bitstreams typically include a series of values ​​for syntax elements representing decoding decisions (e.g., decoding modes) and the partitioning of images into blocks. Therefore, references to the decoding of images or blocks should generally be understood as the decoded values ​​of the syntax elements that form the images or blocks.

[0077] HEVC defines various blocks, including decoding units (CUs), prediction units (PUs), and transform units (TUs). According to HEVC, a video decoder (such as a video encoder 200) divides the decoding tree unit (CTU) into CUs based on a quadtree structure. That is, the video decoder divides the CTU and CU into four equal, non-overlapping squares, and each node of the quadtree has zero or four child nodes. Nodes without child nodes are called "leaf nodes," and the CU of such leaf nodes may include one or more PUs and / or one or more TUs. The video decoder may further divide the PUs and TUs. For example, in HEVC, a residual quadtree (RQT) represents the partitioning of the TU. In HEVC, the PU represents inter-frame prediction data, while the TU represents residual data. The CU after intra-frame prediction includes intra-frame prediction information, such as intra-frame mode indication.

[0078] As another example, video encoder 200 and video decoder 300 can be configured to operate according to VVC. According to VVC, the video decoder (such as video encoder 200) partitions the image into multiple CTUs. Video encoder 200 can partition the CTUs according to a tree structure (such as a quadtree-binary tree (QTBT) structure or a multi-type tree (MTT) structure). The QTBT structure removes the concept of multiple partition types, such as the separation between CUs, PUs, and TUs in HEVC. The QTBT structure includes two levels: a first level partitioned according to quadtree partitioning, and a second level partitioned according to binary tree partitioning. The root node of the QTBT structure corresponds to a CTU. The leaf nodes of the binary tree correspond to CUs.

[0079] In the MTT partitioning structure, blocks can be divided using quadtree (QT) partitioning, binary tree (BT) partitioning, and one or more types of ternary tree (TT) partitioning (also known as triplet tree (TT)). A ternary tree or triplet tree partition is a partition in which a block is divided into three sub-blocks. In some examples, a ternary tree or triplet tree partition divides a block into three sub-blocks without dividing the original block through the center. Partition types in MTT (e.g., QT, BT, and TT) can be symmetric or asymmetric.

[0080] When operating according to the AV1 codec, the video encoder 200 and video decoder 300 can be configured to decode video data in blocks. In AV1, the largest decoded block that can be processed is called a superblock. In AV1, a superblock can be 128x128 luma samples or 64x64 luma samples. However, in subsequent video decoding formats (e.g., AV2), superblocks can be defined by different (e.g., larger) luma sample sizes. In some examples, the superblock is the top level of a block quadtree. The video encoder 200 can further divide the superblock into smaller decoded blocks. The video encoder 200 can use square or non-square partitions to divide the superblock and other decoded blocks into smaller blocks. Non-square blocks can include N / 2xN blocks, NxN / 2 blocks, N / 4xN blocks, and NxN / 4 blocks. The video encoder 200 and video decoder 300 can perform separate prediction and transform processing for each decoded block.

[0081] AV1 also defines video data tiles. A tile is a rectangular array of superblocks that can be decoded independently of other tiles. That is, the video encoder 200 and video decoder 300 can encode and decode the decoding blocks within a tile separately without using video data from other tiles. However, the video encoder 200 and video decoder 300 can perform filtering across tile boundaries. The tile size can be uniform or non-uniform. Tile-based decoding enables parallel processing and / or multithreading in the encoder and decoder implementations.

[0082] In some examples, the video encoder 200 and the video decoder 300 may use a single QTBT or MTT structure to represent each of the luma and chroma components, while in other examples, the video encoder 200 and the video decoder 300 may use two or more QTBT or MTT structures, such as one QTBT / MTT structure for the luma component and another QTBT / MTT structure for the two chroma components (or two QTBT / MTT structures for the respective chroma components).

[0083] The video encoder 200 and the video decoder 300 can be configured to use quadtree partitioning, QTBT partitioning, MTT partitioning, superblock partitioning or other partitioning structures.

[0084] In some examples, a CTU includes a decoded tree block (CTB) of luminance samples, two corresponding CTBs of chrominance samples of an image with three sample arrays, or a CTB of samples of an image decoded using three separate color planes and a syntax structure for decoding the samples. A CTB can be an NxN sample block of some value N, such that a partitioning method divides the components into CTBs. A component is an array or a single sample from one of the three arrays (luminance and two chrominance) constituting a 4:2:0, 4:2:2, or 4:4:4 color format image, or an array or a single sample constituting an array or array constituting a monochrome format image. In some examples, a decoded block is an MxN sample block of values ​​M and N, such that a partitioning method divides the CTB into decoded blocks.

[0085] Blocks (e.g., CTUs or CUs) can be grouped in various ways within an image. As an example, a brick can refer to a rectangular area of ​​a row of CTUs within a specific tile in an image. A tile can be a rectangular area of ​​CTUs within a specific tile column and a specific tile row in an image. A tile column refers to a rectangular area of ​​a CTU having a height equal to the height of the image and a width specified by syntax elements (e.g., such as in an image parameter set). A tile row refers to a rectangular area of ​​a CTU having a height specified by syntax elements (e.g., such as in an image parameter set) and a width equal to the width of the image.

[0086] In some examples, a tile can be divided into multiple bricks, each brick comprising one or more CTU rows within the tile. A tile that is not divided into multiple bricks can also be called a brick. However, bricks that are a true subset of a tile cannot be called a tile. Bricks in an image can also be arranged in slices. A slice can be an integer number of bricks in an image that can be uniquely contained within a single Network Abstraction Layer (NAL) unit. In some examples, a slice comprises multiple complete tiles or a consecutive sequence of complete bricks comprising only one tile.

[0087] This disclosure uses "N×N" and "N by N" interchangeably to refer to the sample size of a block (such as a CU or other video block) in the vertical and horizontal dimensions, for example, 16×16 samples or 16 by 16 samples. Generally, a 16×16 CU will have 16 samples in the vertical direction (y=16) and 16 samples in the horizontal direction (x=16). Similarly, an N×N CU typically has N samples in the vertical direction and N samples in the horizontal direction, where N represents a non-negative integer value. Samples in a CU can be arranged in rows and columns. Furthermore, a CU does not necessarily need to have the same number of samples in the horizontal direction as it does in the vertical direction. For example, a CU may include N×M samples, where M is not necessarily equal to N.

[0088] The video encoder 200 encodes video data representing prediction and / or residual information, as well as other information, for use in the control unit (CU). The prediction information indicates how the CU should be predicted to form a prediction block for the CU. The residual information typically represents the sample-by-sample difference between a sample of the CU before encoding and the prediction block.

[0089] To predict the Cubic Frame (CU), the video encoder 200 typically forms prediction blocks for the CU through inter-frame prediction or intra-frame prediction. Inter-frame prediction typically refers to predicting the CU from data in a previously decoded image, while intra-frame prediction typically refers to predicting the CU from data in a previously decoded image within the same frame. To perform inter-frame prediction, the video encoder 200 can use one or more motion vectors to generate prediction blocks. The video encoder 200 can typically perform a motion search to identify reference blocks that closely match the CU, for example, based on the differences between the CU and a reference block. The video encoder 200 can use sum of absolute differences (SAD), sum of squared differences (SSD), mean absolute difference (MAD), mean squared difference (MSD), or other such difference calculations to compute difference metrics to determine whether a reference block closely matches the current CU. In some examples, the video encoder 200 can use unidirectional or bidirectional prediction to predict the current CU.

[0090] Some examples of VVC also provide an affine motion compensation mode, which can be viewed as an inter-frame prediction mode. In affine motion compensation mode, the video encoder 200 can determine two or more motion vectors representing non-translational motion, such as zooming in or out, rotation, perspective motion, or other irregular motion types.

[0091] To perform intra-frame prediction, the video encoder 200 can select an intra-frame prediction mode to generate prediction blocks. Some examples of VVC provide sixty-seven intra-frame prediction modes, including various directional modes, as well as planar and DC modes. Generally, the video encoder 200 selects an intra-frame prediction mode that describes the neighboring samples of the current block (e.g., the block of the CU) and predicts samples of the current block from them. Assuming the video encoder 200 decodes the CTU and CU in raster scan order (from left to right, from top to bottom), such samples can typically be located above, to the upper left, or to the left of the current block in the same image as the current block.

[0092] The video encoder 200 encodes data representing the prediction mode of the current block. For example, for inter-frame prediction modes, the video encoder 200 may encode data indicating which of the various available inter-frame prediction modes is used, as well as the motion information for the corresponding mode. For example, for unidirectional or bidirectional inter-frame prediction, the video encoder 200 may encode motion vectors using Advanced Motion Vector Prediction (AMVP) or merging modes. The video encoder 200 may use similar modes to encode motion vectors used for affine motion compensation modes.

[0093] AV1 includes two common techniques for encoding and decoding blocks of video data. These two common techniques are intra-frame prediction (e.g., intra-frame prediction or spatial prediction) and inter-frame prediction (e.g., inter-frame prediction or temporal prediction). In the context of AV1, when using intra-frame prediction modes to predict blocks of video data for the current frame, the video encoder 200 and video decoder 300 do not use video data from other frames of the video data. For most intra-frame prediction modes, the video encoder 200 encodes blocks of the current frame based on the difference between sample values ​​in the current block and predicted values ​​generated from reference samples in the same frame. The video encoder 200 determines the predicted values ​​generated from the reference samples based on the intra-frame prediction mode.

[0094] After prediction (such as intra-frame or inter-frame prediction for a block), the video encoder 200 can compute residual data for the block. The residual data (such as a residual block) represents the sample-by-sample difference between the block and the prediction block used to form the block, which is formed using the corresponding prediction mode. The video encoder 200 can apply one or more transforms to the residual block to produce transformed data in the transform domain rather than the sample domain. For example, the video encoder 200 can apply a Discrete Cosine Transform (DCT), an integer transform, a wavelet transform, or a conceptually similar transform to the residual video data. Additionally, the video encoder 200 can apply a secondary transform after the first transform, such as the Mode Correlated Inseparable Secondary Transform (MDNSST), the Signal Correlation Transform, the Karhunen-Loeve Transform (KLT), etc. The video encoder 200 produces transform coefficients after applying one or more transforms.

[0095] As noted above, after any transformation that produces the transform coefficients, the video encoder 200 may perform quantization on the transform coefficients. Quantization generally refers to the process in which the transform coefficients are quantized to reduce the amount of data used to represent them, thereby providing further compression. By performing the quantization process, the video encoder 200 may reduce the bit depth associated with some or all of the transform coefficients. For example, the video encoder 200 may quantize the transform coefficients during quantization. n The place value is rounded down to m Bit value, where n Greater than m In some examples, in order to perform quantization, the video encoder 200 may perform a bit-by-bit right shift of the value to be quantized.

[0096] After quantization, the video encoder 200 can scan the transform coefficients to generate a one-dimensional vector from a two-dimensional matrix containing the quantized transform coefficients. The scan can be designed to place higher-energy (and therefore lower-frequency) transform coefficients before the vector and lower-energy (and therefore higher-frequency) transform coefficients after the vector. In some examples, the video encoder 200 can utilize a predefined scan order to scan the quantized transform coefficients to produce a serialized vector, and then entropy-encode the quantized transform coefficients of that vector. In other examples, the video encoder 200 can perform an adaptive scan. After scanning the quantized transform coefficients to form a one-dimensional vector, the video encoder 200 can entropy-encode the one-dimensional vector, for example, according to context-adaptive binary arithmetic decoding (CABAC). The video encoder 200 can also entropy-encode the values ​​of syntax elements describing metadata associated with the encoded video data, which is used by the video decoder 300 when decoding the video data.

[0097] To perform CABAC, the video encoder 200 can assign context within a context model to the symbols to be transmitted. Context may involve, for example, whether the neighboring values ​​of a symbol are zero. Probability determination can be based on the context assigned to the symbols.

[0098] The video encoder 200 may further generate syntax data for the video decoder 300, such as block-based syntax data, image-based syntax data, and sequence-based syntax data, for example, in image headers, block headers, and slice headers, or generate other syntax data such as sequence parameter sets (SPS), image parameter sets (PPS), or video parameter sets (VPS). The video decoder 300 may also decode such syntax data to determine how to decode the corresponding video data.

[0099] In this way, the video encoder 200 can generate a bitstream that includes encoded video data, such as syntax elements describing the partitioning of images into blocks (e.g., CUs) and prediction and / or residual information for the blocks. Finally, the video decoder 300 can receive the bitstream and decode the encoded video data.

[0100] Generally, the video decoder 300 performs the reverse process of the video encoder 200 to decode the encoded video data of the bitstream. For example, the video decoder 300 can use CABAC to decode the values ​​of syntax elements used for the bitstream in a manner substantially similar to but reversed by the CABAC encoding process of the video encoder 200. Syntax elements can define partitioning information for dividing a picture into CTUs and defining the CUs of each CTU according to a corresponding partitioning structure such as a QTBT structure. Syntax elements can further define prediction and residual information for video data blocks (e.g., CUs).

[0101] The residual information can be represented, for example, by quantized transform coefficients. The video decoder 300 can inversely quantize and inverse transform the quantized transform coefficients of the block to reconstruct the residual block for the block. The video decoder 300 uses a signaling prediction mode (intra-frame prediction or inter-frame prediction) and associated prediction information (e.g., motion information for inter-frame prediction) to form a prediction block for the block. The video decoder 300 can then combine the prediction block and the residual block (on a sample-by-sample basis) to reconstruct the original block. The video decoder 300 can perform additional processing, such as performing a deblocking process to reduce visual artifacts along the block boundaries.

[0102] This disclosure may generally relate to "signaling" certain information (such as syntax elements). The term "signaling" can generally refer to the communication of values ​​and / or other data of syntax elements used to decode encoded video data. That is, video encoder 200 may signal the values ​​of syntax elements in the bitstream. Generally speaking, signaling refers to generating values ​​in the bitstream. As noted above, source device 102 may transmit the bitstream to destination device 116 substantially in real time or not in real time (such as when syntax elements are stored in storage device 112 for later retrieval by destination device 116).

[0103] This disclosure describes methods, techniques, and structures that can reduce the computational complexity and memory bandwidth requirements of neural network (NN)-based video decoding tools. The example techniques described below relate to NN-assisted loop filtering. However, the techniques of this disclosure are applicable to any NN-based video decoding tool using input data with certain statistical properties. The techniques of this disclosure can be used in the context of advanced video codecs, such as extensions to VVC, next-generation video decoding standards, and / or any other video codec.

[0104] According to the techniques disclosed herein, video encoder 200 and video decoder 300 can be configured to perform NN-based video decoding, including NN-based filtering using any combination of techniques described below. Specifically, video encoder 200 and video decoder 300 can implement an NN-based filter architecture in which backbone blocks are configured as pairs of backbone blocks, wherein each pair of backbone blocks includes a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction. The complexity of the NN-based filter can be reduced with minimal performance degradation by using feature channel reduction. In another example, video encoder 200 and video decoder 300 can implement an NN-based filter architecture including a fusion block with two separable convolutions each having applied spatial downsampling. This spatial downsampling reduces the complexity of the NN-based filter with minimal performance degradation.

[0105] In one example, video encoder 200 and video decoder 300 may be configured to receive an image of video data, reconstruct the image of video data, and perform a neural network-based filtering process on one or more blocks of the reconstructed image of video data using a neural network-based filter, wherein the neural network-based filter includes a pair of backbone blocks, each of the backbone blocks including a three-component 1D decomposition of multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0106] In-loop filter technology for video decoding Figure 2 This is a conceptual diagram illustrating a hybrid video decoding framework. Video decoding standards since H.261 are based on the so-called hybrid video decoding principle, which... Figure 2 The example is shown below. The term hybrid refers to a combination of two methods used to reduce redundancy in video signals: predictive and transform decoding with predictive residual quantization. Prediction and transform reduce redundancy in video signals by decorrelation, while quantization reduces the data represented by the transform coefficients by decreasing their precision, ideally by removing only irrelevant details. This hybrid video decoding design principle is also used in two recent standards, ITU-T H.265 / HEVC and ITU-T H.266 / VVC.

[0107] like Figure 2 As shown, a modern hybrid video decoder 130 typically performs block partitioning, motion compensation or inter-picture prediction, intra-picture prediction, transform, quantization, entropy decoding, and post-loop / intra-loop filtering. Figure 2In the example, the video decoder 130 includes a summing unit 134, a transform unit 136, a quantization unit 138, an entropy decoding unit 140, an inverse quantization unit 142, an inverse transform unit 144, a summing unit 146, a loop filter unit 148, a decoded picture buffer (DPB) 150, an intra-frame prediction unit 152, an inter-frame prediction unit 154, and a motion estimation unit 156.

[0108] Typically, the video decoder 130 receives input video data 132 while encoding video data. Block partitioning is used to divide the received video data images into smaller blocks for use in the prediction and transformation processes. Early video decoding standards used a fixed block size, typically 16×16 samples. More recent standards such as HEVC and VVC employ tree-based partitioning structures to provide flexible partitioning.

[0109] Motion estimation unit 156 and inter-frame prediction unit 154 can predict input video data 132, for example, based on previously decoded data from DPB 150. Motion compensation, or inter-picture prediction, utilizes the redundancy present between pictures in the video sequence (hence the term "inter-picture"). According to block-based motion compensation used in modern video codecs, predictions are obtained from one or more previously decoded pictures (i.e., reference pictures). The corresponding regions used to generate inter-frame predictions are indicated by motion information, including motion vectors and reference picture indices.

[0110] The summing unit 134 calculates the residual data as the difference between the input video data 132 and the predicted data from the intra-frame prediction unit 152 or the inter-frame prediction unit 154. The summing unit 134 provides the residual block to the transform unit 136, which applies one or more transforms to the residual block to produce a transform block. The quantization unit 138 quantizes the transform block to form quantized transform coefficients. The entropy decoding unit 140 entropy-encodes the quantized transform coefficients and other syntax elements (such as motion information or intra-frame prediction information) to produce an output bitstream 158.

[0111] Simultaneously, the inverse quantization unit 142 inverse quantizes the quantized transform coefficients, and the inverse transform unit 144 inverse transforms the transform coefficients to reconstruct the residual block. The summation unit 146 combines the residual block with the prediction block (on a sample-by-sample basis) to generate a decoded video data block. The loop filter unit 148 applies one or more filters (e.g., at least one of a neural network-based filter, a neural network-based loop filter, a neural network-based post-loop filter, an adaptive in-loop filter, or a predefined adaptive in-loop filter) to the decoded block to generate a filtered decoded block.

[0112] According to the technology of this disclosure, the neural network filtering unit of the loop filter unit 148 may receive data of a reconstructed image from video data from the summing unit 146 and from one or more other units of the hybrid video decoder 130 (e.g., transform unit 136, quantization unit 138, intra-frame prediction unit 152, inter-frame prediction unit 154, motion estimation unit 156, and / or one or more other filtering units within the loop filter unit 148). For example, the neural network filtering unit may receive data from the deblocking filtering unit (also referred to as the "deblocking unit") of the loop filter unit 148. The neural network filtering unit may receive, for example, boundary strength values, which indicate whether a particular boundary will be filtered for deblocking and, if so, to what extent the boundary will be filtered. For example, the boundary strength values ​​may correspond to multiple samples on either side of a boundary to be modified and / or the degree to which samples are to be modified.

[0113] In other examples, as a supplement to or alternative to the boundary strength values, the neural network filtering unit may receive decoding unit (CU) partitioning data, prediction unit (PU) partitioning data, transform unit (TU) partitioning data, deblocking filtering data, quantization parameter (QP) data, intra-frame prediction data (e.g., reconstructed samples and / or predicted samples), inter-frame prediction data (e.g., reconstructed samples and / or predicted samples), data representing the distance between a decoded image and one or more reference images, or any or all of the motion information of one or more decoded blocks of a decoded image. The deblocking filtering data may also include one or more of the following: whether a long filter or a short filter was used for deblocking, or whether a strong filter or a weak filter was used for deblocking. The data representing the distance between the decoded image and the reference image may be represented as the difference in picture order count (POC) between the POC values ​​of the images.

[0114] Video data blocks (such as CTUs or CUs) can actually include multiple color components, such as a luminance or "luminance" component, a blue hue chroma or "chroma" component, and a red hue chroma (chroma) component. The luminance component can have a larger spatial resolution than the chroma components, and one chroma component can have a larger spatial resolution than the other. Alternatively, the luminance component can have a larger spatial resolution than the chroma components, and the two chroma components can have equal spatial resolutions. For example, in a 4:2:2 format, the luminance component can be twice as large as the chroma component horizontally and equal to the chroma component vertically. Similarly, in a 4:2:0 format, the luminance component can be twice as large as the chroma component both horizontally and vertically. The various operations discussed above can generally be applied individually to each of the luminance and chroma components (but some decoding information, such as motion information or intra-frame prediction direction, can be determined for the luminance component and inherited by the corresponding chroma component).

[0115] In recent video codecs, a hierarchical prediction structure within a group of pictures (GOP) is applied to improve decoding efficiency. Figure 3 An example hierarchical prediction structure 166 with a group of pictures (GOP) of size 16 is shown.

[0116] Refer again Figure 2 In-picture prediction utilizes spatial redundancy present within the picture (hence "within the box") by deriving predictions of blocks from spatially neighboring (reference) samples that have already been decoded / decoded. Angle prediction, DC prediction, and planar or on-plane prediction are used in recent video codecs, including AVC, HEVC, and VVC.

[0117] Hybrid video decoding standards can apply block transforms to prediction residuals (regardless of whether the prediction residuals come from inter-picture or intra-picture predictions). Early standards (including H.261 / 262 / 263) used Discrete Cosine Transform (DCT). In HEVC and VVC, more transform kernels besides DCT can be applied to handle different statistical information in specific video signals.

[0118] Quantization aims to reduce the precision of input values ​​or a set of input values ​​in order to reduce the amount of data required to represent those values. In hybrid video decoding, quantization is typically applied to residual samples of individual transforms, such as transform coefficients, resulting in integer coefficient levels. In recent video decoding standards, the stride is derived from a so-called quantization parameter (QP) that controls fidelity and bit rate. A larger stride reduces the bit rate but also degrades quality, which can cause video images to exhibit blocky artifacts and blurred details, for example.

[0119] The entropy decoding unit 140 can perform context-adaptive binary arithmetic decoding (CABAC) on the encoded video. CABAC is used in recent video codecs such as AVC, HEVC, and VVC due to its high efficiency.

[0120] The loop filter unit 148 can perform post-loop or in-loop filtering. Post-loop / in-loop filtering is a filtering process (or a combination of such processes) applied to the reconstructed image to reduce decoding artifacts. The input to the filtering process is generally the reconstructed image, which is a combination of the reconstructed residual signal (which includes quantization errors) and the prediction. Figure 2 As shown, the reconstructed image after in-loop filtering is stored in the decoded image buffer (DPB) 150 and used as a reference for inter-image prediction of subsequent images. Decoding artifacts mainly depend on the QP. Therefore, QP information is often used in the design of the filtering process. In HEVC, the in-loop filter includes deblocking filtering and sample adaptive offset (SAO) filtering. In the VVC standard, an adaptive loop filter (ALF) is introduced as a third filter. The ALF filtering process is shown below: , (1) in These are samples before the filtering process. These are the sample values ​​after the filtering process. Represents the filter coefficients. It is a trimming function, and This represents the trimming parameters. Variables k and l in... and The changes between them, among which L Indicates the filter length. Trim function. It corresponds to the function The pruning operation introduces nonlinearity to make ALF more efficient by reducing the influence of neighboring sample values ​​that are very different from the current sample value. In VVC, filter parameters can be signaled in the bit stream and selected from a predefined set of filters. The ALF filtering process can also be summarized by the following equation: (2) Neural network (NN) based filtering for video decoding Figure 4 This is a conceptual diagram illustrating a neural network-based filter 170 with four layers. Various studies have shown that embedding neural networks (NNs) into, for example... Figure 2 Compression efficiency can be improved in hybrid video decoding frameworks. Neural networks have been used for intra-frame and inter-frame prediction to improve prediction efficiency. In recent years, NN-based intra-loop filtering has also been an important research topic. In some examples, the filtering process is applied as a post-filter. In such examples, the filtering process is applied to the output image, and the unfiltered image can be used as a reference image.

[0121] In addition to existing filters such as deblocking filters, sample adaptive offset (SAO), and / or adaptive loop filtering (ALF), NN-based filters 170 can also be applied. NN-based filters can also be applied exclusively, where the NN-based filters are designed to replace all existing filters. Additionally or alternatively, NN-based filters (such as NN-based filter 170) can be designed to supplement, enhance, or replace any or all other filters.

[0122] Figure 4 An example of a filter based on a four-layer convolutional neural network (CNN) is shown. Figure 4The NN-based filtering process can take reconstructed luminance and chrominance samples as input, which are wrapped in a 3D volume with six planes. The intermediate output is a residual sample, which is added back to the input to refine the input samples. The NN filter can use all color components (e.g., Y, U, and V, or Y, Cb, and Cr, i.e., luminance data 172A, blue hue chrominance 172B, and red hue chrominance 172C) as input to utilize cross-component correlations. Different color components can share the same filter (including network structure and model parameters), or each component can have its own specific filter.

[0123] The filtering process can also be summarized as follows: (3) The model structure and parameters of NN-based filters can be predefined and stored at the encoder and decoder. The filters can also be signaled in the bitstream.

[0124] exist Figure 4 In the example, a neural network-based filter could include a series of feature extraction layers followed by an output convolution. Figure 4 In a convolutional layer, the feature extraction layer may include a 3×3 convolutional (conv) layer followed by a parameter-corrected linear unit (PReLU) layer. The convolutional layer applies convolution operations to the input data, which involves a filter or kernel sliding across the input data (e.g., a reconstructed sample of input 172) and computing the dot product at each location. The convolutional operation essentially captures local patterns within the input data. For example, in the context of image processing, these patterns could be edges, textures, or other visual features. The filter or kernel is a small matrix of weights that is updated during the training process. By sliding the filter across the input data (or a feature map from a previous layer) and computing the dot product at each location, the convolutional layer creates a feature map that encodes the spatial hierarchy and patterns detected in the input.

[0125] The output of a convolutional layer is a set of feature maps, each corresponding to a filter used to capture different aspects of the input data. As data passes through deeper layers of the network, this layer helps the neural network learn increasingly complex and abstract features. Figure 4 In the naming convention 3×3 conv 3×3×6×8, the first 3×3 indicates that the convolutional layer has a 3×3 filter size (e.g., a 3×3 matrix). 3×3×6×8 refers to both the input and output dimensions of the convolutional layer, where 6 is the number of input channels and 8 is the number of output channels.

[0126] The PReLU layer is an activation function used in neural networks and is introduced as a variant of the ReLU (Modified Linear Unit) activation function. As described above, convolutional layers output feature maps, each corresponding to a filter used to represent the detected features in the input. After the convolutional layers, the PReLU layer applies the PReLU activation function to each element of the feature map produced by that convolutional layer. For positive values, the PReLU layer acts similarly to standard ReLU, allowing the value to pass through. For negative values, the PReLU layer allows small linear negative outputs instead of setting them to zero (e.g., as ReLU does). This keeps neurons active and maintains gradient flow, which can be beneficial for learning in deep networks.

[0127] In summary, when a convolutional layer is followed by a PReLU layer, the convolutional layer first extracts features from the input data through a set of learned filters. The resulting feature map is then passed through the PReLU activation function, which introduces non-linearity and helps avoid neuron death by allowing small gradients when the input is negative. This combination is effective when learning complex patterns in the data while maintaining a robust gradient flow, which is particularly beneficial in deeper network architectures.

[0128] Processing unit When applying neural network-based filtering in video decoding, the entire video signal (pixel data) may be divided into multiple processing units (e.g., 2D blocks), and each processing unit may be processed individually or combined with other information associated with that pixel block. Possible choices for processing units include frames, slices / tiles, CTUs, or any predefined or signaled shape and size. Typically, neural network-based filtering is performed on reconstructed blocks of video data. Here, reconstructed blocks and samples may refer to decoded blocks generated by video decoder 300 and blocks reconstructed in the reconstruction loop of video encoder 200.

[0129] Types of input data To further improve the performance of neural network-based filtering, different types of input data can be jointly processed to produce filtered outputs. Input data may include, but is not limited to, reconstructed pixels / samples, predicted pixels / samples, pixels / samples after loop filtering, partitioning structure information, deblocking parameters (e.g., boundary strength (BS)), QP values, slice or image type, or filter suitability or decoding mode diagrams. Input data can be provided at different granularities. Luminosity reconstruction and predicted samples can be provided at the original resolution, while chromosity samples can be provided at a lower resolution (e.g., for 4:2:0 representation), or can be upsampled to luminosity resolution for per-pixel representation. Similarly, QP, BS, partitioning, or decoding mode information can be provided at a lower resolution, including cases where each frame, slice, or processing block (e.g., QP) has a single value. In other examples, QP, BS, partitioning, or decoding mode information can be expanded (e.g., replicated) for per-pixel / sample representation.

[0130] Figure 5 An example of an architecture that utilizes supplementary data is shown in the figure. Figure 5 This is a block diagram illustrating an example CNN-based filter with padded input samples and supplementary data. The NN-based filter 171 uses pixels / samples of a processing block combined with supplementary data as input 174. Input 174 may include four sub-blocks 174A of interleaved luminance samples (Y×4) and associated blue hue chromaticity (U) data 174B and red hue chromaticity (V) data 174C. Supplementary data includes quantization parameter (QP) step size 176 and boundary strength (BS) 178. The region of input pixels / samples can be expanded from each side using four padded pixels / samples. The resulting size of the processing volume is (4+64+4)×(4+64+4)×(4Y+2UV+1QP+3BS).

[0131] Compared to Figure 4 The neural network-based filter 171 may include two or more hidden layers utilizing both 1×1 convolutions and Leaky ReLU layers. The Leaky ReLU layer, similar to the PReLU layer, allows the output of a small non-zero gradient when the layer is inactive. Instead of outputting zero for negative inputs, Leaky ReLU multiplies these inputs by a small constant. This small slope ensures that even neurons that would otherwise be inactive still contribute a small amount to the network's learning, thus reducing the likelihood of ReLU dying.

[0132] Based on multi-mode design NN Filtering To further improve the performance of neural network-based filtering, multi-mode solutions can be designed. For example, for each processing unit, the video encoder 200 can select from a set of modes based on rate-distortion optimization and can signal this selection in the bitstream. Different modes may include different neural network models, different values ​​of the input information used as the neural network model, and / or other factors. In one example, the video encoder 200 and the video decoder 300 can use a multi-mode neural network-based filtering solution based on a single neural network model by using different QP values ​​as inputs to the neural network model for different modes.

[0133] Examples of NN architectures In one example, a neural network-based filtering solution with multiple modes can be used, as described above. Figure 6 The structure of the network is shown in the figure. Figure 6 The NN-based filter includes a first part (e.g., a feature extraction segment) comprising input 3×3 convolutional filters 510A to 510E and corresponding parameter-corrected linear unit (PReLU) filters 512A to 512E for each of the inputs to generate feature maps (e.g., the feature extraction segment of the NN filter). A cascade unit 514 cascades the feature maps and provides them to a fusion block 516 and a transformation block 522. Figure 6 The NN-based filters also include a set 528 of attention residual (AttRes) blocks 530A to 530N; and a final portion (e.g., a tail segment) comprising a 3×3 convolutional filter 550, a PReLU filter 552, a 3×3 convolutional filter 554, and a pixel recombination unit 556. The AttRes block may also be referred to as the backbone block.

[0134] In the first part (e.g., the feature extraction section), different inputs are received, including quantization parameters (QP) 500, partitioning information (part) 502, boundary strength (BS) 504, predicted samples (pred) 506, and reconstructed samples (rec) 508. Corresponding 3×3 convolutional filters 510A to 510E and PReLU filters 512A to 512E convolve and activate the corresponding inputs to generate feature maps. These feature maps are then concatenated by a cascade unit 514. A fusion block 516, including a 1×1 convolutional filter 518 and a PReLU filter 520, fuses the concatenated feature maps. A transformation block, including a 3×3 convolutional filter 524 and a PReLU filter 526, subsamples the fused input to create an output 188. The output 188 is then fed through a set 528 of attention residual blocks 530A to 530N, which may include a variety of numbers of attention residual blocks, such as eight. The AttRes block can also accept quantization parameters (QP) 500, partitioning information (part) 502, boundary strength (BS) 504, predicted samples (pred) 506, and reconstructed samples (rec) 508 as inputs. This attention block is relative to... Figure 7 Further explanation. The output 189 from the last attention residual block in the set 184 of attention residual blocks is fed into the final part of the NN-based filter. In this final part, a 3×3 convolutional filter 550, a PReLU filter 552, a 3×3 convolutional filter 554, and a pixel reconstruction unit 556 process the output 189, and an adder unit 558 combines this result with the original reconstructed sample input 508. This ultimately forms a filtered output for presentation and storage (e.g., in a decoded image buffer (DPB)) as a reference for subsequent inter-frame prediction. In some examples, Figure 6 The NN-based filter uses 96 feature maps.

[0135] Figure 7 This is an example Figure 6 A conceptual diagram of the attention residual block. That is, Figure 7 An attention residual block 530 is depicted, which may include similar attention residual blocks. Figure 6 The attention residual blocks 530A to 530N are components of the components. In this example, the attention residual block 530 includes a first 3×3 convolutional filter 532, a parameter-corrected linear unit (PReLU) filter 534, a second 3×3 convolutional filter 536, an attention block 538, and an adder unit 540. The adder unit 540 combines the output of the attention block 538 with the output 188 initially received by the convolutional filter 532 to produce an output 189.

[0136] Figure 8 This is an example Figure 7A conceptual diagram of an example spatial attention layer. (See also:) Figure 8 As shown, the spatial attention layer of the attention residual block 530 includes a 3×3 convolutional filter 706, a PReLU filter 708, a 3×3 convolutional filter 710, a size expansion unit 712, a 3×3 convolutional filter 720, a PReLU filter 722, and a 3×3 convolutional filter 724. The 3×3 convolutional filter 706 receives input 702, which corresponds to... Figure 6 The quantization parameters (QP) are 500, the partition information (part) is 502, the boundary strength (BS) is 504, the prediction information (pred) is 506, and the reconstruction block (rec) is 508. A 3×3 convolutional filter 720 receives Z... K Value 704. The outputs of the size expansion unit 712 and the 3×3 convolution filter 724 are combined, and then combined with the R value 730 to generate the S value 732. The S value 732 is then combined with the Z value. K Values ​​704 are combined to produce output Z. K+1 Value 734.

[0137] In other examples, alternative designs of the NN architecture can be used. For example, one could use... Figure 6 The filter backbone has a large number of low-complexity residual blocks, and the number of channels (feature maps) is reduced and the attention module is removed. Figure 9 The alternative convolutional neural network filtering structure (e.g., for brightness filtering) is shown in the figure. Figure 9 This is a block diagram illustrating a simplified CNN-based filter architecture with padded input samples and supplementary data.

[0138] exist Figure 9 In the example, the number of residual blocks used is M=24. The number of feature maps (convolutions) is reduced to 64. In the residual block (ResBlock), the number of channels first increases to 160 before the activation layer and then decreases to 64 after the activation layer. For different performance-complexity tradeoffs, the number of residual blocks and channels can be configured differently (M is set to another value and the number of channels in the residual block can be set to a number other than 160). Chroma filtering can follow... Figure 6 The concept is to modify its core as described above to handle chroma channels.

[0139] Figure 9The NN-based filters include 3×3 convolutional filters 810A to 810E and PReLU filters 812A to 812E, which convolve the corresponding inputs (i.e., QP 800, Part 802, BS 804, Pred 806, and Rec 808) to generate feature maps (e.g., feature extraction segments). A cascade unit 814 concatenates the convolutional inputs (e.g., feature maps). A fusion block 816 then fuses the concatenated feature maps using a 1×1 convolutional filter 818 and a PReLU filter 820. A transformation block 822 then processes the fused data using a 3×3 convolutional filter 824 and a PReLU filter 826.

[0140] In this example, the NN-based filter comprises a set 828 of residual blocks 830A to 830N (also referred to as backbone blocks), each of which can be configured according to the following discussion. Figure 10 The residual block structure 830 is used for construction. Residual blocks 830A to 830N are replaceable. Figure 6 AttRes blocks 530A to 530N. Figure 9 The example can be used for luminance (luminance) filtering, but similar modifications can be made for chrominance (chrominance) filtering as discussed below.

[0141] Figure 9 The number of residual blocks and channels included in set 828 can be configured in different ways. That is, N can be set to different values, and the number of channels in residual block structure 830 can be set to a number other than 160 to achieve different performance-complexity tradeoffs. These modifications can be used to perform chroma filtering to process chroma channels.

[0142] The set 828 of residual blocks 830A to 830N has N instances of residual block structure 830. In one example, N can be equal to 32, such that there are 32 residual block structures. Residual blocks 830A to 830N can use 64 feature maps, which are relative to... Figure 6 The 96 feature maps used in the example have been reduced.

[0143] Figure 10 This is an example Figure 9A conceptual diagram of an example residual block structure 830 is provided. In this example, the residual block structure 830 includes a first 1×1 convolutional filter 832, which increases the number of input channels to 160 before the activation layer (PReLU filter 834) processes the input channels. The PReLU filter 834 can then reduce the number of channels to 64 through this processing. A second 1×1 convolutional filter 836 then processes the reduced channels, followed by a 3×3 convolutional filter 838. Finally, a combining unit 840 combines the output of the 3×3 convolutional filter 838 with the original input received by the residual block structure 830.

[0144] In yet another NN architecture, the residual block can be derived as follows: Figure 11 The filter blocks shown are replaced. In this example, the bypass branches around the convolution and activation layers in the residual blocks of the previous example are removed. The number of channels and the number of filter blocks can be configured (e.g., 64 channels, 24 filter blocks), with 160 channels before and after activation, resulting in a network complexity of 605.93kMAC and 1.5M parameters for the internal luminance model.

[0145] Figure 11 This illustrates alternatives to the technology according to this disclosure. Figure 6 A conceptual diagram of another example filter block structure of an attention residual block set. Figure 11 The NN-based filters include 3×3 convolutional filters 1010A to 1010E and PReLU filters 1012A to 1012E, which convolve the corresponding inputs (i.e., QP 1000, Part 1002, BS 1004, Pred 1006, and Rec 1008) to form feature maps (e.g., feature extraction segments). Concatenation unit 1014 concatenates these feature maps. Then, fusion block 1016 fuses the concatenated inputs using 1×1 convolutional filter 1018 and PReLU filter 1020. Transformation block 1022 then processes the fused data using 3×3 convolutional filter 1024 and PReLU filter 1026.

[0146] In this example, the NN-based filtering unit comprises N filter blocks 1030A to 1030N (also referred to as backbone blocks), each of which may have the features discussed below. Figure 12 The filter block structure 1030 is substantially similar to the residual block structure 830, except that the combination unit 840 is omitted from the filter block structure 1030, so that the input is not combined with the output. Instead, the output of each residual block structure can be directly fed to the subsequent block.

[0147] The number of channels and the number of filter blocks can be configured. In one example, it can be set to 64 channels and 32 filter blocks. The number of additional channels in each filter block 198 can be 160 as discussed above.

[0148] Figure 12 This is an example Figure 11 A conceptual diagram of an example filter block structure 1030 is provided. In this example, the residual block structure 1030 includes a first 1×1 convolutional filter 1032, which increases the number of input channels to 160 before the activation layer (PReLU filter 1034) processes the input channels. The PReLU filter 1034 can then reduce the number of channels to 64 through this processing. A second 1×1 convolutional filter 1036 then processes the reduced number of channels, followed by a 3×3 convolutional filter 1038. As discussed above, with... Figure 10 Compared to the residual block structure 830, the filter block structure 1030 does not include a combination unit.

[0149] In-Loop (ILF) Filters in Multi-Mode CNNs with Separable Convolution Convolutions with 3×3 kernels are popular in neural network-based filters. In the architecture described above, 3×3×N×M convolutions are utilized across multiple segments and blocks, where the 3×3 kernels slide in the spatial (2D) domain. However, multidimensional convolutions such as 2D kernel convolutions introduce significant complexity. According to the techniques disclosed herein, video encoder 200 and video decoder 300 can be configured to utilize separable convolutions instead of multidimensional convolutions (e.g., 3×3×N×M convolutions). For example, two separable one-dimensional convolutions can be used instead of 3×3 convolutions in any segment of a neural network-based filter. The use of separable convolutions reduces computational complexity and memory bandwidth requirements.

[0150] To avoid overcomputation and reduce the parameter set derived from multidimensional convolutions (such as 3×3 convolutions (or 2D convolutional quantities with higher-dimensional kernels) or similar convolutions in the CNN architecture described above), this disclosure describes a technique in which the video encoder 200 and video decoder 300 are configured to utilize separable convolutions (e.g., 1D separable convolutions) generated by low-complexity approximations instead of multidimensional (e.g., 2D) convolutions that slide in the spatial direction. While the technique of this disclosure is described with reference to 3×3 convolutions (e.g., 4×4, 5×5, or larger), the decomposition technique of this disclosure can be used for multidimensional convolutions of any size. Typically, a multidimensional convolution has a kernel size n1×n2 in the spatial dimension, where n1 and n2 are positive integers. The values ​​of n1 and n2 can be the same or different. A multidimensional convolution can also have a size K (e.g., n1×n2×K) in the depth dimension. Furthermore, when the number of output channels is M, a multidimensional convolution can be represented as a 4-D tensor of n1×n2×K×M.

[0151] In one example of this disclosure, the low-rank convolution approximation decomposes a 3×3×M×N convolution into a pixel-wise convolution (1×1×M×R), two separable convolutions (3×1×R×R, 1×3×R×R), and another pixel-wise convolution (1×1×R×N). Here, R is the rank of the approximation and can be used to adjust the performance / complexity of the approximation. The value of R can be an integer. In some examples, R can be derived as a function (ratio) of M or N or max(M,N). In some examples, R can be set to equal A*max(M,N), where A is less than 1 (e.g., 0.2, 0.5, 0.8), greater than 1 (e.g., 1.0, 1.2), or other values.

[0152] In a typical example, a multidimensional convolution can be approximated by multiple separable convolutions in the following way: perform a first convolution of size n1×1 and perform a second convolution of size 1×n2 on the output of the first convolution.

[0153] Figure 13 An example is shown of using separable convolutions to approximate multidimensional convolutions in the backbone of a neural network-based filter. In this example, the multidimensional convolution is a 2D 3×3 convolution. However, the techniques disclosed herein can be extended to convolutions of other dimensions.

[0154] Figure 13 It shows that the backbone block is from Figure 9 and Figure 10 Example of residual block 830A. Figure 13 An example is shown where a 3×3×K×K convolution 838 is decomposed into a series of 1D and separable convolutions in residual blocks 1300. That is, the video encoder 200 and the video decoder 300 can be configured to execute residual blocks 1300, which include performing multiple separable convolutions to approximate multidimensional convolutions. Figure 13 In the examples, multiple separable convolutions include a 3×1×R×R separable convolution (sep.conv) 1304 and a 1×3×R×R separable convolution 1306. In other examples, the order of separable convolutions 1304 and 1306 can be interchanged.

[0155] Therefore, in one example of this disclosure, to perform multiple separable convolutions to approximate multidimensional convolution, the video encoder 200 and the video decoder can be configured to perform a first 1×1 convolution (e.g., a 1×1×K×R convolution 1302), perform a first separable convolution (e.g., a 3×1×R×R separable convolution 1304) on the output of the first 1×2 convolution, perform a second separable convolution (e.g., a 1×3×R×R separable convolution 1306) on the output of the first separable convolution, and perform a second 1×1 convolution (e.g., a 1×1×R×K convolution 1308) on the output of the second separable convolution. The number of output channels of the first 1×1 convolution 1302 is used to control the complexity approximation of the multidimensional convolution. The number of output channels of the separable convolutions 1304 and 1306 can be selected to control the complexity approximation of the multidimensional convolution. The separable convolutions 1304 and 1306 are performing depthwise convolution operations.

[0156] In another example, to perform multiple separable convolutions to approximate multidimensional convolution, the video encoder 200 and the video decoder 300 may receive an input, perform a 1×1×K×M convolution on the input, perform a PReLU layer on the output of the 1×1×K×M convolution, perform a 1×1×M×K convolution on the output of the PReLU layer; perform a 3×1×K×R separable convolution on the output of the 1×1×M×K convolution, perform a 1×3×R×K separable convolution on the output of the 3×1×K×R separable convolution, and perform a 1×1×R×K convolution on the output of the 1×3×R×K separable convolution.

[0157] In a further example, Figure 13 An example is shown where a 3×3×K×K convolution 838 is decomposed into a series of 1D and separable convolutions in residual block 1300, but where the first 1D convolution is fused with another 1D convolution. That is, video encoder 200 and video decoder 300 can be configured to execute residual block 1310, which includes executing 1×1×M×R convolution 1320, which is a fusion of 1×1×M×K convolutions (e.g., convolution 836) and 1×1×K×R convolutions (e.g., convolution 1302). In this way, the decomposition of multidimensional convolutions can be further simplified. The fusion of 1×1 convolutions in residual block 1310 can be used to decompose any cases where convolutions are applied before or after another pixel-wise convolution (e.g., 1×1 fusion) and there are no nonlinearities or residual connections between them. The values ​​of M and R can be selected to control the complexity and accuracy of the approximation.

[0158] In other examples of this disclosure, the NN-based filtering process includes the cascaded (e.g., sequentially utilized) application of backbone blocks. For example, backbone blocks may be applied to multiple different color components. In other examples, the NN-based filtering process includes the cascaded application of backbone blocks applied in two or more parallel processing branches.

[0159] In one or more examples of this disclosure, performing multiple separable convolutions to approximate a neural network-based filtering process within a backbone block involves applying an element-wise activation procedure as part of the multidimensional convolution. Examples of element-wise activation procedures may include the ReLU function and the PReLU function. The PReLU function is an example of a parameter-controlled element-wise activation procedure.

[0160] Different algorithms can be used to determine decompositions to identify separable kernels that can replace 2D or other multidimensional kernels. In some examples, Candecomp / Parafac (CP) tensor decomposition can be used. Examples of other decompositions suitable for use with this disclosure can be found in “Speedingup Convolutional Neural Networks Using Fine-tuned CP-Decomposition” (ICLR 2015) by V. Lebedev, Y. Ganin, M. Rakhuba, I. Oseledets, and V. Lempitsky.

[0161] Alternatively, choose a specific implementation and architecture: In some examples, 2D convolutions of different dimensions (e.g., Z×Y) or high-dimensional convolutions can be used and replaced with corresponding separable convolutions 1×Z and Y×1.

[0162] Figure 14 This is a block diagram illustrating another example of a multidimensional convolution solution. For example... Figure 14 As shown, the 3×3×K×K convolution 838 of residual block 830A is approximated by the 3×1×K×R convolution 1400, and then by the 1×3×R×K convolution 1410. In other examples, the positions of convolution 1400 and the convolution can be interchanged. R is the canonical rank of this decomposition. A lower rank implies a greater reduction in complexity. Figure 13 Compared to the previous example, the video encoder 200 and the video decoder 300 can use two separable convolutions to approximate a 3×3 convolution without performing a leading or trailing 1×1 convolution.

[0163] In one example Figure 10 The architecture (where decomposition is exemplified in) Figure 14The implementation (in the Chinese version) uses parameters K=64, M=160, and R=51, and the total number of 24 residual blocks results in a network complexity of 356.43kMAC and the number of parameters used for the internal luminance model is 1.07M.

[0164] Multimodal CNN ILF with Bicomponent Decomposition for Multiscale Feature Extraction Multi-scale feature extraction with two-component convolutional networks has also been proposed, exemplified in Figure 15 In. Figure 15 In the diagram, the 3×3 convolution decomposes into a 3×1×C1×R convolution followed by a 1×3×R×C2 convolution, where C1 and C2 are the number of input and output channels, respectively, and R is the approximate rank. The parameter R can be derived proportionally to R = C1×C2 / (C1+C2) and controls the approximate complexity.

[0165] Figure 15 The proposed architecture is shown, where 3×3 convolutional blocks are replaced by separable convolutions of dimensions 3×1 and 1×3. In this example, the residual block structure 1530 includes a first 1×1 convolution 1532 preceding a first activation layer (PReLU 1534), and a 3×3 convolution 1540 and a second activation layer (PReLU 1542) parallel to the first 1×1 convolution 1532 and PReLU 1534. The second 1×1 convolution 1536 then processes the combined output of PReLU 1534 and PreLU 1542, followed by a 3×3 convolution 1538. Figure 15 In the example, the 3×3 convolution 1540 can be approximated using multiple separable convolutions, as shown below. Figure 15 The 3×1 convolution 1550 and 1×3 convolution 1552 are shown in the figure. Similarly, the 3×3 convolution 1538 can be approximated by several separable convolutions, as shown in the figure. Figure 15 The 3×1 convolution 1560 and 1×3 convolution 1562 are shown in the figure.

[0166] As an example Figure 15 The illustrated architecture can be implemented with parameters R1=8, R2=44, M1=160 and M2=16 and a total of 24 residual blocks. The network complexity will be 358.43kMAC and the number of parameters used for the internal luminance model will be 1.07M.

[0167] Unified CNN ILF with Two-Component Decomposition A multi-scale feature extraction backbone with two-component decomposition has been integrated into some examples of the unified model. Furthermore, the specifications of such examples may include two model versions: 1) a unified model for joint luminance and chrominance (see [link to model]). Figure 16A and Figure 16B ); and 2) are used for separate models for luminance and chrominance samples respectively (see Figure 17 and Figure 18 ).

[0168] Figure 16A This is a block diagram illustrating an example unified filter with a joint model (joint luminance and chrominance). Figure 16A NN-based filters may include IPB 1600 (which provides information about whether the block is predicted between or within), QP SLICE 1602 (which provides quantization parameters for slices), QP BASE 1604 (which provides quantization parameters for the sequence), 1606 (Boundary Strength (BS), 1608 (Predicted Samples (PRED)), and 1609 (Reconstructed Samples (REC)). Reconstructed Sample 1609 includes brightness samples (REC). EXT Y) and twice the upsampled chromaticity sample (REC) EXT (UV). Video encoder 200 and video decoder 300 can apply the corresponding 3×3 convolutions of 1610A to 1610F to each of the inputs. Video encoder 200 and video decoder 300 can apply the corresponding PReLUs of 1612A to 1612F to the outputs of 3×3 convolutions of 1610A to 1610F. In some examples, backbone blocks 1630A to 1630N can be residual blocks.

[0169] The fusion block 1616 can fuse feature maps using 1×1 convolution 1618 and PReLU 1620. The transformation block 1622 then processes the fused data using 3×3 convolution 1624 and PReLU 1626.

[0170] Video encoder 200 and video decoder 300 may apply backbone block group 1628. This backbone block group 1628 may include multiple backbone blocks 1630A to 1630N. In some examples, N=24, resulting in 24 backbone blocks. Video encoder 200 and video decoder 300 may apply 3×3 convolution 1650 to the output of backbone block 1628. Video encoder 200 and video decoder 300 may apply PReLU 1652 to the output of 3×3 convolution 1650, and may apply 3×3 convolution 1654 to the output of PReLU 1652. Video encoder 200 and video decoder 300 may crop the output of 3×3 convolution 1654 to generate filtered reconstructed UV (REC UV) samples of the image. Video encoder 200 and video decoder 300 can perform pixel reconstruction 1656 on the output of a 3×3 convolution 1654 and crop the output of the 1660 pixel reconstruction 1656 to generate a filtered reconstructed Y sample (RECY) of the image. For example, pixel reconstruction 1656 can upsample its input so that the output of pixel reconstruction 1656 has a size of w*2, h*2, c / 4, where w is the width, h is the input height, and c is the number of input channels to the input of pixel reconstruction 1656. In some examples, Figure 16A The example parameters include one or more of the following: d1=192, d2=32, d3=16, d4=16, d5=16, C=64, and d6=48.

[0171] Figure 16B This is a block diagram illustrating the example main block. Figure 16B The backbone block 1670 can be an example of any of the backbone blocks 1630A to 1630N. The backbone block 1670 may include multi-scale branches 1619 in which convolutions are performed in parallel. For example, a 1×1 convolution 1601 may be performed in parallel with a 3×1 convolution 1603, followed by a 1×3 convolution 1605. The multi-scale branches 1619 may perform multi-scale feature extraction.

[0172] Video encoder 200 and video decoder 300 can apply PReLU 1607 to the output multi-scale branch 1619. Video encoder 200 and video decoder 300 can apply a 1×1 convolution 1611 to the output of PReLU 1607. Video decoder 300 can apply a 1×3 convolution 1613 to the output of 1×1 convolution 1611. Video decoder 300 can apply a 3×1 convolution 1615 to the output of 1×3 convolution 1613. In some examples, the parameters in the example of Figure 16 include C=64, C1=160, C21=32, C22=32, and C31=64.

[0173] Figure 17This is a block diagram illustrating an example unified filter with separate luminance and chrominance models. Specifically, Figure 17 An example model for luminance samples is shown. Figure 17 NN-based filters may include IPB 1700, QP BASE 1702, BS1704, PRED 1706 and reconstructed luminance samples (REC) EXT Y)1708. The video encoder 200 and video decoder 300 can apply the corresponding 3×3 convolutions of 1710A to 1710E to each of the inputs. The video encoder 200 and video decoder 300 can apply the corresponding PReLUs of 1712A to 1712E to the outputs of the 3×3 convolutions of 1710A to 1710E.

[0174] The fusion block 1716 can fuse feature maps using a 1×1 convolution 1718 and PReLU 1720. The transformation block 1722 then processes the fused data using a 3×3 convolution 1724 and PReLU 1726.

[0175] Video encoder 200 and video decoder 300 may apply backbone block group 1728. Backbone block 1728 may include multiple backbone blocks 1730A to 1730N. In some examples, N=20, resulting in 20 filter blocks. Video encoder 200 and video decoder 300 may apply 3×3 convolution 1750 to the output of backbone block group 1728 and may apply PReLU 1752 to the output of 3×3 convolution 1750. Video encoder 200 and video decoder 300 may apply 3×3 convolution 1754 to the output of PReLU 1752. Video encoder 200 and video decoder 300 may perform pixel reconstruction 1756 on the output of 3×3 convolution 1754 and crop 1760 pixels of the reconstructed 1756 output to generate a filtered reconstructed luminance (REC Y) sample of the image.

[0176] One or more of the backbone blocks 1730A to 1730N in the backbone block group 1728 may be Figure 16B An example of the main block 1670. In some examples, Figure 17 Examples and Figure 16B The parameters for the example may include one or more of the following: d1=192, d2=32, d3=16, d4=16, d5=16, C=64, C1=160, C 21 =32、C 22 =32、C 31 =64, N=20, and d6=48.

[0177] Parameters d1, d2, d3, d4, and d5 indicate the number of channels (e.g., features) trained for each convolution from each corresponding input data. For example, the number of d1 channels in a backbone block 1710E (e.g., a 3×3 convolution) is the number of channels for which reconstructed pixel blocks (e.g., RECs) are applied. EXT Y 1708). Parameter C is the number of channels produced as the output of the fusion / transformation block. C can also be used as the number of channels used for the backbone block (e.g., the backbone block may have a number C of input channels). Parameters C1, C 21 C 22 and C 31 This refers to the number of channels at a specific module within the backbone block, as shown in the figure. Parameter N is the number of backbone blocks.

[0178] Figure 18 This is a block diagram illustrating an example unified filter with separate luminance and chrominance models (chrominance). Specifically, Figure 18 An example model for chromaticity (UV) samples is shown. Figure 18 The input to the NN-based filter may include REC EXT Y 1800 (its factor of 2 that can be downsampled), QP BASE 1804, BS 1806, PRED 1808 and reconstructed chromaticity samples (REC) EXT (UV) 1809. The video encoder 200 and video decoder 300 can apply the corresponding 3×3 convolutions of 1810A to 1810E to each of the inputs. The video encoder 200 and video decoder 300 can apply the corresponding PReLUs of 1812A to 1812E to the outputs of the 3×3 convolutions of 1810A to 1810E.

[0179] The fusion block 1816 uses a 1×1 convolution 1818 and PReLU 1820 to fuse feature maps. The transformation block 1822 then processes the fused data using a 3×3 convolution 1824 and PReLU 1826.

[0180] Video encoder 200 and video decoder 300 may apply backbone block group 1828. Backbone block 1828 may include multiple backbone blocks 1830A to 1830N. In some examples, N=16, resulting in 16 backbone blocks. Video encoder 200 and video decoder 300 may apply 3×3 convolution 1850 to the output of backbone block group 1828. Video encoder 200 and video decoder 300 may apply PReLU 1852 to the output of 3×3 convolution 1850. Video encoder 200 and video decoder 300 may apply 3×3 convolution 1854 to the output of PReLU 1852. Video encoder 200 and video decoder 300 may perform pixel reconstruction 1856 on the output of 3×3 convolution 1854 and crop 1860 pixels of the reconstructed 1856 output to generate filtered reconstructed chroma (RECUV) samples of the image.

[0181] One or more of the backbone blocks 1830A to 1830N of backbone block group 1828 can be used. Figure 16B The backbone architecture. In some examples, Figure 18 The example parameters and Figure 16B A group may include one or more of the following: d1=192, d2=32, d3=16, d4=16, d5=16, C=64, C1=160, C 21 =32、C 22 =32、C 31 =64, N=16, and d6=48.

[0182] Low-complexity ILF architecture To achieve a complexity level of less than 20k MAC / pixel and preserve most of the decoding performance, an alternative unified filter architecture with canonical multivariate (CP) decomposition with separable convolutions has been proposed and applied to replace 3×3 convolutions in the backbone network, and its implementation is shown in... Figure 19A middle. Specifically, Figure 19A Examples of the proposed architecture and parameters (such as the number of inputs, features, channels, and residual blocks) are shown. For a given configuration, the complexity is 19.55 kMAC.

[0183] Figure 19A The architecture can be used to implement low-complexity in-loop filters for joint YCbCr 4:2:0 processing. Figure 19A The example also applies to processing the Y,CbCr components separately. Figure 19A An example could be a uniform filter. A uniform filter could include a filter with at least one quality input (such as a QP input) such that the filter can be used to handle different quality levels. (See also:) Figure 19A As observed, the unified filter possesses QP SLICE1902 Input and QP BASE 1904 inputs to both. A unified filter based on neural networks (such as...) Figure 19A The NN-based unified filter can be trained using a variety of image and / or video data at different quality levels.

[0184] Figure 19A The architecture includes a header block 1980, a transition block 1922, one or more trunk block groups 1928, and a tail block 1986. The inputs to the header block 1980 include IPB 1900 and QP. SLICE 1902, QP BASE 1904, BS 1906, PRED 1908, and REC1909. Reconstructed sample (REC) 1909 may include luminance samples (REC... EXT Y) and twice the upsampled chromaticity sample (REC) EXT Both UV and UV.

[0185] Head block 1980 includes blocks preceding fusion block 1916. Figure 19A The filter block. The head block 1980 includes 3×3 convolutions 1910A to 1910F, which are applied to the corresponding inputs and can extract feature maps from the inputs.

[0186] Fusion block 1916 can fuse feature maps using 1×1 convolution 1918 and PReLU 1920. Transformation block 1922 can include blocks that occur after fusion block 1916 and before any backbone or filter blocks (e.g., before backbone blocks 1930A to 1930N). Transformation block 1922 processes the fused data using 3×3 convolution 1924 and PReLU 1926. Backbone block group 1928 can include backbone blocks 1930A to 1930N. In some examples, backbone blocks 1930A to 1930N may provide... Figure 19A The filtering function of the filter. In some examples, the trunk blocks 1930A to 1930N may be located after the transition block 1922 and before the tail block 1986.

[0187] Tail block 1986 may be included after all trunk blocks or filter blocks (e.g., after trunk blocks 1930A to 1930N). Figure 19A The filter block. The tail block 1986 may include a 3×3 convolution 1950, a PReLU 1952, a 3×3 convolution 1954, a cropping unit 1958, a pixel recombination unit 1956, and a cropping unit 1960.

[0188] Examples of the architecture and parameters (such as the number of inputs, features, channels, and residual blocks) are shown in Figure 19A In some examples, Figure 19AThe example parameters include one or more of the following: d1=12, d2=8, d3=4, d4=2, d5=2, C=24, N=11, and d6=24. Figure 19A The complexity of the architecture shown is approximately 19.55kMAC.

[0189] Figure 19B This illustrates one or more aspects of this disclosure. Figure 19A A block diagram of an example backbone block. Backbone block 1970 could be... Figure 19A Examples of any of the backbone blocks 1930A to 1930N. Backbone block 1970 differs from... Figure 16B The UF backbone block 1670 shown is replaced by CP decomposition in backbone block 1970 because backbone block 1970 does not include multi-scale branches like those in backbone block 1670, and the multidimensional convolutions replaced by convolutions 1611, 1613, and 1615 are replaced by CP decompositions in backbone block 1970. Backbone block 1970 includes 1×1 convolutions 1971 and PReLU 1972. The CP decompositions in backbone block 1970 include 1×1 convolutions 1973, 1×3 convolutions 1974a, 3×1 convolutions 1975, and 1×1 convolutions 1976. In some examples, Figure 19B The parameters are C=24, C1=72 and C21=24.

[0190] Figure 19C This illustrates one or more aspects of this disclosure. Figure 19A A block diagram of an example CP decomposition of a 3×3 convolution. CP decomposition 1979 may include 1×1 convolution 1981, 1×3 convolution 1982, 3×1 convolution 1983, and 1×1 convolution 1984. For example, CP decomposition 1979 may replace the 3×3 convolution 1950 of the tail block 1986. In some examples, Figure 19C The parameters are C=24, C1=72 and C21=24.

[0191] Figure 20A This is a block diagram illustrating an example of a low-complexity in-loop filter architecture with luminance / chrominance separation. Figure 20A In this process, the processing branches are divided into separate luminance and chrominance branches, which allows for a better balance of luminance / chrominance performance.

[0192] Figure 20A The architecture includes a header block 2080, a fusion block 2022, one or more backbone blocks 2028 for chroma, one or more backbone blocks 2030 for luma and chroma, a tail block 2086, and a tail block 2088. The inputs to the header block 2080 include IPB 2000 and QP... SLICE 2002, QP BASE2004, BS 2006, PRED 2008 and REC 2009 (e.g., both luminance and chrominance reconstruction samples).

[0193] Header block 2080 is included before cascade 2014. Figure 20A The filter block. The header block 2080 includes 1×1 convolutions 2010A to 2010D and 3×3 convolutions 2010E to 2010F, which are applied to the corresponding inputs.

[0194] The fusion block 2022 can fuse the feature maps generated by the head block 2080 using a 1×1 convolution 2018 and PReLU 2020. The fusion block 2022 can also process the output of PReLU 2020 by applying a 3×3 convolution 2024 and PReLU 2026. One or more backbone blocks 2028 for chroma can provide chroma samples. Figure 20A The filtering function of the filter. One or more backbone blocks 2030 can provide both luminance and chrominance samples together. Figure 20A The filtering function of the filter.

[0195] The tail block 2086 may include the trunk block (e.g., one or more trunk blocks 2028 for chroma and one or more trunk blocks 2030 for luminance and chroma) following the main block. Figure 20A The tail block 2086 may include a separable convolution 2050 (which may include 1×3 and 3×1 convolutions), a PReLU 2052, and a 3×3 convolution 2054. The tail block 2088 may include a separable convolution 2060 (which may include 1×3 and 3×1 convolutions), a PReLU 2062, and a 3×3 convolution 2064. The output of the tail block 2088 may be input to the pixel reconstruction unit 2066.

[0196] Figure 20B Based on one or more aspects of this disclosure Figure 20B Example backbone block. Backbone block 2079 can be... Figure 20A Examples of backbone blocks 2028 for chroma and / or 2030 for luminance and chroma. Backbone block 2079 includes 1×1 convolution 2082, PReLU 2084, 1×1 convolution 2087, 1×3 separable convolution 2089, 3×1 separable convolution 2090, and 1×1 convolution 209. In some examples, Figure 20B The parameters are C=24, C1=72 and C 21 =24.

[0197] question Figure 19A , Figure 19B , Figure 19C , Figure 20A and Figure 20B The low-complexity unified filter (UF) architecture shown typically offers a good complexity-performance tradeoff in the complexity range of 15 kMAC to 500 kMAC / pixel. Targeting even lower complexity (e.g., in the range of 1 kMAC to 5 kMAC / pixel) can result in a significantly reduced backbone block count (e.g., down to 4 backbone blocks for luminance samples and one backbone block for chrominance samples). However, this number of layers in the backbone block does not provide sufficient spatial filtering support (reach range) for most applications.

[0198] Spatial filtering support refers to the effective size or area covered by the filter on the input data. Spatial filtering support can be defined by the filter's dimensions (e.g., a 3×3 filter has 3×3 spatial support). Spatial filtering support determines how much input the filter examines at a time, affecting the granularity of features the filter can process. Spatial filtering support also refers to the maximum value of the tap length. Range of arrival describes the distance the filter's effect can propagate through layers of the network. While spatial filtering support is related to a single layer of the network, range of arrival considers the cumulative effect of applying the filter across multiple layers.

[0199] It is also noted that at the 1kMAC / pixel level, the complexity of the fusion block, which can be made using multidimensional non-separable 3×3 convolutions, contributes up to 40% of the total complexity.

[0200] Example To improve complexity within a very low complexity range (e.g., 1kMAC to 10kMAC / pixel) – a performance tradeoff – alternative ILF architectures are described below. Example architectures and parameter configurations for the backbone and fusion blocks are described below. Each example below can be used individually or in combination with each other.

[0201] Low-complexity backbone First, a low-complexity backbone architecture is described. Figures 21A to 21C and Figures 22A to 22C The example backbone architecture shown can be used to replace any backbone block described in the diagram above. As two examples, Figures 21A to 21C and Figures 22A to 22C Any backbone architecture shown can be used to replace the backbone architecture. Figure 19A The main components 1930A to 1930N, or those that can be used to replace them. Figure 20A The backbone block is 2028 or 2038.

[0202] Generally speaking, according to the technology of this disclosure, the complexity of the backbone block is reduced by resampling a multidimensional separable transformation implementation. To resampling the multidimensional separable transformation implementation, the multidimensional 3×3 convolution is simplified to a 3-component one-dimensional (1D) decomposition. Since the 3-stage decomposition employs feature reduction (e.g., a convolutional layer outputs C1 channels, which is smaller than the input of C channels), the technology of this disclosure uses an alternating order decomposition through a pair of backbone blocks. That is, a single backbone block comprises a backbone block pair (e.g., a first backbone block and a second backbone block). Generally speaking, the following... Figures 21A to 21C and Figures 22A to 22C The described backbone block is based on a filter portion of an NN and includes a pair of backbone blocks, each of which includes a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0203] In one example, the first backbone block in this pair could begin with a 1×3 convolution with feature reduction (e.g., C1 outputs) followed by a 3×1 convolution with feature expansion (e.g., back to C outputs). The second backbone in this pair would use a reverse horizontal / vertical decomposition. In one example, an activation layer (e.g., a PReLU layer) could be applied in the final stage of the decomposition. An example of the proposed low-complexity backbone block is shown in... Figure 21A middle.

[0204] Figure 21A This is a block diagram illustrating a first example of a backbone block pair 2100A according to the technology of this disclosure. The backbone block pair 2100A may be referred to as a type 1 (T1) backbone block pair. The backbone block pair 2100A may include a first backbone block 2103 and a second backbone block 2113. The first backbone block 2103 of the backbone block pair 2100A may receive an input tensor 2102 (e.g., also referred to as a feature map) in the format [H, W, C], where H is the height of the input tensor, W is the width of the input tensor, and C is the number of channels of the input tensor.

[0205] The input tensor is first processed by a convolutional layer 2104, which is a 1×1 convolution with C output channels. Subsequently, the first backbone block 2103 processes the output of the convolutional layer 2104 using a pair of separable convolutions. Figure 21AThe example illustrates a vertically separable convolution 2106 (e.g., a 1×3 convolution) with C1 output channels, followed by a horizontally separable convolution 2108 (e.g., a 3×1 convolution) with C output channels. Since C1 is defined as less than C, the separable convolution 2106 performs feature reduction, thereby reducing the complexity of the backbone block 2100A. The separable convolution 2108 then performs feature expansion back to C output channels. After the separable convolution 2108, the first backbone block 2103 processes the output of the separable convolution 2108 with PReLU 2110. Generally, PReLU 2110 is an activation layer, which in different specific implementations can be implemented as PReLU, ReLU, or another layer that can introduce nonlinearity into the CNN model.

[0206] The output of the first backbone block 2103 is concatenated with the input tensor at the cascade 2112. The concatenated tensor is then fed to the second backbone block 2113 and processed by the convolutional layer 2114, which is a 1×1 convolution with C output channels. The second backbone block 2113 is configured in a manner similar to the first backbone block 2103. However, the order of the separable convolutions in the second backbone block 2113 is reversed. Figure 21A The example illustrates a horizontally separable convolution 2116 (e.g., a 3×1 convolution) with C1 output channels, followed by a vertically separable convolution 2118 (e.g., a 1×3 convolution) with C output channels. Since C1 is defined as less than C, the separable convolution 2116 performs feature reduction, thereby reducing the complexity of the backbone block pair 2100A. The separable convolution 2118 then performs feature expansion back to C output channels. After the separable convolution 2118, a second backbone block 2113 processes the output of the separable convolution 2118 with PReLU 2120. The output of PReLU 2120 is cascaded with the output of cascade 2112 via cascade 2122. The output of the backbone block pair 2100A can then be fed to another backbone block pair or to additional layers in the NN filter.

[0207] all in all, Figure 21A The diagram illustrates a backbone block pair 2100A performing a three-component 1D decomposition. The input to the backbone block pair 2100A is defined by its height (h), width (w), and number of channels (C). The backbone block pair 2100A performs the three-component 1D decomposition using a 1×1 convolution with C output channels, a first separable convolution with C1 output channels in a first direction (where C1 is less than C), and a second separable convolution with C output channels in a second direction. The three-component 1D decomposition of the multidimensional convolutions in the backbone block pair 2100A also includes activation functions (such as PReLU).

[0208] More specifically, the backbone pair 2100A may include a first backbone 2103 having a first 1×1 convolution (e.g., convolutional layer 2104) with C output channels, a first separable convolution (e.g., separable convolution 2106) with C1 output channels (where C1 is less than C) in a first direction, and a second separable convolution (e.g., separable convolution 2108) with C output channels in a second direction. The backbone pair 2100A may also include a second backbone 2113 having a second 1×1 convolution (e.g., convolutional layer 2114) with C output channels, a third separable convolution (e.g., separable convolutional layer 2116) with C1 output channels (where C1 is less than C) in a second direction, and a fourth separable convolution (e.g., separable convolutional layer 2118) with C output channels in the first direction. Figure 21A In the example, the first direction is vertical and the second direction is horizontal.

[0209] Figure 21A The example shows that the first backbone block 2103 performs separable convolutions in the vertical direction and subsequently in the horizontal direction, where feature reduction is performed by separable convolutions in the vertical direction. The second backbone block 2113 then performs separable convolutions in the opposite direction. Figure 21B The text presents an alternative design for the backbone blocks, illustrating a second example of a backbone block pair where the direction of the separable convolutions is reversed.

[0210] like Figure 21B As shown, the backbone block pair 2100B may include a first backbone block 2103 having a first 1×1 convolution (e.g., convolutional layer 2104) with C output channels, a first separable convolution (e.g., separable convolution 2107) with C1 output channels (where C1 is less than C) in a first direction (e.g., 3×1 or horizontal), and a second separable convolution (e.g., separable convolution 2109) with C output channels in a second direction (e.g., 1×3 or vertical). The backbone block pair 2100B may also include a second backbone block 2113 having a second 1×1 convolution (e.g., convolutional layer 2114) with C output channels, a third separable convolution (e.g., separable convolutional layer 2117) with C1 output channels (where C1 is less than C) in the second direction, and a fourth separable convolution (e.g., separable convolutional layer 2119) with C output channels in the first direction. Figure 21B In the example, the first direction is horizontal and the second direction is vertical. The other layers of the main block pair 2100B, which have the same reference numerals as the main block pair 2100A, remain the same.

[0211] Figure 21C This is an example used for Figure 21AA block diagram showing different arrangements of a 1×1 convolution. In some examples, the order of decomposition can be changed; for example, the 1×1 convolution can be placed in the third stage, followed by 3×1 and 1×3 spatial convolutions. Figure 21C An alternative arrangement of convolutional layer 2104 (or more generally, any 1×1 convolutional layer) in a portion of the backbone block 2100C is shown. Specifically, convolutional layer 2104 is positioned after separable convolutions 2106 and 2108 in the first backbone block 2103. Figure 21C The location of convolutional layer 2104 following separable convolution is shown only for the first backbone block 2103. However, it should be understood that convolutional layer 2104 or any 1×1 convolution can be located after the convolution discussed above. Figures 21A to 21C and / or as described below Figures 22A to 22C After two separable convolutions in any backbone block.

[0212] In summary, in one example, the first 1×1 convolution is positioned before the first separable convolution and the second 1×1 convolution is positioned before the third separable convolution. In another example, the first 1×1 convolution is positioned after the second separable convolution and the second 1×1 convolution is positioned before the fourth separable convolution.

[0213] Figures 22A to 22C This is a block diagram illustrating another example of a backbone block pair with different example arrangements of activation layers. Generally, the location of the activation layer (e.g., PReLU) does not need to be limited to the final stage of the backbone block in the backbone block pair, such as... Figures 21A to 21C As shown. In some examples, the activation layer may be placed after a 1×1 convolution within the first depth. In other examples, the activation layer may be placed immediately after the first spatial decomposition layer, or after each decomposition layer.

[0214] Figure 22A This is a block diagram illustrating another example of a backbone block pair with an activation layer, representing a first example arrangement. Figure 22A The structure is shown to be similar to Figure 21A The backbone block pair 2100A is the backbone block pair 2200A. However, in the backbone block pair 2200A, PReLU 2110 and PReLU 2120 are positioned after convolutional layer 2104 and convolutional layer 2114, respectively.

[0215] Figure 22B This is a block diagram illustrating another example of the backbone block arrangement of the 2200B with an activation layer. Figure 22B The location of PReLU 2110 after the separable convolution 2106 is shown only for the first backbone block 2103. However, it should be understood that PReLU 2110 or any activation layer can be located after... Figures 21A to 21CAfter the first separable convolution in any of the main branches of the main block.

[0216] Figure 22C This is a block diagram illustrating another example of a backbone block arrangement with multiple activation layers for the 2200C. Figure 22C Only for the first backbone block 2103 are PReLU 2150 and PReLU 2110 positioned after separable convolutions 2106 and 2108, respectively. However, it should be understood that two PReLUs or any two activation layers can be positioned after... Figures 21A to 21C After each of the two separable convolutions in any of the backbone blocks.

[0217] In summary, in one example, the first backbone block includes a first PReLU after a first 1×1 convolution, and the second backbone block includes a second PReLU after a second 1×1 convolution. In another example, the first backbone block includes a first PReLU after a second separable convolution, and the second backbone includes a second PReLU after a fourth separable convolution. In yet another example, the first backbone block includes a first PReLU after a first separable convolution and a second PReLU after a second separable convolution, and the second backbone block includes a third PReLU after a third separable convolution and a fourth PReLU after a fourth separable convolution.

[0218] Low-complexity fusion blocks In the second example of this disclosure, a method is proposed to reduce the complexity of the fusion block by employing separable convolutions, similar to the separable convolutions used in the backbone block described above. Figures 23A to 23E The example fusion block shown can be used to replace any fusion block in any example NN-based filter, including replacing... Figure 19A Merging block 1916 and Figure 20A The fusion block 2022 in the middle. Generally speaking, Figures 23A to 23E A fusion block for a neural network-based filter is shown, where the fusion block has two separable convolutions, each of which applies spatial downsampling. The implementation complexity of the fusion block is reduced by combining the separable convolutions and spatial downsampling.

[0219] exist Figures 23A to 23E The downsampling process, denoted as d2, is implemented at each stage of spatial processing (e.g., separable convolution). Aside from employing spatial downsampling at each stage of 1D decomposition, the design of the fusion block follows the same design principles as the backbone block described above. In some examples, fusion blocks with different decomposition orders may be used, similar to... Figures 21A to 21C and Figures 22A to 22C The main design shown.

[0220] In some examples, the dX parameter, which specifies the amount of downsampling in the blend block, can be an integer not equal to 2 (e.g., greater than 2). This downsampling parameter can take different values ​​for the luma or chroma components. In other examples, the blend block can be performed separately for the luma and chroma branches, thus allowing for optimal parameter selection for each component.

[0221] Therefore, in one example, the NN-based filter also includes a first fusion block for luminance samples, which has two separable convolutions, each applying a first spatial downsampling. The NN filter also includes a second fusion block for chrominance samples, which has two separable convolutions, each applying a second spatial downsampling. In one example, the first and second spatial downsampling apply the same downsampling amount. In another example, the first and second spatial downsampling apply different downsampling amounts.

[0222] Generally speaking, Figures 23A to 23C Examples of fusion blocks include 1×1 convolution, PReLU, a first separable convolution in a first direction (horizontal or vertical) with a first spatial downsampling amount, and a second separable convolution in a second direction (e.g., the opposite direction to the first direction) with a first spatial downsampling amount.

[0223] Figure 23A This is a block diagram illustrating a first example of a concrete implementation of the fusion block. Figure 23A An example fusion block 2300A is shown, which receives an input tensor T in (H, W, C4) format, where H is the height of the tensor, W is the width of the tensor, and C4 is the number of input channels to the module. Fusion block 2300A includes a convolutional layer 2302, which is a 1×1 convolution with C output channels. Following convolutional layer 2302 is PReLU 2304. After PReLU 2304, fusion block 2300A includes a first separable convolution 2306 (e.g., a 1×3 convolution in the vertical direction) with a d² downsampling amount (e.g., a downsampling amount of 2 or another integer). The output of the first separable convolution 2306 is a tensor with dimensions (H / 2, W, C). Fusion block 2300A also includes a second separable convolution 2308 (e.g., a 3×1 convolution in the horizontal direction) with the same d² downsampling amount. The output of the second separable convolution 2308 is a tensor with dimensions (H / 2, W / 2, C).

[0224] Figure 23B This is a block diagram illustrating a second example of a concrete implementation of the fusion block. (and) Figure 23A Compared to the fusion block 2300A, in Figure 23BIn this configuration, the fusion block 2300B has swapped separable convolutional layers 2308 and 2306. That is, the horizontal separable convolution is performed before the vertical separable convolution.

[0225] Figure 23C This is a block diagram illustrating a third example of a concrete implementation of the fusion block. Figure 23C In the process, the fusion block 2300C has PReLU 2304 as the final layer of the fusion block.

[0226] Figure 23D This is a block diagram illustrating a fourth example of a concrete implementation of the fusion block. Figure 23D In this example, fusion block 2300D has PReLU 2304 as the final layer of the fusion block. Furthermore, an additional PReLU 2305 is included after the separable convolution 2306. In other examples, the order of separable convolution 2306 and separable convolution 2308 may vary. Figure 23D The example swaps.

[0227] Figure 23E This is a block diagram illustrating the fifth example of a concrete implementation of the fusion block. Figure 23E In the middle, the fusion block 2300E has a convolutional layer 2302 after the separable convolution 2306 and the separable convolution 2308.

[0228] Example parameters In the third example of this disclosure, the complexity of the header block and the number of trunk blocks can be optimized to achieve the target complexity. In some examples, for a complexity level below 5k MAC / pixel, the following parameters are proposed. (Refer to...) Figure 20A The block diagram proposes the following configuration: Configuration 1: "D1":8, "D2":2, "D3":1, "D4":1, "D5":1, "D6":8, "N_Y":10, "N_UV":4, "C":4, "C1_Y":8, "C1_UV":8, "C21":4. This configuration offers a wider range of filter support at the cost of a reduced number of channels (C=4).

[0229] Configuration 2: "D1":8, "D2":2, "D3":1, "D4":1, "D5":1, "D6":8, "N_Y":4, "N_UV":1, "C":8, "C1_Y":8, "C1_UV":8, "C21":4. This configuration is characterized by a higher number of channels at the cost of a reduced number of backbone blocks, with N_Y=4 and N_UV=1.

[0230] Based on evaluation, the techniques proposed in this disclosure are applicable to in-loop filters with a wide range of complexities by selecting different parameter configurations.

[0231] The NNVC architecture employing the techniques disclosed herein reduces computational complexity and memory bandwidth requirements while providing competitive performance. The examples described in this document involve NN-assisted loop filtering. However, they are applicable to any NN-based video decoding tool that consumes input data with certain statistical properties, such as static content or sparse representation.

[0232] Figure 24 This is a block diagram illustrating an example video encoder 200 that can perform the techniques of this disclosure for low-complexity NN-based filtering. Figure 24 This disclosure is provided for illustrative purposes and should not be construed as a limitation on the techniques extensively illustrated and described herein. For illustrative purposes, this disclosure describes the video encoder 200 in accordance with the techniques of VVC and HEVC. However, the techniques of this disclosure can be performed by video encoding devices configured for other video decoding standards and video decoding formats, such as AV1 and subsequent formats of AV1 video decoding.

[0233] exist Figure 24 In the example, the video encoder 200 includes a video data memory 230, a mode selection unit 202, a residual generation unit 204, a transform processing unit 206, a quantization unit 208, an inverse quantization unit 210, an inverse transform processing unit 212, a reconstruction unit 214, a filter unit 216, a decoded picture buffer (DPB) 218, and an entropy coding unit 220. Any or all of the video data memory 230, mode selection unit 202, residual generation unit 204, transform processing unit 206, quantization unit 208, inverse quantization unit 210, inverse transform processing unit 212, reconstruction unit 214, filter unit 216, DPB 218, and entropy coding unit 220 may be implemented in one or more processors or in processing circuitry. For example, the units of the video encoder 200 may be implemented as one or more circuit or logic elements as part of hardware circuitry, or as part of a processor, ASIC, or FPGA. Furthermore, the video encoder 200 may include additional or alternative processors or processing circuitry to perform these and other functions.

[0234] Video data storage 230 is an example of a storage system capable of storing video data to be encoded by components of video encoder 200. Video encoder 200 can receive data from, for example, video source 104 (…). Figure 1The video data memory 230 receives video data stored in the video data memory 230. The DPB 218 is an example of a memory system that can act as a reference picture memory, storing reference video data for use when the video encoder 200 predicts subsequent video data. The video data memory 230 and DPB 218 can each be formed from any of one or more memory devices or memory cells, such as dynamic random access memory (DRAM) (including synchronous DRAM (SDRAM)), magnetoresistive RAM (MRAM), resistive RAM (RRAM), or other types of memory devices. The video data memory 230 and DPB 218 can be provided by the same memory device or separate memory devices. In various examples, the video data memory 230 can be on-chip (as illustrated) with other components of the video encoder 200, or off-chip relative to those components.

[0235] In this disclosure, references to video data memory 230 should not be construed as limited to memory inside video encoder 200 (unless specifically described) or memory outside video encoder 200 (unless specifically described). Rather, references to video data memory 230 should be understood as a reference memory that stores video data received by video encoder 200 for encoding (e.g., video data for the current block to be encoded). Figure 1 The memory 106 can also provide temporary storage for the outputs from various units of the video encoder 200.

[0236] Examples Figure 24 Various units help understand the operations performed by the video encoder 200. Units can be implemented as fixed-function circuits, programmable circuits, or combinations thereof. Fixed-function circuits are circuits that provide specific functionality and are pre-defined for the operations that can be performed. Programmable circuits are circuits that can be programmed to perform various tasks and provide flexible functionality for the operations that can be performed. For example, a programmable circuit can execute software or firmware that causes the programmable circuit to operate in a manner defined by the instructions of the software or firmware. Fixed-function circuits can execute software instructions (e.g., to receive or output parameters), but the type of operation performed by a fixed-function circuit is generally immutable. In some examples, one or more units in the unit may be different circuit blocks (fixed-function or programmable), and in some examples, one or more units in the unit may be integrated circuits.

[0237] The video encoder 200 may include an arithmetic logic unit (ALU), an essential function unit (EFU), digital circuitry, analog circuitry, and / or a programmable core, all formed by programmable circuitry. In an example where the operation of the video encoder 200 is performed using software executed by programmable circuitry, memory 106 ( Figure 1 The video encoder 200 may store instructions (e.g., target code) of the software received and executed by the video encoder 200, or another memory (not shown) within the video encoder 200 may store such instructions.

[0238] The video data storage unit 230 is configured to store received video data. The video encoder 200 can retrieve images of the video data from the video data storage unit 230 and provide the video data to the residual generation unit 204 and the mode selection unit 202. The video data in the video data storage unit 230 can be raw video data to be encoded.

[0239] The mode selection unit 202 includes a motion estimation unit 222, a motion compensation unit 224, and an intra-frame prediction unit 226. The mode selection unit 202 may include additional functional units for performing video prediction based on other prediction modes. As an example, the mode selection unit 202 may include a palette unit, an intra-frame block copying unit (which may be part of the motion estimation unit 222 and / or the motion compensation unit 224), an affine unit, a linear model (LM) unit, etc.

[0240] Mode selection unit 202 typically coordinates multiple coding channels to test combinations of coding parameters and the resulting rate-distortion values ​​for such combinations. Coding parameters may include the CTU-CU partitioning, the prediction mode for the CU, the transformation type of the residual data for the CU, the quantization parameters of the residual data for the CU, etc. Mode selection unit 202 can ultimately select a combination of coding parameters that has a better rate-distortion value compared to other tested combinations.

[0241] The video encoder 200 can divide an image retrieved from the video data storage 230 into a series of CTUs and encapsulate one or more CTUs within a slice. The mode selection unit 202 can divide the image's CTUs according to the tree structure described above (such as an MTT structure, QTBT structure, superblock structure, or the quadtree structure described above). As described above, the video encoder 200 can form one or more CUs by dividing the CTUs according to the tree structure. Such CUs are also commonly referred to as "video blocks" or "blocks".

[0242] Generally, mode selection unit 202 also controls its components (e.g., motion estimation unit 222, motion compensation unit 224, and intra-prediction unit 226) to generate prediction blocks for the current block (e.g., the current CU, or, in HEVC, the overlapping portion of PU and TU). To perform inter-frame prediction for the current block, motion estimation unit 222 may perform a motion search to identify one or more closely matching reference blocks in one or more reference pictures (e.g., one or more previously decoded pictures stored in DPB 218). Specifically, motion estimation unit 222 may calculate values ​​representing the similarity between a potential reference block and the current block, for example, based on sum of absolute differences (SAD), sum of squared differences (SSD), mean absolute difference (MAD), mean squared difference (MSD), etc. Motion estimation unit 222 may typically perform these calculations using sample-by-sample differences between the current block and the reference blocks under consideration. Motion estimation unit 222 may identify reference blocks with the lowest values ​​produced by these calculations to indicate the reference block that best matches the current block.

[0243] Motion estimation unit 222 can generate one or more motion vectors (MVs) that define the location of a reference block in a reference image relative to the location of the current block in the current image. Motion estimation unit 222 can then provide the motion vectors to motion compensation unit 224. For example, for unidirectional inter-frame prediction, motion estimation unit 222 can provide a single motion vector, while for bidirectional inter-frame prediction, motion estimation unit 222 can provide two motion vectors. Motion compensation unit 224 can then use the motion vectors to generate a prediction block. For example, motion compensation unit 224 can use the motion vectors to retrieve data for the reference block. As another example, if the motion vectors have fractional sample accuracy, motion compensation unit 224 can interpolate the prediction block according to one or more interpolation filters. Furthermore, for bidirectional inter-frame prediction, motion compensation unit 224 can retrieve data for two reference blocks identified by corresponding motion vectors and combine the retrieved data, for example, by per-sample averaging or weighted averaging.

[0244] When operating according to the AV1 video decoding format, the motion estimation unit 222 and the motion compensation unit 224 can be configured to encode the decoded blocks of video data (e.g., both luma and chroma decoded blocks) using translational motion compensation, affine motion compensation, overlap block motion compensation (OBMC), and / or composite inter-intra-frame prediction.

[0245] As another example, for intra-prediction or intra-prediction decoding, intra-prediction unit 226 may generate a prediction block from samples adjacent to the current block. For example, for directional mode, intra-prediction unit 226 may typically mathematically combine the values ​​of adjacent samples and fill these calculated values ​​across the current block in a defined direction to produce a prediction block. As another example, for DC mode, intra-prediction unit 226 may calculate the average of the adjacent samples of the current block and generate a prediction block to include the resulting average for each sample of the prediction block.

[0246] When operating according to the AV1 video decoding format, the intra-frame prediction unit 226 can be configured to encode decoded blocks of video data (e.g., both luma and chroma decoded blocks) using directional intra-frame prediction, non-directional intra-frame prediction, recursive filter intra-frame prediction, luma-chroma (CFL) prediction, intra-block copying (IBC), and / or palette modes. The mode selection unit 202 may include additional functional units for performing video prediction based on other prediction modes.

[0247] Mode selection unit 202 provides a prediction block to residual generation unit 204. Residual generation unit 204 receives an uncoded raw version of the current block from video data memory 230 and a prediction block from mode selection unit 202. Residual generation unit 204 calculates the sample-by-sample difference between the current block and the prediction block. The resulting sample-by-sample difference defines the residual block for the current block. In some examples, residual generation unit 204 may also determine the differences between sample values ​​in the residual block to generate the residual block using residual differential pulse decoding modulation (RDPCM). In some examples, residual generation unit 204 may be formed using one or more subtractor circuits performing binary subtraction.

[0248] In the example where mode selection unit 202 divides a CU into PUs, each PU can be associated with a luma prediction unit and a corresponding chroma prediction unit. Video encoder 200 and video decoder 300 can support PUs of various sizes. As noted above, the size of a CU can refer to the size of its luma decoding block, while the size of a PU can refer to the size of its luma prediction unit. Assuming a particular CU size is 2N×2N, video encoder 200 can support PU sizes of 2N×2N or N×N for intra-frame prediction, and symmetric PU sizes of 2N×2N, 2N×N, N×2N, N×N, or similar for inter-frame prediction. Video encoder 200 and video decoder 300 can also support asymmetric partitioning for PU sizes of 2N×nU, 2N×nD, nL×2N, and nR×2N for inter-frame prediction.

[0249] In an example where mode selection unit 202 does not further divide the CU into PUs, each CU can be associated with a luminance decoding block and a corresponding chrominance decoding block. As mentioned above, the size of the CU can refer to the size of the luminance decoding block of the CU. The video encoder 200 and the video decoder 300 can support CU sizes of 2N×2N, 2N×N, or N×2N.

[0250] For other video decoding techniques, such as intra-block copy mode decoding, affine mode decoding, and linear model (LM) mode decoding, as some examples, mode selection unit 202 generates a prediction block for the current block being encoded via a corresponding unit associated with the decoding technique. In some examples (such as palette mode decoding), mode selection unit 202 may not generate a prediction block, but instead generate syntax elements indicating how the block is reconstructed based on a selected palette. In such modes, mode selection unit 202 may provide these syntax elements to entropy coding unit 220 for encoding.

[0251] As described above, the residual generation unit 204 receives video data for the current block and the corresponding prediction block. Then, the residual generation unit 204 generates a residual block for the current block. To generate the residual block, the residual generation unit 204 calculates the sample-by-sample difference between the prediction block and the current block.

[0252] Transform processing unit 206 applies one or more transformations to the residual block to generate a block of transform coefficients (referred to herein as a "transform coefficient block"). Transform processing unit 206 may apply various transformations to the residual block to form the transform coefficient block. For example, transform processing unit 206 may apply a discrete cosine transform (DCT), direction transform, Karhunen-Loeve transform (KLT), or conceptually similar transformations to the residual block. In some examples, transform processing unit 206 may perform multiple transformations on the residual block, such as primary and secondary transformations (e.g., rotation transformations). In some examples, transform processing unit 206 does not apply any transformations to the residual block.

[0253] When operating according to AV1, transform processing unit 206 may apply one or more transforms to the residual block to generate a block of transform coefficients (referred to herein as a "transform coefficient block"). Transform processing unit 206 may apply various transforms to the residual block to form the transform coefficient block. For example, transform processing unit 206 may apply a combination of horizontal / vertical transforms, which may include the Discrete Cosine Transform (DCT), the Asymmetric Discrete Sine Transform (ADST), the Reversed ADST (e.g., ADST in reverse order), and the Identity Transform (IDTX). When using the Identity Transform, the transform is skipped in either the vertical or horizontal direction. In some examples, transform processing may be skipped entirely.

[0254] Quantization unit 208 quantizes the transform coefficients in a transform coefficient block to produce a quantized transform coefficient block. Quantization unit 208 quantizes the transform coefficients of the transform coefficient block according to the quantization parameter (QP) value associated with the current block. Video encoder 200 (e.g., via mode selection unit 202) can adjust the degree of quantization applied to the transform coefficient block associated with the current block by adjusting the QP value associated with the CU. Quantization may cause information loss, and therefore, the quantized transform coefficients may have lower accuracy compared to the original transform coefficients produced by transform processing unit 206.

[0255] The inverse quantization unit 210 and the inverse transform processing unit 212 can apply inverse quantization and inverse transform, respectively, to the quantized transform coefficient block to reconstruct the residual block based on the transform coefficient block. The reconstruction unit 214 can generate a reconstructed block corresponding to the current block (although potentially with some degree of distortion) based on the reconstructed residual block and the prediction block generated by the mode selection unit 202. For example, the reconstruction unit 214 can add samples of the reconstructed residual block to corresponding samples of the prediction block generated by the mode selection unit 202 to generate the reconstructed block.

[0256] Filter unit 216 may perform one or more filtering operations on the reconstructed block. For example, filter unit 216 may perform a deblocking operation to reduce block artifacts along the edges of the CU. In some examples, the operation of filter unit 216 may be skipped.

[0257] When operating according to AV1, filter unit 216 may perform one or more filtering operations on the reconstructed block. For example, filter unit 216 may perform a deblocking operation to reduce block artifacts along the edges of the CU. In other examples, filter unit 216 may apply a constrained direction enhancement filter (CDEF) after deblocking and may include the application of a non-separable, nonlinear, low-pass directional filter based on the estimated edge direction. Filter unit 216 may also include a loop recovery filter applied after CDEF and may include a separable symmetric normalized Wiener filter or a dual-guided filter.

[0258] Filter unit 216 can also be configured to implement any combination of the NN-based filters described above. Specifically, filter unit 216 can be configured to perform an NN-based filtering process on one or more blocks of reconstructed images of video data using NN-based filters, wherein the NN-based filters include a pair of backbone blocks, each of which includes a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction. Filter unit 216 can be further configured to perform an NN-based filtering process using NN-based filters, which also include a fusion block with two separable convolutions, each of which applies spatial downsampling.

[0259] The video encoder 200 stores reconstructed blocks in the DPB 218. For example, in an example where the filter unit 216 is not operated, the reconstruction unit 214 may store reconstructed blocks in the DPB 218. In an example where the filter unit 216 is operated, the filter unit 216 may store filtered reconstructed blocks in the DPB 218. The motion estimation unit 222 and the motion compensation unit 224 may retrieve reference images formed by the reconstructed (and potentially filtered) blocks from the DPB 218 to perform inter-frame prediction for blocks of subsequent encoded images. Additionally, the intra-frame prediction unit 226 may use the reconstructed blocks of the current image in the DPB 218 to perform intra-frame prediction for other blocks in the current image.

[0260] Generally, entropy coding unit 220 can entropy code syntax elements received from other functional components of video encoder 200. For example, entropy coding unit 220 can entropy code quantized transform coefficient blocks from quantization unit 208. As another example, entropy coding unit 220 can entropy code predictive syntax elements (e.g., motion information for inter-frame prediction or intra-frame mode information for intra-frame prediction) from mode selection unit 202. Entropy coding unit 220 can perform one or more entropy coding operations on syntax elements (another example of video data) to generate entropy-coded data. For example, entropy coding unit 220 can perform context-adaptive variable-length decoding (CAVLC), CABAC, variable-to-variable (V2V) length decoding, syntax-based context-adaptive binary arithmetic decoding (SBAC), probability interval partitioning entropy (PIPE) decoding, exponential Golomb coding, or another type of entropy coding operation on the data. In some examples, entropy coding unit 220 can operate in a bypass mode where syntax elements are not entropy-coded.

[0261] The video encoder 200 can output a bitstream that includes the entropy coding syntax elements required to reconstruct slices or blocks of images. Specifically, the entropy coding unit 220 can output a bitstream.

[0262] According to AV1, entropy coding unit 220 can be configured as a symbol-to-symbol adaptive multi-symbol arithmetic decoder. The syntax elements in AV1 consist of an N-element alphabet, and the context (e.g., a probability model) consists of a set of N probabilities. Entropy coding unit 220 can store the probabilities as an n-bit (e.g., 15-bit) cumulative distribution function (CDF). Entropy coding unit 220 can perform recursive scaling using an update factor based on the alphabet size to update the context.

[0263] The operations described above are relative to blocks. This description should be understood as operations applied to luma decoding blocks and / or chroma decoding blocks. As described above, in some examples, the luma decoding block and chroma decoding block are the luma and chroma components of the CU. In some examples, the luma decoding block and chroma decoding block are the luma and chroma components of the PU.

[0264] In some examples, it is not necessary to repeat the operations performed relative to the luma decoder for the chroma decoder block. As an example, the operations for identifying the motion vector (MV) and reference image of the luma decoder block do not need to repeat the MV and reference image used to identify the chroma block. Instead, the MV used for the luma decoder block can be scaled to determine the MV used for the chroma block, and the reference image can be the same. As another example, the intra-frame prediction process can be the same for both the luma and chroma decoders.

[0265] Video encoder 200 represents an example of a device configured to encode video data, the device including a memory configured to store the video data, and one or more processing units implemented in circuitry and configured to implement the techniques of this disclosure for NN-based in-loop filtering. For example, video encoder 200 may be configured to receive a picture of video data, reconstruct the picture of the video data, and perform an NN-based filtering process on one or more blocks of the reconstructed picture of the video data using an NN-based filter, wherein the NN-based filter includes a pair of backbone blocks, each of the backbone blocks including a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0266] Figure 25 This is a block diagram illustrating an example video decoder 300 that can perform the techniques of this disclosure. Figure 25 This disclosure is provided for illustrative purposes and not for limiting the techniques extensively illustrated and described herein. For illustrative purposes, the video decoder 300 is described in accordance with VVC and HEVC techniques. However, the techniques of this disclosure can be implemented by video decoding devices configured for other video decoding standards.

[0267] exist Figure 25 In the example, the video decoder 300 includes a decoded picture buffer (CPB) memory 320, an entropy decoding unit 302, a prediction processing unit 304, an inverse quantization unit 306, an inverse transform processing unit 308, a reconstruction unit 310, a filter unit 312, and a DPB 314. Any or all of the CPB memory 320, entropy decoding unit 302, prediction processing unit 304, inverse quantization unit 306, inverse transform processing unit 308, reconstruction unit 310, filter unit 312, and DPB 314 can be implemented in one or more processors or in processing circuitry. For example, the units of the video decoder 300 can be implemented as one or more circuit or logic elements as part of hardware circuitry, or as part of a processor, ASIC, or FPGA. Furthermore, the video decoder 300 may include additional or alternative processors or processing circuitry to perform these and other functions.

[0268] The prediction processing unit 304 includes a motion compensation unit 316 and an intra-prediction unit 318. The prediction processing unit 304 may include additional units that perform predictions based on other prediction modes. As an example, the prediction processing unit 304 may include a palette unit, an intra-block copying unit (which may form part of the motion compensation unit 316), an affine unit, a linear model (LM) unit, etc. In other examples, the video decoder 300 may include more, fewer, or different functional components.

[0269] When operating according to AV1, motion compensation unit 316 can be configured to decode video data blocks (e.g., both luma and chroma blocks) using translational motion compensation, affine motion compensation, OBMC, and / or composite inter-intra-frame prediction, as described above. Intra-frame prediction unit 318 can be configured to decode video data blocks (e.g., both luma and chroma blocks) using directional intra-frame prediction, non-directional intra-frame prediction, recursive filter intra-frame prediction, CFL, IBC, and / or palette mode, as described above.

[0270] CPB memory 320 is an example of a memory system capable of storing video data (such as encoded video bitstreams) to be decoded by components of video decoder 300. For example, it can be stored from computer-readable medium 110 ( Figure 1The video data stored in the CPB memory 320 is obtained. The CPB memory 320 may include a CPB that stores encoded video data (e.g., syntax elements) from the encoded video bitstream. Furthermore, the CPB memory 320 may store video data other than the syntax elements of the decoded picture, such as temporary data representing the output from various units of the video decoder 300. The DPB 314 is an example of a memory system that typically stores decoded pictures, which the video decoder 300 may output, and / or uses as reference video data when decoding subsequent data or pictures from the encoded video bitstream. The CPB memory 320 and the DPB 314 may each be formed from any of various memory devices or memory cells, such as DRAM (including SDRAM), MRAM, RRAM, or other types of memory devices. The CPB memory 320 and the DPB 314 may be provided by the same memory device or separate memory devices. In various examples, the CPB memory 320 may be on-chip with other components of the video decoder 300, or off-chip relative to those components.

[0271] Additionally or alternatively, in some examples, the video decoder 300 may be from the memory 120 ( Figure 1 The decoded video data can be retrieved from the memory. In other words, memory 120 can utilize CPB memory 320 to store data as discussed above. Similarly, when some or all of the functionality of video decoder 300 is implemented in software to be executed by the processing circuitry of video decoder 300, memory 120 can store instructions to be executed by video decoder 300.

[0272] Examples Figure 25 The various units shown help to understand the operations performed by the video decoder 300. The units can be implemented as fixed-function circuits, programmable circuits, or a combination thereof. Similar to... Figure 24 Fixed-function circuits are circuits that provide specific functionality and are pre-defined for the operations they can perform. Programmable circuits are circuits that can be programmed to perform various tasks and provide flexible functionality for the operations they can perform. For example, a programmable circuit can execute software or firmware that causes the programmable circuit to operate in a manner defined by the instructions of the software or firmware. Fixed-function circuits can execute software instructions (e.g., to receive or output parameters), but the type of operation performed by a fixed-function circuit is typically immutable. In some examples, one or more units in a cell may be different circuit blocks (fixed-function or programmable), and in some examples, one or more units in a cell may be integrated circuits.

[0273] The video decoder 300 may include an ALU, EFU, digital circuitry, analog circuitry, and / or a programmable core formed by programmable circuitry. In an example where the operation of the video decoder 300 is performed by software executed on programmable circuitry, on-chip or off-chip memory may store instructions (e.g., object code) of the software received and executed by the video decoder 300.

[0274] The entropy decoding unit 302 can receive encoded video data from the CPB and perform entropy decoding on the video data to reproduce the syntax elements. The prediction processing unit 304, the inverse quantization unit 306, the inverse transform processing unit 308, the reconstruction unit 310, and the filter unit 312 can generate decoded video data based on the syntax elements extracted from the bitstream.

[0275] Generally speaking, the video decoder 300 reconstructs the image block by block. The video decoder 300 can perform the reconstruction operation on each block individually (where the block currently being reconstructed (i.e., decoded) can be referred to as the "current block").

[0276] Entropy decoding unit 302 can entropy decode the syntax elements of the quantized transform coefficients defining the quantized transform coefficient block, as well as transform information (such as quantization parameters (QP) and / or transform mode indications). Inverse quantization unit 306 can use the QP associated with the quantized transform coefficient block to determine the degree of quantization, and similarly, determine the degree of inverse quantization to be applied by inverse quantization unit 306. Inverse quantization unit 306 can, for example, perform a bit-by-bit left shift operation to inverse quantize the quantized transform coefficients. Inverse quantization unit 306 can thereby form a transform coefficient block including the transform coefficients.

[0277] After the inverse quantization unit 306 forms the transform coefficient block, the inverse transform processing unit 308 may apply one or more inverse transforms to the transform coefficient block to generate a residual block associated with the current block. For example, the inverse transform processing unit 308 may apply an inverse DCT, an inverse integer transform, an inverse Karhunen-Loeve transform (KLT), an inverse rotation transform, an inverse direction transform, or another inverse transform to the transform coefficient block.

[0278] Furthermore, the prediction processing unit 304 generates a prediction block based on the prediction information syntax elements entropy decoded by the entropy decoding unit 302. For example, if the prediction information syntax elements indicate that the current block is an inter-frame prediction, the motion compensation unit 316 can generate the prediction block. In this case, the prediction information syntax elements may indicate the reference picture from which the reference block is to be retrieved in the DPB 314, and a motion vector identifying the position of the reference block in the reference picture relative to the position of the current block in the current picture. The motion compensation unit 316 may generally follow the same procedure as relative to the motion compensation unit 224 ( Figure 24 The method described is essentially the same as the method used to perform the inter-frame prediction process.

[0279] As another example, when the prediction information syntax element indicates that the current block is intra-predictive, intra-predictive unit 318 may generate a prediction block according to the intra-predictive mode indicated by the prediction information syntax element. Similarly, intra-predictive unit 318 may generally follow the same procedure as relative to intra-predictive unit 226 ( Figure 24 The intra-prediction process is performed in a manner substantially similar to that described above. The intra-prediction unit 318 can retrieve data of neighboring samples of the current block from the DPB 314.

[0280] Reconstruction unit 310 can use the prediction block and the residual block to reconstruct the current block. For example, reconstruction unit 310 can add samples from the residual block to the corresponding samples from the prediction block to reconstruct the current block.

[0281] Filter unit 312 can perform one or more filtering operations on the reconstructed block. For example, filter unit 312 can perform a deblocking operation to reduce block artifacts along the edges of the reconstructed block. The operation of filter unit 312 is not necessarily performed in all examples.

[0282] Filter unit 312 can also be configured to implement any combination of the NN-based filters described above. Specifically, filter unit 312 can be configured to perform an NN-based filtering process on one or more blocks of reconstructed images of video data using NN-based filters, wherein the NN-based filters include a pair of backbone blocks, each of which includes a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction. Filter unit 216 can be further configured to perform an NN-based filtering process using NN-based filters, which also include a fusion block with two separable convolutions, each of which applies spatial downsampling.

[0283] The video decoder 300 can store reconstructed blocks in the DPB 314. For example, in an example where the operation of the filter unit 312 is not performed, the reconstruction unit 310 can store the reconstructed blocks in the DPB 314. In an example where the operation of the filter unit 312 is performed, the filter unit 312 can store the filtered reconstructed blocks in the DPB 314. As discussed above, the DPB 314 can provide reference information (such as samples of the current image for intra-frame prediction and samples of previously decoded images for subsequent motion compensation) to the prediction processing unit 304. Furthermore, the video decoder 300 can output decoded images (e.g., decoded video) from the DPB 314 for use in applications such as... Figure 1 The subsequent presentation on display devices such as display device 118.

[0284] In this manner, video decoder 300 represents an example of a video decoding device, which includes a memory configured to store video data and one or more processing units implemented in circuitry and configured to implement the techniques of this disclosure for NN-based in-loop filtering. For example, video decoder 300 may be configured to receive a picture of video data, reconstruct the picture of the video data, and perform an NN-based filtering process on one or more blocks of the reconstructed picture of the video data using an NN-based filter, wherein the NN-based filter includes a pair of backbone blocks, each of the backbone blocks including a three-component 1D decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

[0285] Figure 26 This is a flowchart illustrating an example method for encoding the current block according to the technology of this disclosure. The current block may be or may include the current CU. Although regarding video encoder 200 ( Figure 1 and Figure 24 This description is provided, but it should be understood that other devices can be configured to perform similar actions. Figure 26 Similar to the method.

[0286] In this example, the video encoder 200 initially predicts the current block (400). For example, the video encoder 200 may form a prediction block for the current block. The video encoder 200 may then compute a residual block for the current block (402). To compute the residual block, the video encoder 200 may compute the difference between the unencoded original block for the current block and the prediction block. The video encoder 200 may then transform the residual block and quantize the transform coefficients of the residual block (404). Next, the video encoder 200 may scan the quantized transform coefficients of the residual block (406). During or after the scan, the video encoder 200 may entropy encode the transform coefficients (408). For example, the video encoder 200 may use CAVLC or CABAC to encode the transform coefficients. The video encoder 200 may then output the entropy-encoded data of the block (410).

[0287] Figure 27 This is a flowchart illustrating an example method for decoding a current block of video data according to the technology of this disclosure. The current block may be or may include the current CU. Although regarding the video decoder 300 ( Figure 1 and Figure 25 This description is provided, but it should be understood that other devices can be configured to perform similar actions. Figure 27 Similar to the method.

[0288] The video decoder 300 may receive entropy-coded data for the current block, such as entropy-coded prediction information and entropy-coded data for the transform coefficients of the residual block corresponding to the current block (500). The video decoder 300 may entropy decode the entropy-coded data to determine prediction information for the current block and reproduce the transform coefficients of the residual block (502). The video decoder 300 may predict the current block, for example, using an intra-frame prediction mode or inter-frame prediction mode indicated by the prediction information of the current block (504), to compute a prediction block for the current block. The video decoder 300 may then inverse scan the reproduced transform coefficients (506) to create a block of quantized transform coefficients. The video decoder 300 may then inverse quantize the transform coefficients and apply an inverse transform to the transform coefficients to produce a residual block (508). The video decoder 300 may finally decode the current block by combining the prediction block and the residual block (510).

[0289] Figure 28 This is a flowchart illustrating an example method for decoding the current block according to the technology of this disclosure. Figure 28 The technology can be performed by a video encoder 200 (e.g., including filter unit 216) and a video decoder 300 (e.g., including filter unit 312).

[0290] In one example of this disclosure, video encoder 200 and video decoder 300 may be configured to receive a picture (2800) of video data, a picture (2802) of reconstructed video data, and perform a NN-based filtering process on one or more blocks of the reconstructed picture of video data using a NN-based filter, wherein the NN-based filter includes a pair of backbone blocks, each of which includes a three-component 1D decomposition of multidimensional convolution, wherein the 1D decomposition includes at least one layer (2804) with feature channel reduction.

[0291] The input to a three-component 1D decomposition can be defined by its height (h), width (w), and number of channels (C). In one example, the three-component 1D decomposition of multidimensional convolution includes a 1×1 convolution with C output channels, a first separable convolution with C1 output channels in a first direction (where C1 is less than C), and a second separable convolution with C output channels in a second direction. The three-component 1D decomposition of multidimensional convolution may also include an activation function (such as PReLU).

[0292] In a further example of this disclosure, the backbone blocks include a first backbone block and a second backbone block, and the input to the first backbone block is defined by height (h), width (w), and the number of channels (C). In this example, the first backbone block includes a first 1×1 convolution with C output channels, a first separable convolution with C1 output channels in a first direction (where C1 is less than C), and a second separable convolution with C output channels in a second direction. The second backbone block includes a second 1×1 convolution with C output channels, a third separable convolution with C1 output channels in the second direction (where C1 is less than C), and a fourth separable convolution with C output channels in the first direction. In one example, the first direction is horizontal and the second direction is vertical. In another example, the first direction is vertical and the second direction is horizontal.

[0293] Various arrangements can be used for the backbone block. In one example, the first 1×1 convolution is positioned before the first separable convolution, and the second 1×1 convolution is positioned before the third separable convolution. In another example, the first 1×1 convolution is positioned after the second separable convolution, and the second 1×1 convolution is positioned before the fourth separable convolution.

[0294] In yet another example, the first backbone block includes a first PReLU following a first 1×1 convolution, and the second backbone includes a second PReLU following a second 1×1 convolution. In other examples, the first backbone block includes a first PReLU following a second separable convolution, and the second backbone includes a second PReLU following a fourth separable convolution.

[0295] In another example, the first backbone block includes a first PReLU following a first separable convolution and a second PReLU following a second separable convolution. The second backbone includes a third PReLU following a third separable convolution and a fourth PReLU following a fourth separable convolution.

[0296] In a further example of this disclosure, the NN-based filter further includes a fusion block having two separable convolutions, each of which applies spatial downsampling. The fusion block may include a 1×1 convolution, PReLU, a first separable convolution with a first spatial downsampling amount in a first direction, and a second separable convolution with a first spatial downsampling amount in a second direction. In one example, the first spatial downsampling amount is 2. In other examples, the first spatial downsampling amount is an integer not equal to 2 (e.g., not equal to 2). In one example, the first direction is horizontal and the second direction is vertical. In another example, the first direction is vertical and the second direction is horizontal.

[0297] Various arrangements can be used for the fusion block. In one example, a 1×1 convolution is positioned before the first separable convolution. In another example, a 1×1 convolution is positioned after the second separable convolution. In one example, PReLU is positioned after the 1×1 convolution. In another example, PReLU is positioned after the second separable convolution. In yet another example, PReLU is positioned after the first separable convolution. In a further example, PReLU is a first PReLU and the fusion block includes a first PReLU positioned after the first separable convolution and a second PReLU positioned after the second separable convolution.

[0298] In other examples of this disclosure, the NN-based filter further includes a first fusion block for luminance samples, the first fusion block having two separable convolutions that are first spatially downsampled; and a second fusion block for chrominance samples, the second fusion block having two separable convolutions, each of which is second spatially downsampled. In one example, the first and second spatial downsampling are applied with the same downsampling amount. In another example, the first and second spatial downsampling are applied with different downsampling amounts.

[0299] In other examples of this disclosure, the following parameters may be used. In one example, the NN-based filter includes 10 pairs of backbone blocks for luminance samples, 4 pairs of backbone blocks for chrominance samples, and 4 channels. In another example, the NN-based filter includes 4 pairs of backbone blocks for luminance samples, 1 pair of backbone blocks for chrominance samples, and 8 channels.

[0300] The following numbered clauses illustrate one or more aspects of the devices and technologies described in this disclosure.

[0301] Aspect 1A. A method for decoding video data, the method comprising: receiving a picture of the video data; reconstructing the picture of the video data; and performing a neural network (NN)-based filtering process on the reconstructed picture of the video data, wherein the NN-based filtering process includes using a backbone block having a three-component one-dimensional (1D) decomposition.

[0302] Aspect 2A. The method according to any one of Aspect 1A, wherein decoding includes decoding, and wherein reconstruction includes decoding.

[0303] Aspect 3A. The method according to any one of Aspect 1A, wherein decoding includes encoding.

[0304] Aspect 4A. An apparatus for decoding video data, the apparatus comprising one or more components for performing the method according to any one of Aspects 1A to 3A.

[0305] Aspect 5A. The device according to aspect 4A, wherein the one or more components include one or more processors implemented in a circuit.

[0306] Aspect 6A. The device according to any one of Aspects 4A and 5A, the device further comprising: a memory for storing the video data.

[0307] Aspect 7A. The device according to any one of aspects 4A to 6A, the device further comprising: a display configured to display decoded video data.

[0308] Aspect 8A. The device according to any one of Aspects 4A to 7A, wherein the device includes one or more of a camera, computer, mobile device, broadcast receiver device or set-top box.

[0309] Aspect 9A. The device according to any one of aspects 4A to 8A, wherein the device includes a video decoder.

[0310] Aspect 10A. The device according to any one of aspects 4A to 9A, wherein the device includes a video encoder.

[0311] Aspect 11A. A computer-readable storage medium having instructions stored thereon, which, when executed, cause one or more processors to perform the method according to any one of aspects 1A to 3A.

[0312] Aspect 1B. A method for decoding video data, the method comprising: receiving an image of the video data; reconstructing the image of the video data; and performing a neural network (NN)-based filtering process on one or more blocks of the reconstructed image of the video data using a neural network (NN)-based filter, wherein the NN-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0313] Aspect 2B. According to the method of aspect 1B, the input of the three-component 1D decomposition is defined by height (h), width (w) and number of channels (C), and the three-component 1D decomposition of the multidimensional convolution includes: a 1×1 convolution with C output channels, a first separable convolution in a first direction with C1 output channels, and a second separable convolution in a second direction with C output channels, wherein C1 is less than C.

[0314] Aspect 3B. According to the method of aspect 2B, the three-component 1D decomposition of the multidimensional convolution further includes an activation function.

[0315] Aspect 4B. The method according to aspect 3B, wherein the activation function is a parameter-corrected linear unit (PReLU).

[0316] Aspect 5B. The method according to any one of Aspects 1B to 4B, wherein the pair of backbone blocks comprises a first backbone block and a second backbone block, wherein the input of the first backbone block is defined by a height (h), a width (w), and a number of channels (C), and wherein the first backbone block comprises: a first 1×1 convolution having C output channels, a first separable convolution in a first direction having C1 output channels, and a second separable convolution in a second direction having C output channels, wherein C1 is less than C, and wherein the second backbone block comprises: a second 1×1 convolution having C output channels, a third separable convolution in the second direction having C1 output channels, and a fourth separable convolution in the second direction having C output channels, wherein C1 is less than C.

[0317] Aspect 6B. The method according to aspect 5B, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

[0318] Aspect 7B. The method according to aspect 5B, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

[0319] Aspect 8B. The method according to any one of Aspects 5B to 7B, wherein the first 1×1 convolution is positioned before the first separable convolution, and wherein the second 1×1 convolution is positioned before the third separable convolution.

[0320] Aspect 9B. The method according to any one of Aspects 5B to 7B, wherein the first 1×1 convolution is positioned after the second separable convolution, and wherein the second 1×1 convolution is positioned before the fourth separable convolution.

[0321] Aspect 10B. The method according to any one of Aspects 5B to 9B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first 1×1 convolution, and wherein the second backbone comprises a second PReLU following the second 1×1 convolution.

[0322] Aspect 11B. The method according to any one of Aspects 5B to 9B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the second separable convolution, and wherein the second backbone comprises a second PReLU following the fourth separable convolution.

[0323] Aspect 12B. The method according to any one of Aspects 5B to 9B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) after the first separable convolution and a second PReLU after the second separable convolution, and wherein the second backbone comprises a third PReLU after the third separable convolution and a fourth PReLU after the fourth separable convolution.

[0324] Aspect 13B. The method according to any one of Aspects 1B to 12B, wherein the NN-based filter further comprises a fusion block having two separable convolutions, each of which applies spatial downsampling.

[0325] Aspect 14B. The method according to aspect 13B, wherein the fusion block comprises: a 1×1 convolution, a parameter-corrected linear unit (PReLU), a first separable convolution having a first spatial downsampling amount in a first direction, and a second separable convolution having the first spatial downsampling amount in a second direction.

[0326] Aspect 15B. The method according to aspect 14B, wherein the first spatial downsampling amount is 2.

[0327] Aspect 16B. The method according to aspect 14B, wherein the first spatial downsampling amount is an integer not equal to 2.

[0328] Aspect 17B. The method according to any one of aspects 14B to 16B, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

[0329] Aspect 18B. The method according to any one of aspects 14B to 16B, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

[0330] Aspect 19B. The method according to any one of aspects 14B to 18B, wherein the 1×1 convolution is positioned before the first separable convolution.

[0331] Aspect 20B. The method according to any one of aspects 14B to 18B, wherein the 1×1 convolution is positioned after the second separable convolution.

[0332] Aspect 21B. The method according to any one of Aspects 14B to 20B, wherein the PReLU is positioned after the 1×1 convolution.

[0333] Aspect 22B. The method according to any one of aspects 14B to 20B, wherein the PReLU is positioned after the second separable convolution.

[0334] Aspect 23B. The method according to any one of aspects 14B to 20B, wherein the PReLU is positioned after the first separable convolution.

[0335] Aspect 24B. The method according to any one of aspects 14B to 20B, wherein the PReLU is a first PReLU, and wherein the fusion block comprises the first PReLU positioned after the first separable convolution and the second PReLU positioned after the second separable convolution.

[0336] Aspect 25B. The method according to any one of Aspects 1B to 24B, wherein the NN-based filter further comprises: a first fusion block for luminance samples, the first fusion block having two separable convolutions, each of the two separable convolutions being applied to a first spatial downsampling; and a second fusion block for chrominance samples, the second fusion block having two separable convolutions, each of the two separable convolutions being applied to a second spatial downsampling.

[0337] Aspect 26B. The method according to aspect 25B, wherein the first spatial downsampling and the second spatial downsampling apply the same downsampling amount.

[0338] Aspect 27B. The method according to aspect 25B, wherein the first spatial downsampling and the second spatial downsampling apply different downsampling amounts.

[0339] Aspect 28B. The method according to any one of Aspects 1B to 27B, wherein the NN-based filter comprises 10 pairs of backbone blocks for luminance samples, 4 pairs of backbone blocks for chrominance samples, and 4 channels.

[0340] Aspect 29B. The method according to any one of Aspects 1B to 27B, wherein the NN-based filter comprises 4 pairs of backbone blocks for luminance samples, 1 pair of backbone blocks for chrominance samples, and 8 channels.

[0341] Aspect 30B. The method according to any one of aspects 1B to 29B, wherein decoding includes decoding, and wherein the method further includes: displaying a decoded image comprising the one or more blocks processed by the NN-based filter.

[0342] Aspect 31B. The method according to any one of aspects 1B to 29B, wherein decoding includes encoding, and wherein the method further includes: using a camera to capture the image of video data.

[0343] Aspect 32B. An apparatus configured to encode video data, the apparatus comprising: a memory; and one or more processors in communication with the memory, the one or more processors being configured to: receive video data images; reconstruct the images of the video data; and perform a neural network (NN) based filtering process on one or more blocks of the reconstructed images of the video data using a neural network (NN) based filter, wherein the NN-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0344] Aspect 33B. The apparatus according to aspect 32B, wherein the input of the three-component 1D decomposition is defined by height (h), width (w), and number of channels (C), and wherein the three-component 1D decomposition of the multidimensional convolution comprises: a 1×1 convolution having C output channels, a first separable convolution in a first direction having C1 output channels, and a second separable convolution in a second direction having C output channels, wherein C1 is less than C.

[0345] Aspect 34B. The apparatus according to aspect 33B, wherein the three-component 1D decomposition of the multidimensional convolution further includes an activation function.

[0346] Aspect 35B. The apparatus according to aspect 34B, wherein the activation function is a parameter-corrected linear unit (PReLU).

[0347] Aspect 36B. The apparatus according to any one of aspects 32B to 35B, wherein the pair of backbone blocks comprises a first backbone block and a second backbone block, wherein the input of the first backbone block is defined by a height (h), a width (w), and a number of channels (C), and wherein the first backbone block comprises: a first 1×1 convolution having C output channels, a first separable convolution in a first direction having C1 output channels, and a second separable convolution in a second direction having C output channels, wherein C1 is less than C, and wherein the second backbone block comprises: a second 1×1 convolution having C output channels, a third separable convolution in the second direction having C1 output channels, and a fourth separable convolution in the second direction having C output channels, wherein C1 is less than C.

[0348] Aspect 37B. The apparatus according to aspect 36B, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

[0349] Aspect 38B. The apparatus according to aspect 36B, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

[0350] Aspect 39B. The apparatus according to any one of aspects 36B to 38B, wherein the first 1×1 convolution is positioned prior to the first separable convolution, and wherein the second 1×1 convolution is positioned prior to the third separable convolution.

[0351] Aspect 40B. The apparatus according to any one of aspects 36B to 38B, wherein the first 1×1 convolution is positioned after the second separable convolution, and wherein the second 1×1 convolution is positioned before the fourth separable convolution.

[0352] Aspect 41B. The apparatus according to any one of Aspects 36B to 40B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first 1×1 convolution, and wherein the second backbone comprises a second PReLU following the second 1×1 convolution.

[0353] Aspect 42B. The apparatus according to any one of Aspects 36B to 40B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the second separable convolution, and wherein the second backbone comprises a second PReLU following the fourth separable convolution.

[0354] Aspect 43B. The apparatus according to any one of Aspects 36B to 40B, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first separable convolution and a second PReLU following the second separable convolution, and wherein the second backbone comprises a third PReLU following the third separable convolution and a fourth PReLU following the fourth separable convolution.

[0355] Aspect 44B. The apparatus according to any one of aspects 32B to 43B, wherein the NN-based filter further comprises a fusion block having two separable convolutions, each of the two separable convolutions applying spatial downsampling.

[0356] Aspect 45. The apparatus according to aspect 44, wherein the fusion block comprises: A 1×1 convolution, a parameter-corrected linear unit (PReLU), a first separable convolution in the first direction having a first spatial downsampling amount, and a second separable convolution in the second direction having the first spatial downsampling amount.

[0357] Aspect 46B. The apparatus according to aspect 45B, wherein the first spatial downsampling amount is 2.

[0358] Aspect 47B. The apparatus according to aspect 45B, wherein the first spatial downsampling amount is an integer not equal to 2.

[0359] Aspect 48B. The apparatus according to any one of aspects 45B to 47B, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

[0360] Aspect 49B. The apparatus according to any one of aspects 45B to 47B, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

[0361] Aspect 50B. The apparatus according to any one of aspects 45B to 49B, wherein the 1×1 convolution is positioned prior to the first separable convolution.

[0362] Aspect 51B. The apparatus according to any one of aspects 45B to 49B, wherein the 1×1 convolution is positioned after the second separable convolution.

[0363] Aspect 52B. The apparatus according to any one of aspects 45B to 51B, wherein the PReLU is positioned after the 1×1 convolution.

[0364] Aspect 53B. The apparatus according to any one of aspects 45B to 51B, wherein the PReLU is positioned after the second separable convolution.

[0365] Aspect 54B. The apparatus according to any one of aspects 45B to 51B, wherein the PReLU is positioned after the first separable convolution.

[0366] Aspect 55B. The apparatus according to any one of aspects 45B to 51B, wherein the PReLU is a first PReLU, and wherein the fusion block includes the first PReLU positioned after the first separable convolution and the second PReLU positioned after the second separable convolution.

[0367] Aspect 56B. The apparatus according to any one of Aspects 32B to 55B, wherein the NN-based filter further comprises: a first fusion block for luminance samples, the first fusion block having two separable convolutions, each of the two separable convolutions applying a first spatial downsampling; and a second fusion block for chrominance samples, the second fusion block having two separable convolutions, each of the two separable convolutions applying a second spatial downsampling.

[0368] Aspect 57B. The apparatus according to aspect 56B, wherein the first spatial downsampling and the second spatial downsampling apply the same downsampling amount.

[0369] Aspect 58B. The apparatus according to aspect 56B, wherein the first spatial downsampling and the second spatial downsampling apply different downsampling amounts.

[0370] Aspect 59B. The apparatus according to any one of Aspects 32B to 58B, wherein the NN-based filter comprises 10 pairs of backbone blocks for luminance samples, 4 pairs of backbone blocks for chrominance samples, and 4 channels.

[0371] Aspect 60B. The apparatus according to any one of Aspects 32B to 58B, wherein the NN-based filter comprises four pairs of backbone blocks for luminance samples, one pair of backbone blocks for chrominance samples, and eight channels. Aspect 61B. The apparatus according to any one of aspects 32B to 60B, wherein the apparatus is a video decoder, and wherein the apparatus further comprises: a display configured to display a decoded image comprising the one or more blocks processed by the NN-based filter.

[0372] Aspect 62B. The apparatus according to any one of aspects 32B to 60B, wherein the apparatus is a video decoder, and wherein the apparatus further comprises: a camera configured to capture the image of video data using the camera.

[0373] Aspect 63B. An apparatus configured to decode video data, the apparatus comprising: a component for receiving an image of the video data; a component for reconstructing the image of the video data; and a component for performing a neural network (NN)-based filtering process on one or more blocks of the reconstructed image of the video data using a neural network (NN)-based filter, wherein the NN-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer having feature channel reduction.

[0374] Aspect 64B. A non-transitory computer-readable storage medium storing instructions, which, when executed, cause one or more processors of a device configured to decode video data to: receive a picture of the video data; reconstruct the picture of the video data; and perform a neural network (NN)-based filtering process on one or more blocks of the reconstructed picture of the video data, wherein the NN-based filter comprises a pair of backbone blocks, each of the pair of backbone blocks comprising a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition comprises at least one layer with feature channel reduction.

[0375] It should be recognized that, based on the examples, certain actions or events of any technique described herein may be performed in a different sequence, and may be added, combined, or omitted entirely (e.g., not all actions or events described are necessary for implementing the technique). Furthermore, in some examples, actions or events may be performed concurrently (e.g., through multithreading, interrupt handling, or multiple processors) rather than sequentially.

[0376] In one or more examples, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functionality may be stored as one or more instructions or code on a computer-readable medium or transmitted via a computer-readable medium and executed by a hardware-based processing unit. A computer-readable medium may include a computer-readable storage medium (which corresponds to a tangible medium such as a data storage medium) or a communication medium, including, for example, any medium that facilitates the transfer of a computer program from one place to another according to a communication protocol. Thus, a computer-readable medium may generally correspond to (1) a non-transitory tangible computer-readable storage medium, or (2) a communication medium such as a signal or carrier wave. A data storage medium may be any available medium that can be accessed by one or more computers or one or more processors to retrieve instructions, code, and / or data structures for implementing the techniques described in this disclosure. Computer program products may include computer-readable media.

[0377] By way of example, and not limitation, such computer-readable storage media may include one or more of RAM, ROM, EEPROM, CD-ROM or other optical disc storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer. Furthermore, any connection is appropriately referred to as a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technology (such as infrared, radio, and microwave), then coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technology (such as infrared, radio, and microwave) is included in the definition of medium. However, it should be understood that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but instead refer to non-transient tangible storage media. As used herein, disks and optical discs include compact optical discs (CDs), laser optical discs, optical discs, digital versatile optical discs (DVDs), floppy disks, and Blu-ray discs, where disks typically reproduce data magnetically, while optical discs use lasers to reproduce data optically. Combinations of these should also be included within the scope of computer-readable media.

[0378] Instructions can be executed by one or more processors, such as one or more DSPs, general-purpose microprocessors, ASICs, FPGAs, or other equivalent integrated or discrete logic circuits. Therefore, the terms "processor" and "processing circuit" as used herein can refer to any of the foregoing structures or any other structure suitable for implementing the techniques described herein. Additionally, in some aspects, the functionality described herein can be provided within dedicated hardware and / or software modules configured for encoding and decoding, or incorporated into combined codecs. Furthermore, these techniques can be fully implemented in one or more circuit or logic elements.

[0379] The techniques disclosed herein can be implemented in a wide variety of devices or apparatuses, including wireless mobile phones, integrated circuits (ICs), or IC sets (e.g., chip sets). Various components, modules, or units are described in this disclosure to emphasize functional aspects of a device configured to perform the disclosed techniques, but implementation by different hardware units is not necessarily required. Specifically, as described above, various units may be combined in a codec hardware unit, or various units may be provided by a collection of interoperable hardware units (including one or more processors as described above) combined with appropriate software and / or firmware.

[0380] Various examples have been described. These and other examples are within the scope of the following claims.

Claims

1. A method for decoding video data, the method comprising: Images that receive video data; The image described in the reconstructed video data; as well as A neural network (NN)-based filter is used to perform a NN-based filtering process on one or more blocks of the reconstructed image of the video data, wherein the NN-based filter includes a pair of backbone blocks, each of the pair of backbone blocks including a three-component one-dimensional (1D) decomposition of multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

2. The method according to claim 1, wherein the input of the three-component 1D decomposition is defined by height (h), width (w), and number of channels (C), and The three-component 1D decomposition of the multidimensional convolution includes: A 1×1 convolution with C output channels, A first separable convolution in a first direction having C1 output channels, where C1 is less than C, and A second separable convolution with C output channels in the second direction.

3. The method according to claim 2, wherein the three-component 1D decomposition of the multidimensional convolution further includes an activation function.

4. The method of claim 3, wherein the activation function is a parameter-corrected linear unit (PReLU).

5. The method of claim 1, wherein the pair of backbone blocks comprises a first backbone block and a second backbone block, wherein the input of the first backbone block is defined by height (h), width (w), and number of channels (C), and The first backbone block includes: The first 1×1 convolution with C output channels, A first separable convolution in a first direction having C1 output channels, where C1 is less than C, and A second separable convolution with C output channels in the second direction, and The second backbone block includes: The second 1×1 convolution with C output channels, A third separable convolution in the second direction with C1 output channels, where C1 is less than C, and A fourth separable convolution in the first direction with C output channels.

6. The method of claim 5, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

7. The method of claim 5, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

8. The method of claim 5, wherein the first 1×1 convolution is positioned before the first separable convolution, and wherein the second 1×1 convolution is positioned before the third separable convolution.

9. The method of claim 5, wherein the first 1×1 convolution is positioned after the second separable convolution, and wherein the second 1×1 convolution is positioned before the fourth separable convolution.

10. The method of claim 5, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first 1×1 convolution, and The second backbone includes the second PReLU following the second 1×1 convolution.

11. The method of claim 5, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the second separable convolution, and The second backbone includes the second PReLU following the fourth separable convolution.

12. The method of claim 5, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first separable convolution and a second PReLU following the second separable convolution, and The second backbone includes the third PReLU after the third separable convolution and the fourth PReLU after the fourth separable convolution.

13. The method of claim 1, wherein the NN-based filter further comprises a fusion block having two separable convolutions, each of the two separable convolutions applying spatial downsampling.

14. The method of claim 13, wherein the fusion block comprises: 1×1 convolution, Parameter Corrected Linear Unit (PReLU) A first separable convolution in the first direction having a first spatial downsampling amount, and A second separable convolution in the second direction having the first spatial downsampling amount.

15. The method of claim 14, wherein the first spatial downsampling amount is 2.

16. The method of claim 14, wherein the first spatial downsampling amount is an integer not equal to 2.

17. The method of claim 14, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

18. The method of claim 14, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

19. The method of claim 14, wherein the 1×1 convolution is positioned prior to the first separable convolution.

20. The method of claim 14, wherein the 1×1 convolution is positioned after the second separable convolution.

21. The method of claim 14, wherein the PReLU is positioned after the 1×1 convolution.

22. The method of claim 14, wherein the PReLU is positioned after the second separable convolution.

23. The method of claim 14, wherein the PReLU is positioned after the first separable convolution.

24. The method of claim 14, wherein the PReLU is a first PReLU, and wherein the fusion block comprises the first PReLU positioned after the first separable convolution and the second PReLU positioned after the second separable convolution.

25. The method of claim 1, wherein the NN-based filter further comprises: A first fusion block for luminance samples, the first fusion block having two separable convolutions, each of which applies a first spatial downsampling; And a second fusion block for chroma samples, the second fusion block having two separable convolutions, each of which applies a second spatial downsampling.

26. The method of claim 25, wherein the first spatial downsampling and the second spatial downsampling apply the same downsampling amount.

27. The method of claim 25, wherein the first spatial downsampling and the second spatial downsampling apply different downsampling amounts.

28. The method of claim 1, wherein the NN-based filter comprises 10 pairs of backbone blocks for luminance samples, 4 pairs of backbone blocks for chrominance samples, and 4 channels.

29. The method of claim 1, wherein the NN-based filter comprises 4 pairs of backbone blocks for luminance samples, 1 pair of backbone blocks for chrominance samples, and 8 channels.

30. The method of claim 1, wherein decoding includes decoding, and wherein the method further comprises: Displaying a decoded image, the decoded image comprising the one or more blocks processed by the NN-based filter.

31. The method of claim 1, wherein decoding includes encoding, and wherein the method further comprises: The image is captured using a camera to capture video data.

32. An apparatus configured to decode video data, the apparatus comprising: Memory; and One or more processors communicating with the memory, the one or more processors being configured to: Images that receive video data; The image described in the reconstructed video data; as well as A neural network (NN)-based filter is used to perform a NN-based filtering process on one or more blocks of the reconstructed image of the video data, wherein the NN-based filter includes a pair of backbone blocks, each of the pair of backbone blocks including a three-component one-dimensional (1D) decomposition of multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

33. The apparatus of claim 32, wherein the input to the three-component 1D decomposition is defined by height (h), width (w), and the number of channels (C), and The three-component 1D decomposition of the multidimensional convolution includes: A 1×1 convolution with C output channels, A first separable convolution in a first direction having C1 output channels, where C1 is less than C, and A second separable convolution with C output channels in the second direction.

34. The apparatus of claim 33, wherein the three-component 1D decomposition of the multidimensional convolution further comprises an activation function.

35. The apparatus of claim 34, wherein the activation function is a parameter-corrected linear unit (PReLU).

36. The apparatus of claim 32, wherein the pair of backbone blocks comprises a first backbone block and a second backbone block, wherein the input of the first backbone block is defined by height (h), width (w), and number of channels (C), and The first backbone block includes: The first 1×1 convolution with C output channels, A first separable convolution in a first direction having C1 output channels, where C1 is less than C, and A second separable convolution with C output channels in the second direction, and The second backbone block includes: The second 1×1 convolution with C output channels, A third separable convolution in the second direction with C1 output channels, where C1 is less than C, and A fourth separable convolution in the first direction with C output channels.

37. The apparatus of claim 36, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

38. The apparatus of claim 36, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

39. The apparatus of claim 36, wherein the first 1×1 convolution is positioned before the first separable convolution, and wherein the second 1×1 convolution is positioned before the third separable convolution.

40. The apparatus of claim 36, wherein the first 1×1 convolution is positioned after the second separable convolution, and wherein the second 1×1 convolution is positioned before the fourth separable convolution.

41. The apparatus of claim 36, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first 1×1 convolution, and The second backbone includes the second PReLU following the second 1×1 convolution.

42. The apparatus of claim 36, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the second separable convolution, and The second backbone includes the second PReLU following the fourth separable convolution.

43. The apparatus of claim 36, wherein the first backbone block comprises a first parameter-corrected linear unit (PReLU) following the first separable convolution and a second PReLU following the second separable convolution, and The second backbone includes the third PReLU after the third separable convolution and the fourth PReLU after the fourth separable convolution.

44. The apparatus of claim 32, wherein the NN-based filter further comprises a fusion block having two separable convolutions, each of the two separable convolutions applying spatial downsampling.

45. The apparatus of claim 44, wherein the fusion block comprises: 1×1 convolution, Parameter Corrected Linear Unit (PReLU) A first separable convolution in the first direction having a first spatial downsampling amount, and A second separable convolution in the second direction having the first spatial downsampling amount.

46. ​​The apparatus of claim 45, wherein the first spatial downsampling amount is 2.

47. The apparatus of claim 45, wherein the first spatial downsampling amount is an integer not equal to 2.

48. The apparatus of claim 45, wherein the first direction is a horizontal direction and the second direction is a vertical direction.

49. The apparatus of claim 45, wherein the first direction is a vertical direction and the second direction is a horizontal direction.

50. The apparatus of claim 45, wherein the 1×1 convolution is positioned prior to the first separable convolution.

51. The apparatus of claim 45, wherein the 1×1 convolution is positioned after the second separable convolution.

52. The apparatus of claim 45, wherein the PReLU is positioned after the 1×1 convolution.

53. The apparatus of claim 45, wherein the PReLU is positioned after the second separable convolution.

54. The apparatus of claim 45, wherein the PReLU is positioned after the first separable convolution.

55. The apparatus of claim 45, wherein the PReLU is a first PReLU, and wherein the fusion block comprises the first PReLU positioned after the first separable convolution and the second PReLU positioned after the second separable convolution.

56. The apparatus of claim 32, wherein the NN-based filter further comprises: A first fusion block for luminance samples, the first fusion block having two separable convolutions, each of which applies a first spatial downsampling; And a second fusion block for chroma samples, the second fusion block having two separable convolutions, each of which applies a second spatial downsampling.

57. The apparatus of claim 56, wherein the first spatial downsampling and the second spatial downsampling apply the same downsampling amount.

58. The apparatus of claim 56, wherein the first spatial downsampling and the second spatial downsampling apply different downsampling amounts.

59. The apparatus of claim 32, wherein the NN-based filter comprises 10 pairs of backbone blocks for luminance samples, 4 pairs of backbone blocks for chrominance samples, and 4 channels.

60. The apparatus of claim 32, wherein the NN-based filter comprises four pairs of backbone blocks for luminance samples, one pair of backbone blocks for chrominance samples, and eight channels.

61. The apparatus of claim 32, wherein the apparatus is a video decoder, and wherein the apparatus further comprises: A display configured to display a decoded image, the decoded image comprising the one or more blocks processed by the NN-based filter.

62. The apparatus of claim 32, wherein the apparatus is a video decoder, and wherein the apparatus further comprises: A camera configured to capture the image using video data.

63. An apparatus configured to decode video data, the apparatus comprising: A component used to receive video data from images; Components used to reconstruct the image from the video data; and A component for performing a neural network (NN) based filtering process on one or more blocks of a reconstructed image of video data using a neural network (NN) based filter, wherein the NN-based filter includes a pair of backbone blocks, each of the pair of backbone blocks including a three-component one-dimensional (1D) decomposition of a multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

64. A non-transitory computer-readable storage medium storing instructions, which, when executed, cause a device to be configured as one or more processors to decode video data: Images that receive video data; The image described in the reconstructed video data; as well as A neural network (NN)-based filter is used to perform a NN-based filtering process on one or more blocks of the reconstructed image of the video data, wherein the NN-based filter includes a pair of backbone blocks, each of the pair of backbone blocks including a three-component one-dimensional (1D) decomposition of multidimensional convolution, wherein the 1D decomposition includes at least one layer with feature channel reduction.

Images