Method and apparatus for video coding

Abstract

A method of coding video data, the method comprising: reconstructing a block of video data; and applying a convolutional neural network (CNN) based filter to the reconstructed block, wherein the CNN based filter uses a LeakyReLU activation function.

Description

[0001] This application claims priority to U.S. Patent Application No. 17 / 489,459, filed September 29, 2021, and U.S. Provisional Patent Application No. 63 / 085,936, filed September 30, 2020, the entire content of each of which is incorporated herein by reference. U.S. Patent Application No. 17 / 489,459, filed September 29, 2021, claims the benefit of U.S. Provisional Patent Application No. 63 / 085,936, filed September 30, 2020. TECHNICAL FIELD

[0002] The present disclosure relates to video encoding and video decoding. BACKGROUND

[0003] Digital video capabilities can be incorporated into a wide range of devices, including digital televisions, digital direct broadcast systems, wireless broadcast systems, personal digital assistants (PDAs), laptop or desktop computers, tablet computers, e-book readers, digital cameras, digital recording devices, digital media players, video gaming devices, video game consoles, cellular or satellite radio telephones, so-called “smart phones,” video teleconferencing devices, video streaming devices, and the like. Digital video devices implement video coding techniques, such as those described in the standards defined by MPEG-2, MPEG-4, ITU-T H.263, ITU-T H.264 / MPEG-4, Part 10, Advanced Video Coding (AVC), ITU-T H.265 / High Efficiency Video Coding (HEVC), and extensions of such standards. The video devices can more efficiently send, receive, encode, decode, and / or store digital video information by implementing such video coding techniques.

[0004] Video coding 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 coding, a video slice (e.g., a video picture or a portion of a video picture) can be partitioned into video blocks, which can also be referred to as coding tree units (CTUs), coding units (CUs), and / or coding nodes. Video blocks in an intra-coded (I) slice of a picture are encoded using spatial prediction with respect to reference samples in neighboring blocks within the same picture. Video blocks in an inter-coded (P or B) slice of a picture can use spatial prediction with respect to reference samples in neighboring blocks in the same picture or temporal prediction with respect to reference samples in other reference pictures. Pictures can be referred to as frames, and reference pictures can be referred to as reference frames. SUMMARY

[0005] In general, this disclosure describes techniques related to a filtering process for distorted pictures. The filtering process can be based on neural network techniques. For example, the filtering process can apply a convolutional neural network (CNN) based filter to a reconstructed block of video data. The CNN based filter can use a Leaky Rectified Linear Unit (ReLU) activation function. Using a LeakyReLU activation function can result in better performance than a Rectified Linear Unit (ReLU) activation function and higher stability during training than a Parametric Rectified Linear Unit (PReLU) activation function. The filtering process can be used in the context of advanced video codecs, such as extensions of Versatile Video Coding (VVC) or next generation video coding standards, as well as any other video codecs.

[0006] In one example, this disclosure describes a method of coding video data, the method comprising: reconstructing a block of the video data; and applying a convolutional neural network (CNN) based filter to the reconstructed block, wherein the CNN based filter uses a LeakyReLU activation function.

[0007] In another example, a device for coding video data comprises: a memory to store the video data; and one or more processors implemented in circuitry, the one or more processors configured to: reconstruct a block of the video data; and apply a convolutional neural network (CNN) based filter to the reconstructed block, wherein the CNN based filter uses a LeakyReLU activation function.

[0008] In another example, this disclosure describes a device for coding video data, the device comprising: means for reconstructing a block of the video data; and means for applying a convolutional neural network (CNN) based filter to the reconstructed block, wherein the CNN based filter uses a LeakyReLU activation function.

[0009] In another example, this disclosure describes a computer-readable storage medium having stored thereon instructions that, when executed, cause one or more processors to: reconstruct a block of the video data; and apply a convolutional neural network (CNN) based filter to the reconstructed block, wherein the CNN based filter uses a LeakyReLU activation function.

[0010] The details of one or more examples are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will be apparent from the description, drawings, and claims. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a block diagram illustrating an example video encoding and decoding system that can perform the techniques of this disclosure.

[0012] Figure 2A and Figure 2B is a conceptual diagram illustrating an example quadtree binary tree (QTBT) structure and corresponding coding tree unit (CTU).

[0013] Figure 3 is a block diagram illustrating an example hybrid video coding framework.

[0014] Figure 4 is a conceptual diagram illustrating an example hierarchical prediction structure with a group of pictures (GOP) size equal to 16.

[0015] Figure 5 is a conceptual diagram illustrating a convolutional neural network (CNN) based filter with four layers.

[0016] Figure 6A is a conceptual diagram illustrating a ReLU activation function.

[0017] Figure 6B is a conceptual diagram illustrating a PReLU activation function.

[0018] Figure 7 is a conceptual diagram illustrating an example CNN based filter with 14 layers and LeakyReLU as the activation function according to one or more techniques of this disclosure.

[0019] Figure 8 is a block diagram illustrating an example video encoder that can perform the techniques of this disclosure.

[0020] Figure 9 is a block diagram illustrating an example video decoder that can perform the techniques of this disclosure.

[0021] Figure 10 is a flowchart illustrating an example method for encoding a current block according to the techniques of this disclosure.

[0022] Figure 11 is a flowchart illustrating an example method for decoding a current block according to the techniques of this disclosure.

[0023] Figure 12 is a flowchart illustrating example operations of a video coder according to one or more techniques of this disclosure. DETAILED DESCRIPTION

[0024] Video encoders and video decoders can implement in-loop filters that can improve reconstructed picture quality. Common types of in-loop filters include a deblocking filter and an adaptive loop filter (ALF). Recently, neural network (NN) based filters have been proposed. A NN based filter includes a neural network that takes a block of a reconstructed picture as input. The neural network outputs a filtered block.

[0025] A neural network-based filter can include an input layer, one or more hidden layers, and an output layer. These layers can include a set of neurons that receive inputs and generate outputs. The input layer and each hidden layer are associated with an activation function that is applied to the outputs of the neurons of that layer. Rectified linear unit (ReLU) and parametric ReLU (PReLU) are examples of activation functions. The activation function used in a neural network can have a significant impact on the performance and trainability of the neural network. For example, PReLU can have better performance than ReLU. However, PReLU can make the training process of the neural network unstable because PReLU itself contains a trainable variable that can become extreme.

[0026] This disclosure describes techniques that can address these technical problems of previously proposed NN-based filters in video coding. For example, a video coder (e.g., a video encoder or a video decoder) can reconstruct a block of video data. In addition, the video coder can apply a convolutional neural network (CNN)-based filter to the reconstructed block. The CNN-based filter uses a LeakyReLU activation function. Using a LeakyReLU activation function can enable the CNN-based filter to have better performance than PReLU while avoiding instability during the training process. In contrast to other types of activation functions, the use of the LeakyReLU activation function is based on specific characteristics exhibited during the filtering of reconstructed pictures, and can bring unexpected significant improvements for video coding in both performance and stability.

[0027] Figure 1 FIG. 1 is a block diagram illustrating an example video encoding and decoding system 100 that can perform the techniques of this disclosure. The techniques of this disclosure generally relate to coding (encoding and / or decoding) video data. In general, video data includes any data for processing video. Thus, video data can include raw, uncoded video, coded video, decoded (e.g., reconstructed) video, and video metadata, such as signaling data.

[0028] As Figure 1As shown, in this example, system 100 includes source device 102, which provides encoded video data to be decoded and displayed by 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 include any of a wide range of devices, including desktop computers, laptop computers, mobile devices, tablet computers, set-top boxes, handsets such as smartphones, televisions, cameras, display devices, digital media players, video game terminals, video streaming devices, broadcast receiver devices, etc. In some cases, source device 102 and destination device 116 may be configured for wireless communication and therefore may be referred to as wireless communication devices.

[0029] exist Figure 1 In one example, source device 102 includes a video source 104, memory 106, a video encoder 200, and an output interface 108. Destination device 116 includes an input interface 122, a video decoder 300, memory 120, and a 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 using neural network-based filters. 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 and destination devices 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 be coupled to an external display device, rather than including an integrated display device.

[0030] like Figure 1 The system 100 shown is merely an example. Typically, any digital video encoding and / or decoding device can perform techniques using neural network-based filters. The source device 102 and destination device 116 are simply examples of such decoding devices, where the source device 102 generates decoded video data for transmission to the destination device 116. This disclosure refers to a “decoding” device as a device that performs the decoding (encoding and / or decoding) of data. Thus, the video encoder 200 and video decoder 300 represent examples of decoding devices, specifically, a video encoder and a video decoder, respectively. In some examples, the source device 102 and destination device 116 may operate in a substantially symmetrical manner, such that each of the source device 102 and destination device 116 includes video encoding and decoding components. Therefore, system 100 can support one-way or two-way video transmission between the source device 102 and destination device 116, for example, for video streaming, video playback, video broadcasting, or video telephony.

[0031] In general, video source 104 represents a source of video data (i.e., raw, unencoded video) and provides a continuous stream of pictures (also referred to as “frames”) of the video data to video encoder 200, which encodes data for the pictures. Video source 104 of source device 102 can include a video capture device, such as a video camera, a video archive containing previously captured raw video, and / or a video feed interface to receive video from a video content provider. As a further alternative, video source 104 can generate computer graphics-based data as the 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 can rearrange the pictures from the received order (sometimes referred to as “display order”) into the coding order for coding. Video encoder 200 can generate a bitstream including encoded video data. Source device 102 can then output the encoded video data via output interface 108 onto computer-readable medium 110 for reception and / or retrieval by an input interface 122 of destination device 116, for example.

[0032] Memory 106 of source device 102 and memory 120 of destination device 116 represent general purpose memories. In some examples, memories 106, 120 can 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, 120 can store software instructions executable by video encoders 200 and video decoders 300, respectively. Although memory 106 and memory 120 are shown separately from video encoders 200 and video decoders 300 in this example, it should be understood that video encoders 200 and video decoders 300 can also include internal memories for similar or equivalent purposes. Furthermore, memories 106, 120 can store encoded video data, e.g., output from video encoder 200 and input to video decoder 300. In some examples, portions of memories 106, 120 can be allocated as one or more video buffers, e.g., to store raw, decoded, and / or encoded video data.

[0033] Computer-readable medium 110 can represent any type of medium or device capable of transporting encoded video data from source device 102 to destination device 116. In one example, computer-readable medium 110 represents a communication medium to enable source device 102 to transmit encoded video data directly to destination device 116 in real-time, e.g., via a radio frequency network or computer-based network. Output interface 108 can modulate transmission signals including the encoded video data, and input interface 122 can demodulate received transmission signals according to communication standards, such as wireless communication protocols. The communication medium can comprise any wireless or wired communication medium, such as a radio frequency (RF) spectrum or one or more physical transmission lines. The communication medium can 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 can include routers, switches, base stations, or any other equipment that can be useful to facilitate communication from source device 102 to destination device 116.

[0034] 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 can include any of a variety of distributed or locally accessed data storage media such as a hard drive, Blu-ray discs, DVDs, CD-ROMs, flash memory, volatile or non-volatile memory, or any other suitable digital storage media for storing encoded video data.

[0035] In some examples, source device 102 can output encoded video data to file server 114, which can store the encoded video data for later access by destination device 116. Destination device 116 can access stored video data from file server 114 via streaming or download.

[0036] The file server 114 can be any type of server device capable of storing encoded video data and transmitting that encoded video data to the destination device 116. The file server 114 can represent a web server (e.g., for a website), a server configured to provide file transfer protocol services (e.g., File Transfer Protocol (FTP) or File Delivery over Unidirectional Transport (FLUTE) 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. The file server 114 can additionally or alternatively implement one or more HTTP streaming protocols, such as Dynamic Adaptive Streaming over HTTP (DASH), HTTP Live Streaming (HLS), Real Time Streaming Protocol (RTSP), HTTP Dynamic Streaming, and the like.

[0037] The destination device 116 can access the encoded video data from the file server 114 through any standard data connection, including an Internet connection. This can include a wireless channel (e.g., a Wi-Fi connection), a wired connection (e.g., a digital subscriber line (DSL), a cable modem, or the like), or a combination of both that is suitable for accessing the encoded video data stored on the file server 114. The input interface 122 can be configured to operate according to any one or more of various protocols

[0038] The output interface 108 and the input interface 122 can represent wireless transmitters / receivers, modems, wired networking components (e.g., Ethernet cards), wireless communication components operating according to any of a variety of IEEE 802.11 standards, or other physical components. In examples where the output interface 108 and the input interface 122 include wireless components, the output interface 108 and the input interface 122 can be configured to transmit data, e.g., encoded video data, according to a cellular communication standard, such as 4G, 4G-LTE (Long-Term Evolution), LTE Advanced, 5G, or the like. In some examples where the output interface 108 includes a wireless transmitter, the output interface 108 and the input interface 122 can be configured to transmit data, e.g., encoded video data, according to other wireless standards (e.g., IEEE 802.11 specifications, IEEE 802.15 specifications (e.g., ZigBee TM TM ​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 to perform functions belonging to video encoder 200 and / or output interface 108, and destination device 116 may include an SoC device to perform functions belonging to video decoder 300 and / or input interface 122.

[0039] The technology disclosed herein can be applied to video decoding that supports any of a variety of multimedia applications, such as over-the-air television broadcasting, cable television transmission, satellite television transmission, internet streaming video transmission, such as Dynamic Adaptive Streaming (DASH) via HTTP, digital video encoded onto a data storage medium, decoding digital video stored on a data storage medium, or other applications.

[0040] 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, which is 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 decoded images 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 other types of display devices.

[0041] Although Figure 1 Not shown, but in some examples, the video encoder 200 and video decoder 300 may each be integrated with the audio encoder and / or audio decoder, and may include appropriate MUX-DEMUX units, or other hardware and / or software, to handle the multiplexed streams of both audio and video included in a common data stream. Where applicable, the MUX-DEMUX unit may conform to the ITU H.223 multiplexer protocol or other protocols, such as User Datagram Protocol (UDP).

[0042] Video encoder 200 and video decoder 300 each can be implemented as any of a variety of suitable encoder and / or decoder circuitry, such as one or more microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), discrete logic, software, hardware, firmware or any combinations thereof. When the techniques are implemented partially in software, a device can 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 techniques of this disclosure. Each of video encoder 200 and video decoder 300 can be included in one or more encoders or decoders, any of which alone can be a part of a

[0043] Video encoder 200 and video decoder 300 can operate according to a video coding standard, such as ITU-T H.265, also referred to as High Efficiency Video Coding (HEVC), or extensions thereof, such as multi-view and / or scalable video coding extensions. Alternatively, video encoder 200 and video decoder 300 can operate according to other proprietary or industry standards, such as ITU-T H.266, also referred to as Versatile Video Coding (VVC). Version 1 of the VVC specification has been finalized (hereinafter referred to as VVC FDIS), and is available from http: / / phenix.int- evry.fr / jvet / doc_end_user / documents / 19_Teleconference / wg11 / JVET-S2001-v17.zip. The techniques of this disclosure, however, are not limited to any particular coding standard.

[0044] In general, video encoder 200 and video decoder 300 can perform block-based coding of pictures. The term “block” generally refers to a structure containing data to be processed (e.g., encoded, decoded, or otherwise used in encoding and / or decoding processes). For example, a block can include a two-dimensional matrix of luma and / or chroma samples. In general, video encoder 200 and video decoder 300 can code video data represented in a YUV (e.g., Y, Cb, Cr) format. That is, rather than coding picture samples of red, green, and blue (RGB) data, video encoder 200 and video decoder 300 can code luma and chroma components, where the chroma components can include both red hue and blue hue chroma components. In some examples, video encoder 200 converts received RGB formatted data to a YUV representation prior to encoding, and video decoder 300 converts the YUV representation to the RGB format. Alternatively, pre- and post-processing units (not shown) can perform these conversions.

[0045] This disclosure can generally refer to coding (e.g., encoding and decoding) of pictures to include the process of encoding or decoding data of the pictures. Similarly, this disclosure can refer to coding of blocks of pictures to include the process of encoding or decoding data of the blocks, e.g., prediction and / or residual coding. An encoded video bitstream generally includes a series of values representing coding decisions (e.g., coding modes) and syntax elements that partition a picture into blocks. Thus, a reference to coding of a picture or block should generally be understood to refer to coding values of syntax elements that form the picture or block.

[0046] HEVC defines various blocks, including coding units (CUs), prediction units (PUs), and transform units (TUs). According to HEVC, a video coder (such as video encoder 200) partitions a coding tree unit (CTU) into CUs according to a quadtree structure. That is, the video coder divides a CTU and CUs into four equal, non overlapping squares, and each node of the quadtree has either zero or four child nodes. Nodes with zero child nodes can be referred to as “leaf nodes,” and CUs of such leaf nodes can include one or more PUs and / or one or more TUs. The video coder can further partition PUs and TUs. For example, in HEVC, a residual quadtree (RQT) represents partitioning of TUs. In HEVC, PUs represent inter-predicted data, while TUs represent residual data. Intra-predicted CUs include intra-prediction information, such as an intra-mode indication.

[0047] As another example, video encoder 200 and video decoder 300 can be configured to operate according to VVC. According to VVC, a video coder (such as video encoder 200) partitions a picture into a plurality of coding tree units (CTUs). Video encoder 200 can partition a CTU according to a tree structure such as a quad-tree binary tree (QTBT) structure or a multi-type tree (MTT) structure. The QTBT structure removes the concepts of multiple partition types, such as the splitting of CUs, PUs, TUs of HEVC. The QTBT structure includes two levels: a first level partitioned according to quad-tree partitioning, and a second level partitioned according to binary tree partitioning. A root node of the QTBT structure corresponds to a CTU. Leaf nodes of the binary trees correspond to coding units (CUs).

[0048] In the MTT partitioning structure, a block can be partitioned using quad-tree (QT) partitioning, binary tree (BT) partitioning, and one or more types of ternary tree (TT) (also referred to as tri-tree) partitioning. Ternary or tri-tree partitioning is a partitioning that splits one block into three sub-blocks. In some examples, ternary or tri-tree partitioning divides a block into three sub-blocks without dividing the original block through a center split. The partitioning types (e.g., QT, BT, and TT) in the MTT can be symmetric or asymmetric.

[0049] In some examples, video encoder 200 and video decoder 300 can use a single QTBT or MTT structure to represent each of luma and chroma components, while in other examples, video encoder 200 and video decoder 300 can use two or more QTBT or MTT structures, such as one QTBT / MTT structure for luma components and another QTBT / MTT structure for two chroma components (or two QTBT / MTT structures for respective chroma components).

[0050] Video encoder 200 and video decoder 300 can be configured to use quad-tree partitioning per HEVC, QTBT partitioning, MTT partitioning, or other partition structures. For purposes of explanation of the techniques of this disclosure, the description of the techniques of this disclosure is presented with respect to QTBT partitioning. However, it should be understood that the techniques of this disclosure also apply to video coders configured to use quad-tree partitioning or other types of partitioning.

[0051] In some examples, a CTU includes a coding tree block (CTB) of luma samples, two corresponding CTBs of chroma samples of a picture having three sample arrays, or a CTB of samples of a monochrome picture or a picture coded using three separate color planes and syntax structures for coding samples. A CTB can be an NxN block of samples for some value of N, such that partitioning a component into CTBs is a partitioning. A component is one array or a single sample from one of three arrays (luma and two chroma) that make up a picture in a 4:2:0, 4:2:2, or 4:4:4 color format or a single sample array or group of arrays that make up a monochrome format picture. In some examples, a coding block is an MxN block of samples for some values of M and N, such that partitioning a CTB into coding blocks is a partitioning.

[0052] Blocks (e.g., CTUs or CUs) can be grouped in various ways in a picture. As one example, a brick can refer to a rectangular region of CTU rows within a particular tile in a picture. A tile can be a rectangular region of CTUs within a particular tile column and a particular tile row in a picture. A tile column refers to a rectangular region of CTUs whose height is equal to the height of the picture and whose width is specified by a syntax element (e.g., in a picture parameter set). A tile row refers to a rectangular region of CTUs whose height is specified by a syntax element (e.g., in a picture parameter set) and whose width is equal to the width of the picture.

[0053] In some examples, a tile can be partitioned into multiple bricks, each of which can include one or more CTU rows within the tile. A tile that is not partitioned into multiple bricks can also be referred to as a brick. However, a brick that is a proper subset of a tile can not be referred to as a tile.

[0054] Bricks in a picture can also be arranged in slices. A slice can be an integer number of bricks of a picture that can be exclusively contained in a single network abstraction layer (NAL) unit. In some examples, a slice includes a plurality of complete tiles or a contiguous sequence of complete bricks of only one tile.

[0055] The disclosure can use “NxN” and “N by N” interchangeably to refer to the sample dimensions of a block (e.g., a CU or other video block) in terms of vertical and horizontal dimensions, e.g., 16x16 samples or 16 by 16 samples. In general, a 16x16 CU will have 16 samples in the vertical direction (y = 16) and 16 samples in the horizontal direction (x = 16). Likewise, an NxN CU generally has N samples in the vertical direction and N samples in the horizontal direction, where N represents a nonnegative integer value. The samples in a CU can be arranged in rows and columns. Further, a CU need not necessarily have the same number of samples in the horizontal direction as in the vertical direction. For example, a CU can include NxM samples, where M need not necessarily equal N.

[0056] Video encoder 200 encodes video data of CUs that represent prediction and / or residual information, among other information. Prediction information indicates how to predict a CU in order to form a prediction block for the CU. Residual information generally represents sample-by-sample differences between the CU prior to encoding and the prediction block.

[0057] To predict a CU, video encoder 200 can generally form a prediction block for the CU through inter prediction or intra prediction. Inter prediction generally refers to predicting the CU from data of a previously coded picture, whereas intra prediction generally refers to predicting the CU from previously coded data of the same picture. To perform inter prediction, video encoder 200 can use one or more motion vectors to generate the prediction block. Video encoder 200 can generally perform a motion search to identify a reference block that closely matches the CU, e.g., according to a difference between the CU and the reference block. Video encoder 200 can calculate a difference metric using sum of absolute difference (SAD), sum of squared difference (SSD), mean absolute difference (MAD), mean squared difference (MSD), or other such difference calculations to determine whether a reference block closely matches a current CU. In some examples, video encoder 200 can use uni -prediction or bi-prediction to predict a current CU.

[0058] Some examples of VVC also provide an affine motion compensation mode, which can be considered an inter prediction mode. In affine motion compensation mode, video encoder 200 can determine two or more motion vectors that represent non-translational motion, such as scaling or zooming, rotation, perspective motion, or other irregular types of motion.

[0059] To perform intra prediction, video encoder 200 can select an intra prediction mode to generate the prediction block. Some examples of VVC provide sixty-seven intra prediction modes, including various directional modes, as well as a planar mode and a DC mode. Generally, video encoder 200 selects an intra prediction mode that describes neighboring samples of a current block (e.g., a block of a CU) to predict samples of the current block from the neighboring samples. Assuming video encoder 200 is coding CTUs and CUs in a raster scan order (from left to right, top to bottom), such samples can generally be above, above-left, and to the left of the current block in the same picture as the current block.

[0060] Video encoder 200 encodes data that represents a prediction mode for a current block. For example, for inter prediction modes, video encoder 200 can encode data that indicates which of various available inter prediction modes to use, as well as motion information for the corresponding mode. For example, for uni - or bi-prediction, video encoder 200 can encode motion vectors using advanced motion vector prediction (AMVP) or merge mode. Video encoder 200 can use similar modes to encode motion vectors for affine motion compensation modes.

[0061] Following prediction, e.g., intra prediction or inter prediction of a block, video encoder 200 can calculate residual data for the block. Residual data, such as a residual block, represents sample-by-sample differences between the block and a prediction block formed using the corresponding prediction mode. Video encoder 200 can apply one or more transforms to the residual block to produce transformed data in a transform domain instead of the sample domain. For example, 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, video encoder 200 can apply a secondary transform following the first transform, e.g., a mode-dependent non-separable secondary transform (MDNSST), a signal dependent transform, a Karhunen-Loeve transform (KLT), etc. Video encoder 200 produces transform coefficients following application of the one or more transforms.

[0062] As described above, following any transforms to produce transform coefficients, video encoder 200 can perform quantization of the transform coefficients. Quantization generally refers to a process in which transform coefficients are quantized to possibly reduce the amount of data used to represent the transform coefficients, provide further compression. By performing the quantization process, video encoder 200 can reduce the bit depth of some or all of the transform coefficients. For example, video encoder 200 can round n-bit values down to m-bit values during quantization, where n is greater than m. In some examples, to perform quantization, video encoder 200 can perform a bitwise right-shift of the values to be quantized.

[0063] Following quantization, video encoder 200 can scan the transform coefficients, producing a one-dimensional vector from the two-dimensional matrix including the quantized transform coefficients. The scan can be designed to place higher energy (and thus lower frequency) transform coefficients earlier in the vector and lower energy (and thus higher frequency) transform coefficients later in the vector. In some examples, 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 the vector. In other examples, video encoder 200 can perform an adaptive scan. Following scanning of the quantized transform coefficients to form a one-dimensional vector, video encoder 200 can entropy encode the one-dimensional vector, e.g., according to context adaptive binary arithmetic coding (CABAC). Video encoder 200 can also entropy encode values for syntax elements describing metadata associated with encoded video data used by video decoder 300 when decoding the video data.

[0064] To perform CABAC, video encoder 200 can assign a context within a context model to a symbol to be transmitted. The context can relate to, for example, whether neighboring values of the symbol are zero-valued or not. The probability determination can be based on a context assigned to the symbol.

[0065] Video encoder 200 can also generate syntax data, such as block-based syntax data, picture-based syntax data, and sequence-based syntax data, e.g., in picture headers, block headers, slice headers, or other syntax data, such as sequence parameter sets (SPS), picture parameter sets (PPS), or video parameter sets (VPS), to video decoder 300. Video decoder 300 can likewise decode such syntax data to determine how to decode corresponding video data.

[0066] In this way, video encoder 200 can generate a bitstream that includes encoded video data, e.g., syntax elements that describe partitioning of a picture into blocks (e.g., CUs) and prediction and / or residual information for the blocks. Ultimately, video decoder 300 can receive the bitstream and decode the encoded video data.

[0067] In general, video decoder 300 performs a reciprocal process to that performed by video encoder 200 to decode the encoded video data of the bitstream. For example, video decoder 300 can decode values for syntax elements of the bitstream using CABAC in a manner substantially similar to, but reciprocal to, the CABAC encoding process of video encoder 200. The syntax elements can define partitioning information for partitioning a picture into CTUs, and partitioning each CTU according to a corresponding partition structure, such as a QTBT structure, to define CUs of the CTU. The syntax elements can further define prediction and residual information for blocks (e.g., CUs).

[0068] The residual information can be represented by, for example, quantized transform coefficients. Video decoder 300 can inverse quantize and inverse transform the quantized transform coefficients of a block to reproduce a residual block for the block. Video decoder 300 forms a prediction block for the block using the signaled prediction mode (intra- or inter-prediction) and related prediction information (e.g., motion information for inter-prediction). Video decoder 300 can then combine the prediction block and the residual block (on a sample-by-sample basis) to reproduce the original block. Video decoder 300 can perform additional processing, such as performing a deblocking process to reduce visual artifacts along block boundaries.

[0069] The disclosure can generally refer to“signaling” certain information, such as syntax elements. The term“signaling” can generally refer to a communication of values for syntax elements and / or other data used to decode encoded video data. That is, video encoder 200 can signal values for syntax elements in a bitstream. Generally, signaling refers to generating a value in a bitstream. As described above, source device 102 can transmit the bitstream to destination device 116 in substantially real time or non-real time, such as can occur, for example, when storing syntax elements to storage device 112 for later retrieval by destination device 116.

[0070] Figure 2A and Figure 2B is a conceptual diagram illustrating an example quad-tree binary-tree (QTBT) structure 130 and a corresponding coding tree unit (CTU) 132. Solid lines represent quad-tree splits, and dashed lines represent binary-tree splits. In each split (i.e., non-leaf) node of the binary tree, one flag is signaled to indicate which split type (i.e., horizontal or vertical) is used, where 0 represents a horizontal split and 1 represents a vertical split in this example. For quad-tree splits, no split type indication is needed because a quad-tree node splits one block horizontally and vertically into 4 equally-sized sub-blocks. Thus, video encoder 200 can encode and video decoder 300 can decode syntax elements (e.g., split information) for the region tree level (i.e., solid lines) of the QTBT structure 130 and syntax elements (e.g., split information) for the prediction tree level (i.e., dashed lines) of the QTBT structure 130. For a CU represented by a terminal leaf node of the QTBT structure 130, video encoder 200 can encode video data, such as prediction and transform data, and video decoder 300 can decode video data, such as prediction and transform data.

[0071] Generally, Figure 2B The CTU 132 of FIG. 13A can be associated with parameters defining the sizes of blocks at the first and second levels corresponding to nodes of the QTBT structure 130. These parameters can include a CTU size (representing the size of the CTU 132 in samples), a minimum quad-tree size (MinQTSize, representing the allowed minimum quad-tree leaf node size), a maximum binary-tree size (MaxBTSize, representing the allowed maximum binary-tree root node size), a maximum binary-tree depth (MaxBTDepth, representing the allowed maximum binary-tree depth), and a minimum binary-tree size (MinBTSize, representing the allowed minimum binary-tree leaf node size).

[0072] The root node of the QTBT structure corresponding to a CTU can have four child nodes at the first level of the QTBT structure, each of which can be partitioned according to quadtree partitioning. That is, a node at the first level is either a leaf node (having no child nodes) or has four child nodes. The example of QTBT structure 130 represents such nodes as including a parent node and child nodes having solid lines for branches. If a node at the first level is not larger than a maximum allowed binary tree root node size (MaxBTSize), the node can be further partitioned by a corresponding binary tree. Binary tree splitting for a node can be iterated until the split produces nodes that reach an allowed minimum binary tree leaf node size (MinBTSize) or an allowed maximum binary tree depth (MaxBTDepth). The example of QTBT structure 130 represents such nodes as having dashed lines for branches. Binary tree leaf nodes are referred to as coding units (CUs), which are used for prediction (e.g., intra- or inter-prediction) and transform without any further partitioning. As noted above, a CU can also be referred to as a “video block” or “block.”

[0073] In one example of a QTBT partitioning structure, the CTU size is set to 128x128 (luma samples and two corresponding 64x64 chroma samples), MinQTSize is set to 16x16, MaxBTSize is set to 64x64, MinBTSize (for both width and height) is set to 4, and MaxBTDepth is set to 4. Quadtree partitioning is first applied to the CTU to generate quadtree leaf nodes. The quadtree leaf nodes can be from 16x16 (i.e., MinQTSize) to 128x128 (i.e., the CTU size) in size. If a quadtree leaf node is 128x128, the leaf quadtree node will not be further split by a binary tree because the size exceeds MaxBTSize (i.e., 64x64 in this example). Otherwise, the quadtree leaf node will be further partitioned by a binary tree. Thus, the quadtree leaf node is also a root node of a binary tree, and the binary tree depth is 0. When the binary tree depth reaches MaxBTDepth (4 in this example), no further splitting is allowed. A binary tree node with a width equal to MinBTSize (4 in this example) means that no further vertical split (i.e., division of the width) is allowed for that binary tree node. Similarly, a binary tree node with a height equal to MinBTSize means that no further horizontal split (i.e., division of the height) is allowed for that binary tree node. As noted above, leaf nodes of a binary tree are referred to as CUs and are further processed according to prediction and transform without further partitioning.

[0074] Most video coding standards since H.261 are based on the so-called hybrid video coding principle, which combines the advantages of Figure 3The term hybrid refers to the combination of two means for reducing redundancy in video signals, namely prediction with prediction residual quantization and transform coding. Prediction and transform reduce redundancy in video signals by decorrelation, while quantization reduces the data represented by transform coefficients by lowering their precision, ideally by removing only uncorrelated details. This hybrid video coding design principle is also used in the two recent standards HEVC and VVC.

[0075] A modern hybrid video coder can consist of the following building blocks as shown in the example of Figure 3 In other words, Figure 3 is a block diagram illustrating an example hybrid video coding framework 320. In the example of Figure 3 The framework 320 receives pictures 322 of input video data and splits the pictures 322 into blocks. Block partitioning is used to divide an image into smaller blocks for the operations of the prediction and transform processes. Early video coding standards used fixed block sizes, typically 16x16 samples. Recent standards such as HEVC and VVC employ tree-based partitioning structures to provide flexible partitioning, for example, as shown in Figure 2A and Figure 2B

[0076] The framework 320 includes a motion estimation unit 324, an intra prediction unit 326, and an inter prediction unit 328. The motion estimation unit 324 can perform motion estimation based on previously reconstructed pictures 330 of the video data. The inter prediction unit 328 can generate a prediction block using motion information generated by the motion estimation unit 324. Motion-compensated prediction or inter-picture prediction (i.e., inter prediction) exploits the redundancy that exists between pictures of a video sequence (hence “inter- picture”). In block-based motion compensation used in modern video codecs, the prediction is obtained from one or more previously decoded pictures (i.e., reference pictures). The corresponding region used to generate the inter prediction is indicated by motion information, including a motion vector and a reference picture index.

[0077] The intra prediction unit 326 can generate a prediction block using intra prediction. Intra-picture prediction exploits spatial redundancy that exists within a picture (hence called “intra-picture”) by deriving a prediction for a block from spatially neighboring (reference) samples that have been coded / decoded. Directional angular prediction, DC prediction, and planar or plane-up prediction are used in recent video codecs, including AVC, HEVC, and VVC. The framework 320 can select between the prediction block generated by the inter prediction unit 328 and the prediction block generated by the intra prediction unit 326.

[0078] ​A residual generation unit 332 of the framework 320 can generate residual data based on the prediction block and the corresponding block of the picture 322. A transform unit 334 can apply a transform to the residual data to generate transform coefficients. Hybrid video coding standards apply a block transform to the prediction residual, whether the prediction residual is from inter-picture or intra-picture prediction. In early standards, including H.261 / 262 / 263, a discrete cosine transform (DCT) was employed. In HEVC and VVC, more transform kernels can be applied in addition to DCT in order to cope with different statistical information in specific video signals.

[0079] A quantization unit 336 can then apply quantization to the transform coefficients. 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 these values. In hybrid video coding, quantization is typically applied to individual transformed residual samples, i.e., transform coefficients, resulting in integer coefficient levels. In recent video coding standards, the step size used in quantization is derived from a so-called quantization parameter (QP) that controls the fidelity and bitrate. A larger step size reduces the bitrate, but also reduces the quality, which can result in blocking artifacts and fuzzy details in the video pictures.

[0080] An entropy encoding unit 338 can apply entropy encoding to syntax elements representing the transform coefficients. The framework 320 can output the entropy encoded syntax elements in a bitstream. Entropy encoding techniques, such as context adaptive binary arithmetic coding (CABAC), are used in recent video codecs, such as AVC, HEVC, and VVC, due to their high efficiency.

[0081] Furthermore, in Figure 3 In an example, an inverse quantization unit 340 can inverse quantize the transform coefficients. An inverse transform unit 342 can apply an inverse transform to the inverse quantized transform coefficients to generate residual data. A reconstruction unit 344 can reconstruct the block based on the prediction block and the residual data. A loop filter unit 346 can apply one or more filters to the picture containing the reconstructed block.

[0082] In recent video codecs, a hierarchical prediction structure within a group of pictures (GOP) is applied to improve coding efficiency. Figure 4 is a conceptual diagram of an example hierarchical prediction structure 400 that illustrates a group of pictures (GOP) size equal to 16. In Figure 4 In an example, the picture I0 is an intra picture that is not coded using inter prediction. The other pictures (i.e., pictures B1 to B16) in Figure 4 may be coded using inter prediction. Specifically, Figure 4 Each arrow in points from a first picture to a second picture indicates that the second picture can be coded using the first picture.

[0083] Post-loop / in-loop filtering is a filtering process (or a combination of such processes) applied to reconstructed images to reduce decoding artifacts. The input to the filtering process is typically the reconstructed image, which is a combination of the reconstructed residual signal (including quantization errors) and the prediction. For example... Figure 3 As shown, the reconstructed image 330 after in-loop filtering is stored and used as a reference for inter-image prediction of subsequent decoded images. Decoding artifacts are mainly determined by QP, therefore QP information is generally used in the design of the filtering process. In HEVC, the in-loop filter includes deblocking filtering and Sample Adaptive Offset (SAO) filtering. In VVC, an Adaptive Loop Filter (ALF) is introduced as a third filter. The ALF filtering process is shown below.

[0084] R′(i,j)=R(i,j)+((∑ k≠0 ∑ l≠0 f(k,l)×K(R(i+k,j+l)-R(i,j),c(k,l))+64)>>7) (1)

[0085] In equation (1), R(i,j) represents the sample before the filtering process, R'(i,j) represents the sample value after the filtering process, f(k,l) represents the filter coefficients, K(x,y) is the clipping function, and c(k,l) represents the clipping parameter. The variables k and l are... arrive The values ​​vary between L and L, where L represents the filter length. The clipping function K(x,y) = min(y,max(-y,x)), which corresponds to the function Clip3(-y,y,x). The clipping operation introduces nonlinearity to make ALF more efficient by reducing the influence of neighboring sample values ​​that differ too much from the current sample value. In VVC, the filter parameters can be signaled in the bitstream or selected from a predefined set of filters. The ALF filtering process can also be summarized by the following equation.

[0086] R'(i,j)=R(i,j)+ALF_residual_ouput(R) (2)

[0087] Multiple studies have demonstrated that embedding neural networks (NNs) into hybrid video decoding frameworks can improve compression efficiency. NNs have been used in modules for intra-frame and inter-frame prediction to enhance prediction efficiency. In recent years, NN-based intra-loop filters have also been an active research topic. In some examples, the filtering process is applied as a post-filter. When the filtering process is applied as a post-filter, it is only applied to the output image, and the unfiltered image is used as a reference image. In contrast, when the filtering process is applied as an intra-loop filter, it is applied to the reconstructed image used as the output and then as the reference image.

[0088] The NN-based filter can additionally be applied on top of existing filters, such as the deblocking filter, SAO, or ALF. The NN-based filter can also be exclusively applied, where the NN-based filter is designed to replace all existing filters.

[0089] Figure 5 is a conceptual diagram illustrating a convolutional neural network (CNN)-based filter 500 with four layers. The CNN-based filter 500 is a particular type of NN-based filter that uses convolutional layers. As shown, the NN-based filtering process can take reconstructed samples as input, and the intermediate output 501 is residual samples that are added back to the input to refine the input samples. The CNN-based filter 500 can use all color components (e.g., luma, Cb chroma, Cr chroma) as input to exploit cross-component correlation. Different color components can share the same filter (including network structure and model parameters), or each color component can have its own specific filter. Figure 5

[0090] In particular, the input to the CNN-based filter 500 includes a plurality of sample blocks 502 of a reconstructed picture. Each sample block 502 has a size of NxN. The sample blocks 502 are rectangular arrays of samples. The sample blocks 502 can or can not correspond to coding blocks, prediction blocks, transform blocks, coding tree blocks, or other types of blocks used in the coding process. However, the present disclosure is not limited to square sample blocks, but can generally apply to rectangular sample blocks with size NxM. Chroma blocks have half the number of samples in the vertical direction and half the number of samples in the horizontal direction as luma blocks. Thus, for each NxN chroma block, the input to the filter 500 can include four NxN luma blocks. In the example of Figure 5 The luma blocks are shaded, and the chroma blocks are white. Thus, the sample blocks received as input to the filter 500 can be represented as Yx4 + U + V, where Y corresponds to luma, U corresponds to a first chroma component, and V corresponds to a second chroma component.

[0091] In addition, the CNN-based filter 500 includes an input layer 504, hidden layers 506A-506B (collectively, “hidden layers 506”), and an output layer 508. In particular, the input layer 504 receives the input sample blocks 502. The hidden layers 506A-506B are configured to process the input sample blocks 502 to generate the output 508. The output layer 508 outputs the output 508. Figure 5 ​In the example, the CNN-based filter 500 includes two hidden layers, for a total of four layers. Each of the input layer 504, hidden layer 506, and output layer 508 includes one of the convolutional layers 510A to 510D (collectively referred to as "convolutional layer 510"). Additionally, the input layer 504 and each hidden layer 506 include one of the PReLU activation layers 512A to 512C (collectively referred to as "PReLU activation layer 512"). Each PReLU activation layer 512 takes the output of the convolutional layer as input and applies the PReLU activation function to the input.

[0092] Throughout this disclosure, the notation axbxcxd is used to describe a convolutional layer. The values ​​a and b can correspond to the width and height of a two-dimensional array of input locations. Each input location has c input channels. Therefore, the convolutional layer receives axbxc input values. The value d represents the number of output channels of the convolutional layer. Each output channel corresponds to a distinct set of trainable weights applied by the convolutional layer. The number of weights in the weight set corresponding to an output channel is equal to axbxc. The value of an output channel (i.e., the output value) can be computed by convolving the input value with the set of weights corresponding to the output channel. For example, the value of an output channel can be computed as the dot product of the input value and the set of weights corresponding to the output channel. Because there are d output channels, the convolutional layer is associated with axbxcxd weights.

[0093] In order to discuss Figure 5 The example illustrates this; convolutional layer 510A is described as 3x3x6x8. Therefore, the input to convolutional layer 510A can be a set of six values ​​corresponding to each chroma sample position in a 3x3 square of chroma sample positions centered around the current chroma sample position. Each chroma sample position corresponds to the position of a chroma sample in block 502. In this example, the chroma block is half the width and half the height of the luma block. Therefore, each chroma sample position has four luma sample positions. Each chroma sample position has two types of chroma samples, resulting in six sample values ​​per chroma sample position. Convolutional layer 510A has eight output channels and thus generates eight output values. Figure 5In the example of FIG. 5, PReLU activation layer 512A applies a PReLU activation function to the 8 output values generated by convolution layer 510A. Convolution layer 510A can repeat this operation for all or some of the chroma sample positions of block 502 as the current chroma sample position. The set of 8 values output by PReLU activation layer 512A can be treated as the values corresponding to a single position in the two-dimensional array of positions used as input to convolution layer 510B. Convolution layers 510B, 510C, and 510D operate in a similar manner. Convolution layer 510D has 6 output channels, corresponding to the 6 sample values (i.e., 4 luma and 2 chroma) corresponding to the current chroma sample position. In other examples, convolution layer 510A can have different values for a, b, c, and d than 3, 3, 6, and 8. However, in other words, the present disclosure is not limited to 3x3 filter kernels for chroma sample positions, but can apply to a x b filter kernels in general, including non-square filter kernels. Furthermore, padding can be used at block boundaries, CTU boundaries, and / or picture boundaries.

[0094] Output layer 508 of CNN-based filter 500 generates intermediate output 501. Intermediate output 501 can include residual samples. These residual samples can be different than the residual samples generated by residual generation unit 332. CNN-based filter 500 generates filtered block 514 by adding the residual samples of intermediate output 501 to the corresponding samples of sample block 502. Filtered block 514 can include two NxN chroma blocks and four NxN luma blocks.

[0095] Figure 5 The filtering process of FIG. 6 can be summarized as follows:

[0096] R'(i,j) = R(i,j) + NN_filter_residual_output(R) (3)

[0097] Figure 6A is a conceptual diagram illustrating a ReLU activation function 600. As shown in Figure 6A ReLU activation function 600 can be defined as:

[0098]

[0099] Figure 6B is a conceptual diagram illustrating a PReLU activation function 650. As shown in Figure 6B PReLU activation function 650 can be defined as:

[0100]

[0101] In Equation 5, a is a value trained as part of training a neural network that includes a PReLU activation function.

[0102] The model structure and model parameters of the NN-based filter can be predefined and stored in the video encoder 200 and the video decoder 300. The model structure and model parameters of the NN-based filter can also be signaled (e.g., by the video encoder 200) in the bitstream. For example, the video encoder 200 can encode one or more syntax elements with values indicative of the model structure and model parameters (e.g., of the NN-based filter), and the video decoder 300 can decode one or more syntax elements with values indicative of the model structure and model parameters (e.g., of the NN-based filter). The video encoder 200 can encode the syntax elements in any suitable syntax structure, such as a sequence parameter set (SPS), a picture parameter set (PPS), a video parameter set (VPS), a slide title, and so on.

[0103] It has been found that the performance of PReLU is superior to ReLU in most cases. However, PReLU also sometimes makes the training process unstable. Therefore, training a NN-based filter using a PReLU activation function can take a longer time, or can deviate from the optimal solution.

[0104] In this disclosure, it is proposed to use LeakyReLU as the activation function in the NN-based filtering process. LeakyReLU is shown in the following equation (6).

[0105]

[0106] In equation (6), the parameter Alpha is a fixed value. That is, the parameter Alpha does not change during the training process of the neural network that contains the LeakyReLU activation function. In contrast, the parameter a in PReLU is a trainable variable, which can change during the training process.

[0107] As one example, it is proposed Figure 7 a CNN-based filter shown in the example of Figure 7 is a conceptual diagram of an example CNN-based filter 700 with 14 layers and LeakyReLU as the activation function according to one or more techniques of this disclosure. In Figure 7 In the example of Figure 7In the example of FIG. 7, the number of samples in the vertical direction of the chroma blocks is half of the luma blocks, and the number of samples in the horizontal direction of the chroma blocks is half of the luma blocks. Thus, for each NxN chroma block, the input to the CNN-based filter 700 can include four NxN luma blocks. In the example of FIG. 7, the CNN-based filter 700 includes an input layer 704, hidden layers 706A-706N (collectively, “hidden layers 706”), and an output layer 708. In the example of FIG. 7, the input layer 704 includes one of the four NxN luma blocks as input. In other examples, the input layer 704 can include more than one luma block as input. Figure 7 In the example of FIG. 7, the luma blocks are shaded and the chroma blocks are white. Thus, the sample block 702 received as input to the filter 700 can be represented as Yx4 + U + V, where Y corresponds to luma, U corresponds to a first chroma component, and V corresponds to a second chroma component.

[0108] In addition, the CNN-based filter 700 includes the input layer 704, the hidden layers 706A-706N (collectively, “hidden layers 706”), and the output layer 708. In the example of FIG. 7, the input layer 704 includes one of the four NxN luma blocks as input. In other examples, the input layer 704 can include more than one luma block as input. Figure 7 In the example of FIG. 7, the CNN-based filter 700 includes 12 hidden layers, for a total of 14 layers. In other examples, the CNN-based filter 700 can have more or fewer layers than 14. Each of the input layer 704, the hidden layers 706, and the output layer 708 includes one of the convolutional layers 710A-710N (collectively, “convolutional layers 710”). In addition, the input layer 704 and each of the hidden layers 706 includes one of the LeakyReLU activation layers 712A-712M (collectively, “LeakyReLU activation layers 712”). Each of the LeakyReLU activation layers 712 takes as input the output of a convolutional layer and applies a LeakyReLU activation function to the input.

[0109] The output layer 708 of the CNN-based filter 700 generates an intermediate output 701. The intermediate output 701 can include residual samples. These residual samples can be different than the residual samples generated by the residual generation unit 332. The CNN-based filter 700 generates a filtered block 714 by adding the residual samples of the intermediate output 701 to the corresponding samples of the sample block 702. The filtered block 714 can include two NxN chroma blocks and four NxN luma blocks.

[0110] In one example, the optimal value of Alpha is obtained based on statistics of the trainable values of PReLUs, and the same Alpha value is used for all layers. For example, video encoder 200 and video decoder 300 can use the same predetermined value of Alpha in all LeakyReLU activation layers 712 of CNN-based filter 700. In this example, it can not be necessary for video encoder 200 to signal the value of Alpha in the bitstream. In some examples, to obtain the optimal value of Alpha based on statistics of the values of the trainable values (e.g., a) of PReLUs, a system (e.g., by video encoder 200, video decoder 300, or another device or system) can train a network using PReLU activation functions until the training has converged. The system can then use the statistics of the trainable values (e.g., a) of the PReLU activation functions for each layer. For example, the system can determine the optimal value of Alpha to be the average of the trainable values of the PReLU activation functions of the layers.

[0111] As another example, the optimal value of Alpha is obtained based on statistics of the values of the trainable values of PReLUs for each layer, each of which can use its own fixed value of Alpha. For example, for each of LeakyReLU activation layers 712, video encoder 200 and video decoder 300 can use the same predetermined layer-specific value of Alpha in the LeakyReLU activation layer, but different predetermined values of Alpha can be used in different LeakyReLU activation layers 712. In this example, it can not be necessary for video encoder 200 to signal the value of Alpha in the bitstream. In some examples, to obtain the optimal value of Alpha based on statistics of the values of the trainable values (e.g., a) of PReLUs for each layer, a system (e.g., by video encoder 200, video decoder 300, or another device or system) can train a network using PReLU activation functions until the training has converged. The system can then use the statistics of the trainable values (e.g., a) of the PReLU activation functions for each layer. For example, the system can determine the optimal value of Alpha for a layer to be equal to the trainable value of the PReLU activation function for the corresponding layer. In another example, the system can determine the optimal value of Alpha for a layer to be equal to the average of two or more trainable values of PReLU activation functions for two or more layers.

[0112] As another example, the Alpha value in a NN-based filter (e.g., CNN-based filter 700) is a parameter that can be specified when applying the NN-based filter to a picture. For example, in one possible use case involving video coding, at the encoder side, video encoder 200 decides what Alpha value should be used. For example, video encoder 200 can apply a NN-based filter with multiple different predetermined Alpha values and select which one of the Alpha values produces the best results. Video encoder 200 can determine which Alpha value produces the best results by comparing the reconstructed blocks or pictures generated using the different Alpha values (e.g., sum of absolute differences, etc.). Furthermore, video encoder 200 can encode the information needed to derive the Alpha value as a syntax element in the bitstream. Video decoder 300 reconstructs the Alpha value selected by video encoder 200 and performs the filtering process. Examples of ways to signal the Alpha value include signaling the Alpha value in the bitstream or signaling an index of the selected Alpha value within a predefined set. A typical value for the Alpha value can be ¼, but other values are possible as well.

[0113] Figure 8 FIG. 1 is a block diagram illustrating an example video encoder 200 that can perform the techniques of this disclosure. Figure 8 FIG. 1 is a block diagram illustrating an example video encoder 200 that can perform the techniques of this disclosure.

[0114] In Figure 8In the example of FIG. 8, video encoder 200 includes video data memory 830, mode select unit 802, residual generation unit 804, transform processing unit 806, quantization unit 808, inverse quantization unit 810, inverse transform processing unit 812, reconstruction unit 814, filter unit 816, decoded picture buffer (DPB) 818, and entropy encoding unit 820. Any or all of video data memory 830, mode select unit 802, residual generation unit 804, transform processing unit 806, quantization unit 808, inverse quantization unit 810, inverse transform processing unit 812, reconstruction unit 814, filter unit 816, DPB 818, and entropy encoding unit 820 can be implemented in one or more processors or in processing circuitry. For example, the units of video encoder 200 can be implemented as one or more circuits or logic elements as part of hardware circuitry, or as part of a processor, ASIC, or FPGA. Also, video encoder 200 can include additional or alternative processors or processing circuitry to perform these and other functions.

[0115] Video data memory 830 can store video data to be encoded by the components of video encoder 200. Video encoder 200 can receive the video data stored in video data memory 830 from, for example, video source 104 Figure 1 DPB 818 can act as a reference picture memory that stores reference video data for use in prediction of subsequently coded video data by video encoder 200. Video data memory 830 and DPB 818 can be formed by any of a variety of memory devices, such as dynamic random access memory (DRAM) including synchronous DRAM (SDRAM), magneto resistive RAM (MRAM), resistive RAM (RRAM), or other types of memory devices. Video data memory 830 and DPB 818 can be provided by the same memory device or separate memory devices. In various examples, video data memory 830 can be on-chip with other components of video encoder 200, as illustrated, or off-chip relative to those components.

[0116] In this disclosure, reference to video data memory 830 should not be interpreted as being limited to memory internal to video encoder 200 (unless so specifically described) or memory external to video encoder 200 (unless so specifically described). Rather, reference to video data memory 830 should be understood as reference to a memory that stores video data that video encoder 200 receives for encoding (e.g., video data of a current block to be encoded). Figure 1 Memory 106 of source device 102 can also provide temporary storage of the output from the various units of video encoder 200.

[0117] Figure 8 The various units described for video encoder 200 are replicated for each slice of each picture of each view of the multi-view video sequence being encoded. The various units of video encoder 200 are described to assist with understanding the operations performed by video encoder 200. The units can be implemented as fixed- function circuitry, programmable circuitry, or a combination thereof. Fixed-function circuitry refers to circuitry that provides particular functionality, and is preset on the operations that can be performed. Programmable circuitry refers to circuitry that can be programmed to perform various tasks, and provides flexible functionality in the operations that can be performed. For example, programmable circuitry can execute software or firmware that cause the programmable circuitry to operate in the manner defined by instructions of the software or firmware. Fixed-function circuitry can execute software instructions (e.g., to receive parameters or output parameters), but the types of operations that the fixed-function circuitry performs are typically immutable. In some examples, one or more of the units can be distinct circuit blocks (fixed-function or programmable), and in some examples, one or more of the units can be integrated circuitry.

[0118] Video encoder 200 can include arithmetic logic units (ALUs), elementary function units (EFUs), digital circuits, analog circuits, and / or programmable cores, formed from programmable circuits. In examples where the operations of video encoder 200 are performed using software executed by the programmable circuits, memory 106 Figure 1 ) can store the instructions (e.g., object code) of the software that video encoder 200 receives and executes, or another memory within video encoder 200 (not shown) can store such instructions.

[0119] Video data memory 830 is configured to store video data to be encoded. Video encoder 200 can retrieve pictures of the video data from video data memory 830 and provide the video data to residual generation unit 804 and mode selection unit 802. Video data in video data memory 830 can be raw video data that is awaiting encoding.

[0120] Mode selection unit 802 includes motion estimation unit 822, motion compensation unit 824, and intra prediction unit 826. Mode selection unit 802 can include additional functional units to perform video prediction from other prediction modes. As examples, mode selection unit 802 can include a palette unit, an intra-block copy unit (which can be part of motion estimation unit 822 and / or motion compensation unit 824), an affine unit, a linear model (LM) unit, and so on.

[0121] The mode selection unit 802 generally coordinates multiple encoding passes to test combinations of encoding parameters and resulting rate-distortion values for such combinations. The encoding parameters can include partitioning of CTUs into CUs, prediction modes for CUs, transform types for residual data of CUs, quantization parameters for residual data of CUs, and so on. The mode selection unit 802 can ultimately select the combination of encoding parameters that has a better rate-distortion value than other tested combinations.

[0122] The video encoder 200 can partition a picture retrieved from the video data memory 830 into a series of CTUs, and encapsulate one or more CTUs within a slice. The mode selection unit 802 can partition the CTUs of the picture in a tree structure, such as the QTBT structure or the quad-tree structure of HEVC described above. As described above, the video encoder 200 can form one or more CUs by partitioning a CTU according to the tree structure. Such CUs can also be referred to generally as “video blocks” or “blocks.”

[0123] In general, the mode selection unit 802 also controls its components (e.g., the motion estimation unit 822, the motion compensation unit 824, and the intra prediction unit 826) to generate a prediction block for a current block (e.g., a current CU, or in HEVC, an overlapping portion of a PU and a TU). For inter prediction of a current block, the motion estimation unit 822 can 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 coded pictures stored in the DPB 818). In particular, the motion estimation unit 822 can calculate a value that represents how similar a potential reference block is to the current block, e.g., according to a sum of absolute differences (SAD), a sum of squared differences (SSD), a mean absolute difference (MAD), a mean squared difference (MSD), and the like. The motion estimation unit 822 can generally perform these calculations using sample-by-sample differences between the current block and the reference block under consideration. The motion estimation unit 822 can identify the reference block with the lowest value resulting from these calculations, to indicate the reference block that most closely matches the current block.

[0124] Motion estimation unit 822 can form one or more motion vectors (MVs) that define a position of a reference block in a reference picture relative to a position of a current block in a current picture. Motion estimation unit 822 can then provide the motion vector(s) to motion compensation unit 824. For example, for single prediction, motion estimation unit 822 can provide a single motion vector, while for bi-prediction, motion estimation unit 822 can provide two motion vectors. Motion compensation unit 824 can then generate a prediction block using the motion vector(s). For example, motion compensation unit 824 can use the motion vector(s) to retrieve data for a reference block. As another example, if the motion vector(s) have fractional sample precision, motion compensation unit 824 can interpolate values of the prediction block from one or more interpolation filters. Moreover, for bi-prediction, motion compensation unit 824 can retrieve data for two reference blocks identified by the respective motion vectors and combine the retrieved data, e.g., by a sample-wise average or weighted average.

[0125] As another example, for intra prediction or intra-prediction coding, intra-prediction unit 826 can generate a prediction block from samples neighboring the current block. For example, for directional modes, intra-prediction unit 826 can mathematically combine values of the neighboring samples and fill these calculated values across a defined direction of the current block to produce the prediction block. As another example, for DC mode, intra-prediction unit 826 can calculate an average of the neighboring samples of the current block and generate the prediction block to include this resulting average for each sample of the prediction block.

[0126] Mode selection unit 802 provides the prediction block to residual generation unit 804. Residual generation unit 804 receives the original, unencoded version of the current block from video data store 830 and the prediction block from mode selection unit 802. Residual generation unit 804 calculates the sample-wise difference between the current block and the prediction block. The resulting sample-wise difference defines a residual block for the current block. In some examples, residual generation unit 804 can also determine the difference between sample values in the residual block to generate the residual block using residual differential pulse code modulation (RDPCM). In some examples, residual generation unit 804 can form using one or more subtractor circuits that perform binary subtraction.

[0127] In examples where the mode select unit 802 partitions the CU into PUs, each PU can be associated with a luma prediction unit and corresponding chroma prediction units. Video encoder 200 and video decoder 300 can support PUs having various sizes. As described above, the size of a CU can refer to the size of the CU's luma coding block and the size of a PU can refer to the size of the PU's luma prediction unit. Assuming that the size of a particular CU is 2Nx2N, video encoder 200 can support 2Nx2N or NxN PU sizes for intra-prediction, and 2Nx2N, 2NxN, Nx2N, NxN, or similar symmetric PU sizes for inter-prediction. Video encoder 200 and video decoder 300 can also support asymmetric partitioning for inter-prediction of the PU sizes 2NxnU, 2NxnD, nLx2N, and nRx2N.

[0128] In examples where the mode select unit 802 does not further partition the CU into PUs, each CU can be associated with a luma coding block and corresponding chroma coding blocks. As above, the size of a CU can refer to the size of the CU's luma coding block. Video encoder 200 and video decoder 300 can support CU sizes of 2Nx2N, 2NxN, or Nx2N.

[0129] For other video coding techniques, such as intra-block copy mode coding, affine mode coding, and linear model (LM) mode coding, as some examples, the mode select unit 802 generates, via a respective unit associated with the coding technique, a prediction block for the current block being coded. In some examples, such as palette mode coding, the mode select unit 802 can not generate a prediction block, but rather generate syntax elements indicative of a way to reconstruct the block based on a selected palette. In such modes, the mode select unit 802 can provide the syntax elements to the entropy encoding unit 820 for encoding.

[0130] As described above, the residual generation unit 804 receives video data for the current block and the corresponding prediction block. The residual generation unit 804 then generates a residual block for the current block. To generate the residual block, the residual generation unit 804 calculates the sample-by-sample difference between the prediction block and the current block.

[0131] The transform processing unit 806 applies one or more transforms to the residual block to produce a transform coefficient block (referred to herein as a "transform coefficient block"). The transform processing unit 806 can apply various transforms to the residual block to form the transform coefficient block. For example, the transform processing unit 806 can apply a discrete cosine transform (DCT), a directional transform, a Karhunen-Loeve transform (KLT), or a conceptually similar transform. In some examples, the transform processing unit 806 can perform multiple transforms, such as a primary transform and a secondary transform, e.g., a rotational transform, on the residual block. In some examples, the transform processing unit 806 does not apply a transform to the residual block.

[0132] Quantization unit 808 can quantize the transform coefficients in a transform coefficient block, to produce quantized transform coefficient blocks. Quantization unit 808 can quantize transform coefficients of a transform coefficient block according to a quantization parameter (QP) value associated with the current block. Video encoder 200 (e.g., via mode selection unit 802) can adjust the degree of quantization applied to the transform coefficient blocks associated with the current block by adjusting the QP value associated with the CU. Quantization can introduce loss of information, and thus, quantized transform coefficients can have lower precision than the original transform coefficients produced by transform processing unit 806.

[0133] Inverse quantization unit 810 and inverse transform processing unit 812 can apply inverse quantization and inverse transform, respectively, to a quantized transform coefficient block to reconstruct a residual block from the transform coefficient block. Reconstruction unit 814 can produce a reconstructed block corresponding to the current block (albeit potentially with some degree of distortion) based on the reconstructed residual block and the prediction block generated by mode selection unit 802. For instance, reconstruction unit 814 can add samples of the reconstructed residual block to corresponding samples from the prediction block generated by mode selection unit 802 to produce the reconstructed block.

[0134] Filter unit 816 can perform one or more filter operations on the reconstructed block. For instance, filter unit 816 can perform a deblocking operation to reduce blockiness artifacts along the edges of the CU. In some examples, the operations of filter unit 816 can be skipped. In some examples, filter unit 816 can apply a CNN-based filter that uses a LeakyReLU activation function, such as CNN-based filter 700. Thus, in some examples, reconstruction unit 814 can reconstruct a block of a picture, where the block can be a square or rectangular region of the picture and can or can not correspond to a coding block, prediction block, and so forth. Filter unit 816 can apply CNN-based filter 700 to the block, where CNN-based filter 700 uses a LeakyReLU activation function.

[0135] Video encoder 200 stores the reconstructed block in DPB 818. For instance, in examples where the operations of filter unit 816 are not performed, reconstruction unit 814 can store the reconstructed block to DPB 818. In examples where the operations of filter unit 816 are performed, filter unit 816 can store the filtered reconstructed block to DPB 818. Motion estimation unit 822 and motion compensation unit 824 can retrieve reference pictures formed from reconstructed (and potentially filtered) blocks in DPB 818 to inter-predict blocks of subsequently encoded pictures. Moreover, intra-prediction unit 826 can use reconstructed blocks in DPB 818 of the current picture to intra-predict other blocks in the current picture.

[0136] In general, entropy encoding unit 820 can entropy encode syntax elements received from other functional components of video encoder 200. For example, entropy encoding unit 820 can entropy encode quantized transform coefficient blocks from quantization unit 808. As another example, entropy encoding unit 820 can entropy encode prediction syntax elements (e.g., motion information for inter-prediction or intra-mode information for intra-prediction) from mode select unit 802. Entropy encoding unit 820 can perform one or more entropy encoding operations on syntax elements, which are another example of video data, to generate entropy encoded data. For example, entropy encoding unit 820 can perform a context- adaptive variable length coding (CAVLC) operation, a CABAC operation, a variable to variable (V2V) length coding operation, a syntax-based context-adaptive binary arithmetic coding (SBAC) operation, a Probability Interval Partitioning Entropy (PIPE) coding operation, an Exponential-Golomb coding operation, or another type of entropy encoding operation. In some examples, entropy encoding unit 820 can operate in a bypass mode in which syntax elements are not entropy encoded.

[0137] Video encoder 200 can output a bitstream that includes the entropy encoded syntax elements needed to reconstruct blocks of a slice or picture. For example, entropy encoding unit 820 can output the bitstream.

[0138] The operations described above are described with respect to blocks. Such description should be understood to be operations with respect to luma coding blocks and / or chroma coding blocks. As described above, in some examples, the luma coding blocks and chroma coding blocks are luma and chroma components of a CU. In some examples, the luma coding blocks and chroma coding blocks are luma and chroma components of a PU.

[0139] In some examples, operations performed with respect to luma coding blocks need not be repeated with respect to chroma coding blocks. As one example, operations to identify motion vectors (MVs) and reference pictures for luma coding blocks need not be repeated to identify MVs and reference pictures for chroma blocks. Rather, the MVs for luma coding blocks can be scaled to determine MVs for chroma blocks, and the reference pictures can be the same. As another example, the intra-prediction process can be the same for luma coding blocks and chroma coding blocks.

[0140] Video encoder 200 represents an example of a device configured to encode video data, including a memory configured to store video data, and one or more processing units implemented in circuitry and configured to reconstruct blocks of video data and apply a CNN-based filter to the blocks, where the CNN-based filter uses a LeakyReLU activation function.

[0141] Figure 9is a block diagram illustrating an example video decoder 300 that can perform the techniques of this disclosure. Figure 9 is provided for explanatory purposes and does not limit the techniques broadly set forth in this disclosure. For explanatory purposes, this disclosure describes video decoder 300 in terms of the techniques of VVC (ITU-T H.266, under development) and HEVC (ITU-T H.265). However, the techniques of this disclosure can be performed by video coding devices configured to other video coding standards.

[0142] In Figure 9 In the example of FIG. 9, video decoder 300 includes coded picture buffer (CPB) memory 920, entropy decoding unit 902, prediction processing unit 904, inverse quantization unit 906, inverse transform processing unit 908, reconstruction unit 910, filter unit 912, and decoded picture buffer (DPB) 914. Any or all of CPB memory 920, entropy decoding unit 902, prediction processing unit 904, inverse quantization unit 906, inverse transform processing unit 908, reconstruction unit 910, filter unit 912, and DPB 914 can be implemented in one or more processors or in processing circuitry. For instance, the units of video decoder 300 can be implemented as one or more circuits or logic elements as part of hardware circuitry, or as part of a processor, ASIC, or FPGA. Moreover, video decoder 300 can include additional or alternative processors or processing circuitry to perform these and other functions.

[0143] Prediction processing unit 904 includes motion compensation unit 916 and intra-prediction unit 918. Prediction processing unit 904 can include additional units to perform prediction from other prediction modes. As examples, prediction processing unit 904 can include a palette unit, an intra-block copy unit (which can form a part of motion compensation unit 916), an affine unit, a linear model (LM) unit, etc. In other examples, video decoder 300 can include more, less, or different functional components.

[0144] CPB memory 920 can store video data, such as an encoded video bitstream, to be decoded by the components of video decoder 300. The video data stored in CPB memory 920 can be obtained, for example, from computer- readable medium 110 (FIG. 1). In some examples, CPB memory 920 can store data for multiple video bitstreams. Figure 1CPB memory 920 can include a CPB that stores encoded video data (e.g., syntax elements) from the encoded video bitstream. In addition, CPB memory 920 can store video data other than syntax elements of coded pictures, such as temporary data representing outputs from the various units of video decoder 300. DPB 914 generally stores decoded pictures that video decoder 300 can output and / or use as reference video data when decoding subsequent data or pictures of the encoded video bitstream. CPB memory 920 and DPB 914 can be formed by any of a variety of memory devices, such as DRAM, including SDRAM, MRAM, RRAM, or other types of memory devices. CPB memory 920 and DPB 914 can be provided by the same memory devices or separate memory devices. In various examples, CPB memory 920 can be on-chip with other components of video decoder 300, or off-chip relative to those components.

[0145] Additionally or alternatively, in some examples, video decoder 300 can retrieve coded video data from memory 120 Figure 1 ) as discussed above with respect to CPB memory 920. Also, when some or all functions of video decoder 300 are implemented in software to be executed by processing circuitry of video decoder 300, memory 120 can store the instructions to be executed by video decoder 300.

[0146] Explanation Figure 9 The various units shown in FIG. 3A are reproduced in FIG. 3B to help understand the operations performed by video encoder 300. The units can be implemented as fixed- function circuits, programmable circuits, or a combination thereof. Similar to Figure 8 , fixed-function circuits refer to circuits that provide particular functionality and are preset on the operations that can be performed. Programmable circuits refer to circuits that can be programmed to perform various tasks and provide flexible functionality among the operations that can be performed. For instance, the programmable circuits can execute software or firmware that cause the programmable circuits to operate in ways defined by instructions of the software or firmware. Fixed-function circuits can execute software instructions (e.g., to receive parameters or output parameters), but the types of operations that the fixed-function circuits perform are generally immutable. In some examples, one or more of the units can be distinct circuit blocks (fixed-function or programmable) and, in some examples, one or more of the units can be integrated circuits.

[0147] Video decoder 300 can include ALUs, EFUs, digital circuits, analog circuits, and / or programmable cores formed from programmable circuitry. In examples where the operations of video decoder 300 are performed by software executing on the programmable circuitry, on-chip or off-chip memory can store instructions (e.g., object code) of the software that video decoder 300 receives and executes.

[0148] Entropy decoding unit 902 can receive encoded video data from a CPB and entropy decode the video data to reconstruct syntax elements. Prediction processing unit 904, inverse quantization unit 906, inverse transform processing unit 908, reconstruction unit 910, and filter unit 912 can generate decoded video data based on the syntax elements extracted from the bitstream. In some examples, filter unit 912 can apply a CNN-based filter using a LeakyReLU activation function.

[0149] In general, video decoder 300 reconstructs a picture on a block-by-block basis. Video decoder 300 can perform reconstruction operations individually for each block (where the block that is currently being reconstructed (i.e., decoded) can be referred to as the “current block”).

[0150] Entropy decoding unit 902 can entropy decode syntax elements defining quantized transform coefficient blocks, as well as transform information such as a quantization parameter (QP) and / or an indication of a transform mode. Inverse quantization unit 906 can use the QP associated with a quantized transform coefficient block to determine a degree of quantization, and likewise a degree of inverse quantization for inverse quantization unit 906 to apply. Inverse quantization unit 906 may, for example, perform a bitshift operation to inverse quantize a quantized transform coefficient. Inverse quantization unit 906 can thereby form a transform coefficient block comprising transform coefficients.

[0151] After inverse quantization unit 906 forms a transform coefficient block, inverse transform processing unit 908 can apply one or more inverse transforms to the transform coefficient block to generate a residual block associated with the current block. For example, inverse transform processing unit 908 can apply an inverse DCT, an inverse integer transform, an inverse Karhunen-Loeve Transform (KLT), an inverse rotational transform, an inverse directional transform, or another inverse transform to the transform coefficient block.

[0152] Furthermore, the prediction processing unit 904 generates the prediction block from the prediction information syntax elements entropy decoded by the entropy decoding unit 902. For example, if the prediction information syntax elements indicate that the current block is inter predicted, the motion compensation unit 916 can generate the prediction block. In this case, the prediction information syntax elements can indicate a reference picture in the DPB 914 from which to retrieve a reference block, and a motion vector that identifies a position of the reference block in the reference picture relative to a position of the current block in the current picture. The motion compensation unit 916 can generally perform the inter prediction process in a manner substantially similar to that described with respect to the motion compensation unit 824( Figure 8 ) above.

[0153] As another example, if the prediction information syntax elements indicate that the current block is intra predicted, the intra prediction unit 918 can generate the prediction block according to an intra prediction mode indicated by the prediction information syntax elements. Again, the intra prediction unit 918 can generally perform the intra prediction process in a manner substantially similar to that described with respect to the intra prediction unit 826( Figure 8 ) above. The intra prediction unit 918 can retrieve data for neighboring samples of the current block from the DPB 914.

[0154] The reconstruction unit 910 can reconstruct the current block using the prediction block and the residual block. For example, the reconstruction unit 910 can add the samples of the residual block to the corresponding samples of the prediction block to reconstruct the current block.

[0155] The filter unit 912 can perform one or more filter operations on the reconstructed block. For example, the filter unit 912 can perform a deblocking operation to reduce blocking artifact pseudo-ghosts along edges of the reconstructed block. The operations of the filter unit 912 are not necessarily performed in all examples. In some examples, the filter unit 912 can apply a CNN-based filter that uses a LeakyReLU activation function, such as the CNN-based filter 700. Thus, in some examples, the reconstruction unit 910 can reconstruct a block of a picture, where the block can be a square or rectangular region of the picture and can or can not correspond to a coding block, a prediction block, and so on. The filter unit 912 can apply the CNN-based filter 700 to the block, where the CNN-based filter 700 uses a LeakyReLU activation function.

[0156] Video decoder 300 can store the reconstructed blocks in DPB 914. For instance, in examples where the operations of filter unit 912 are not performed, reconstruction unit 910 can store the reconstructed blocks to DPB 914. In examples where the operations of filter unit 912 are performed, filter unit 912 can store the filtered reconstructed blocks to DPB 914. As described above, DPB 914 can provide reference information, such as samples of a current picture for intra prediction and previously decoded pictures for subsequent motion compensation, to prediction processing unit 904. Moreover, video decoder 300 can output decoded pictures (e.g., decoded video) from DPB 914 for subsequent presentation on a display device, such as display device 118 of FIG. 1. Figure 1

[0157] In this way, video decoder 300 represents an example of a video decoding device that includes a memory configured to store video data and one or more processing units implemented in circuitry and configured to reconstruct blocks of the video data; and apply a CNN-based filter to the blocks, where the CNN-based filter uses a LeakyReLU activation function.

[0158] Figure 10 is a flowchart illustrating an example method of encoding a current block in accordance with the techniques of this disclosure. The current block can comprise a current CU. Although described with respect to video encoder 200 Figure 1 and Figure 8 ), it should be understood that other devices can be configured to perform methods similar to those of Figure 10 is a flowchart illustrating an example method of encoding a current block in accordance with the techniques of this disclosure. The current block can comprise a current CU. Although described with respect to video encoder 200

[0159] In this example, video encoder 200 initially predicts the current block (350). For instance, video encoder 200 can form a prediction block for the current block. Video encoder 200 can then calculate a residual block for the current block (352). To calculate the residual block, video encoder 200 can calculate a difference between an original, unencoded block of the current block and the prediction block. Video encoder 200 can then transform the residual block and quantize transform coefficients of the residual block (354). Next, video encoder 200 can scan the quantized transform coefficients of the residual block (356). Video encoder 200 can entropy encode the transform coefficients during or after the scan (358). For instance, video encoder 200 can encode the transform coefficients using CAVLC or CABAC. Video encoder 200 can then output the entropy encoded data for the block (360).

[0160] Figure 11 is a flowchart illustrating an example method of encoding a current block in accordance with the techniques of this disclosure. The current block can comprise a current CU. Although described with respect to video encoder 200 Figure 1 and​Figure 9 described, but it should be understood that other devices can be configured to perform similarly to the methods of Figure 11 described, but it should be understood that other devices can be configured to perform similarly to the methods of

[0161] Video decoder 300 can receive entropy encoded data for a current block, e.g., entropy encoded prediction information and entropy encoded data for transform coefficients of a residual block corresponding to the current block (370). Video decoder 300 can entropy decode the entropy encoded data to determine prediction information for the current block and reproduce transform coefficients of the residual block (372). Video decoder 300 can predict the current block (374), e.g., using intra- or inter-prediction modes as indicated by the prediction information for the current block, to calculate a prediction block for the current block. Video decoder 300 can then inverse scan the reproduced transform coefficients (376) to create a block of quantized transform coefficients. Video decoder 300 can then inverse quantize the transform coefficients and apply an inverse transform to the transform coefficients to produce a residual block (378). Video decoder 300 can finally decode the current block by combining the prediction block and the residual block (380). In some examples, video decoder 300 can further apply a CNN-based filter to the decoded block, where the CNN-based filter uses a LeakyReLU activation function.

[0162] Figure 12 is a flowchart illustrating example operations of a video coder (e.g., video encoder 200 or video decoder 300) in accordance with one or more techniques of this disclosure. In Figure 12 example, the video coder reconstructs a block of video data (1200). The block of video data can be a rectangular array of samples. The block can or can not correspond to a transform block, a prediction block, a coding block, a coding tree block, or a block type used in a coding process. For example, the video coder (e.g., reconstruction unit 814 Figure 8 ) or reconstruction unit 910 Figure 9 ) can reconstruct a picture including the block by adding residual data to a corresponding prediction block.

[0163] Further, the video coder applies a CNN-based filter (e.g., CNN-based filter 700) to the reconstructed block (1202). The CNN-based filter uses a LeakyReLU activation function. As described above, the LeakyReLU activation function is defined as:

[0164]

[0165] where y is an output value of a CNN convolution layer, and Alpha is a fixed parameter. In some examples, the CNN-based filter includes multiple convolution layers and the value of Alpha is the same for each convolution layer. In some examples, the CNN-based filter includes multiple convolution layers and the value of Alpha is different for two or more convolution layers.

[0166] In some examples where the video coder is the video decoder 300, the video decoder 300 can determine the value of Alpha based on one or more syntax elements signaled in a bitstream that includes an encoded representation of the video data. For example, the value of Alpha can be signaled in the bitstream. In other words, a syntax element can directly specify the value of Alpha. In another example, an index to Alpha in a predefined set is signaled in the bitstream. In this example, the video decoder 300 can determine the value of Alpha in the predefined set that corresponds to the index to Alpha.

[0167] In some examples where the video coder is the video encoder 200, the CNN-based filter includes multiple convolution layers (e.g., convolution layers 710) and the video encoder 200 can determine a single value of Alpha for use in a LeakyReLU activation function applied to output values of the convolution layers. The video encoder 200 can encode one or more syntax elements in a bitstream that includes an encoded representation of the video data that provide information needed by a video decoder to derive the value of Alpha. In some examples, the video encoder 200 signals an index to Alpha in a predefined set in the bitstream. In this example, the video encoder 200 can determine the value of Alpha in the predefined set that corresponds to the index to Alpha. In other examples, the video encoder 200 can determine multiple values of Alpha for use in a LeakyReLU activation function applied to output values of two or more different convolution layers of the multiple convolution layers (e.g., convolution layers 710). In such examples, the video encoder 200 can encode one or more syntax elements in a bitstream that includes an encoded representation of the video data that provide information needed by a video decoder to derive the values of Alpha.

[0168] The following is a non-limiting list of aspects that can be in accordance with one or more aspects of the present disclosure.

[0169] Aspect 1A. A method of coding video data, the method comprising: reconstructing a block of the video data; and applying a convolutional neural network (CNN)-based filter to the block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0170] Aspect 2A. The method of aspect 1A, wherein the LeakyReLU activation function is defined as:

[0171]

[0172] where y is an output value of a convolutional layer of the CNN, and Alpha is a fixed parameter.

[0173] Aspect 3A. The method of aspect 2A, wherein the CNN includes multiple convolutional layers and the value of Alpha is the same for each convolutional layer.

[0174] Aspect 4A. The method of aspect 2A, wherein the CNN includes multiple convolutional layers and the value of Alpha is different for two or more convolutional layers.

[0175] Aspect 5A. The method of any of aspects 2A-4A, further comprising: determining the value of Alpha based on one or more syntax elements signaled in a bitstream that includes an encoded representation of the video data.

[0176] Aspect 6A. The method of aspect 5A, wherein the value of Alpha is signaled in the bitstream.

[0177] Aspect 7A. The method of aspect 5A, wherein an index of Alpha within a predefined set is signaled in the bitstream.

[0178] Aspect 8A. The method of any of aspects 1A-7A, wherein coding comprises decoding.

[0179] Aspect 9A. The method of any of aspects 1A-7A, wherein coding comprises encoding.

[0180] Aspect 10A. A device for coding video data, the device comprising one or more means for performing the methods of aspects 1A-9A.

[0181] Aspect 11A. The device of aspect 10A, wherein the one or more means comprise one or more processors implemented in circuitry.

[0182] Aspect 12A. The device of any of aspects 10A and 11A, further comprising a memory for storing video data.

[0183] Aspect 13A. The device of any of aspects 10A-12A, further comprising a display configured to display decoded video data.

[0184] Aspect 14A. The device of any of aspects 10A-13A, wherein the device comprises one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.

[0185] Aspect 15A. The device of any of aspects 10A-14A, wherein the device comprises a video decoder.

[0186] Aspect 16A. The device of any of aspects 10A-15A, wherein the device comprises a video encoder.

[0187] Aspect 17A. A computer-readable storage medium having stored thereon instructions that, when executed, cause one or more processors to perform the method of any of aspects 1A-9A.

[0188] Aspect 1B. A method of encoding or decoding video data comprising: reconstructing a block of the video data; and applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0189] Aspect 2B. The method of aspect 1B, wherein the LeakyReLU activation function is defined as:

[0190]

[0191] where y is an output value of a CNN convolutional layer of the CNN-based filter, and Alpha is a fixed parameter.

[0192] Aspect 3B. The method of aspect 2B, wherein the CNN comprises a plurality of convolutional layers and the value of Alpha is the same for each convolutional layer.

[0193] Aspect 4B. The method of aspect 2B, wherein the CNN comprises a plurality of convolutional layers and the value of Alpha is different for two or more convolutional layers.

[0194] Aspect 5B. The method of aspect 2B, further comprising: determining the value of Alpha based on one or more syntax elements signaled in a bitstream that comprises an encoded representation of the video data.

[0195] Aspect 6B. The method of aspect 5B, wherein the value of Alpha is signaled in the bitstream.

[0196] Aspect 7B. The method of aspect 5B, wherein an index of Alpha in a predefined set is signaled in the bitstream, and wherein determining the value of Alpha comprises determining the value of Alpha that corresponds to the index of Alpha in the predefined set.

[0197] Aspect 8B. The method of aspect 2B, wherein the CNN comprises a plurality of convolution layers and the method further comprises: determining a single value of Alpha for the LeakyReLU activation function to be applied to output values of the convolution layers; and encoding one or more syntax elements in a bitstream that includes an encoded representation of the video data that provide information needed for a video decoder to derive the value of Alpha.

[0198] Aspect 9B. The method of aspect 8B, wherein the one or more syntax elements directly specify the value of Alpha.

[0199] Aspect 10B. The method of aspect 8B, wherein the one or more syntax elements specify an index of Alpha within a predefined set.

[0200] Aspect 11B. The method of aspect 2B, wherein the CNN comprises a plurality of convolution layers and the method further comprises: determining a plurality of values of Alpha for the LeakyReLU activation function to be applied to output values of two or more different convolution layers of the plurality of convolution layers; and encoding one or more syntax elements in a bitstream that includes an encoded representation of the video data that provide information needed for a video decoder to derive the values of Alpha.

[0201] Aspect 12B. The method of aspect 11B, wherein the one or more syntax elements directly specify the values of Alpha.

[0202] Aspect 13B. The method of aspect 11B, wherein the one or more syntax elements specify an index of Alpha within a predefined set.

[0203] Aspect 14B. The method of aspect IB, wherein the method comprises decoding the video data, decoding the video data comprising reconstructing a block of the video data and applying a filter based on the CNN to the reconstructed block.

[0204] Aspect 15B. The method of aspect IB, wherein the method comprises encoding the video data, and encoding the video data comprises reconstructing a block of the video data and applying the filter based on the CNN to the reconstructed block.

[0205] Aspect 16B. A device for encoding or decoding video data comprising: a memory for storing the video data; and one or more processors implemented in circuitry, the one or more processors configured to: reconstruct a block of the video data; and apply a filter based on a convolutional neural network (CNN) to the reconstructed block, wherein the filter based on the CNN uses a LeakyReLU activation function.

[0206] Aspect 17B: The device of aspect 16B, wherein the LeakyReLU activation function is defined as:

[0207]

[0208] where y is an output value of a CNN convolution layer of the CNN-based filter, and Alpha is a fixed parameter.

[0209] Aspect 18B: The device of aspect 17B, wherein the CNN comprises a plurality of convolution layers and the value of Alpha is the same for each of the convolution layers.

[0210] Aspect 19B: The device of aspect 17B, wherein the CNN comprises a plurality of convolution layers, and the value of Alpha is different for two or more convolution layers.

[0211] Aspect 20B: The device of aspect 17B, wherein the one or more processors are further configured to determine the value of Alpha based on one or more syntax elements signaled in a bitstream that comprises an encoded representation of the video data.

[0212] Aspect 21B: The device of aspect 20B, wherein the value of Alpha is signaled in the bitstream.

[0213] Aspect 22B: The device of aspect 20B, wherein an index of Alpha in a predefined set is signaled in the bitstream, and wherein determining the value of Alpha comprises determining a value of Alpha that corresponds to the index of Alpha in the predefined set.

[0214] Aspect 23B: The device of aspect 17B, wherein the CNN-based filter contains a plurality of convolution layers and the one or more processors are further configured to: determine a single value of Alpha for a LeakyReLU activation function to be applied to output values of the convolution layers; and encode, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information needed by a video decoder to derive the value of Alpha.

[0215] Aspect 24B: The device of aspect 17B, wherein the CNN comprises a plurality of convolution layers and the one or more processors are further configured to: determine a plurality of values of Alpha for a LeakyReLU activation function to be applied to output values of two or more different convolution layers of the plurality of convolution layers; and encode, in a bitstream that comprises an encoded representation of the video data, one or more syntax elements that provide information needed by a video decoder to derive the values of Alpha.

[0216] Aspect 25B: The device of aspect 16B, further comprising a display configured to display the decoded video data.

[0217] Aspect 26B: The device of aspect 16B, wherein the device comprises one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.

[0218] Aspect 27B: The device of aspect 16B, wherein the device comprises a video decoder.

[0219] Aspect 28B: The device of aspect 16B, wherein the device comprises a video encoder.

[0220] Aspect 29B: A device for coding video data comprising: means for reconstructing a block of video data; and means for applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0221] Aspect 30B: A computer-readable storage medium having stored thereon instructions that, when executed, cause one or more processors to: reconstruct a block of video data; and apply a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0222] Aspect 1C: A method of encoding or decoding video data comprising: reconstructing a block of the video data; and applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0223] Aspect 2C: The method of aspect 1C, wherein the LeakyReLU activation function is defined as:

[0224]

[0225] where y is an output value of a CNN convolution layer of the CNN-based filter, and Alpha is a fixed parameter.

[0226] Aspect 3C: The method of aspect 2C, wherein the CNN comprises a plurality of convolution layers and the value of Alpha is the same for each convolution layer.

[0227] Aspect 4C: The method of aspect 2C, wherein the CNN comprises a plurality of convolution layers and the value of Alpha is different for two or more convolution layers.

[0228] Aspect 5C: The method of any of aspects 2C-4C, further comprising: determining the value of Alpha based on one or more syntax elements signaled in a bitstream that includes an encoded representation of the video data.

[0229] Aspect 6C: The method of aspect 5C, wherein the value of Alpha is signaled in the bitstream.

[0230] Aspect 7C: The method of aspect 5C, wherein an index of Alpha in a predefined set is signaled in the bitstream, and wherein determining the value of Alpha comprises determining a value of Alpha that corresponds to the index of Alpha in the predefined set.

[0231] Aspect 8C: The method of any of aspects 2C-3C or 5C-7C, wherein the CNN comprises a plurality of convolution layers and the method further comprises: determining a single value of Alpha for the LeakyReLU activation function to be applied to output values of the convolution layers; and encoding in a bitstream that includes an encoded representation of the video data one or more syntax elements that provide information needed by a video decoder to derive the value of Alpha.

[0232] Aspect 9C. The method of aspect 8C, wherein the one or more syntax elements directly specify the value of Alpha.

[0233] Aspect 10C. The method of aspect 8C, wherein the one or more syntax elements specify an index of Alpha within a predefined set.

[0234] Aspect 11C: The method of any of aspects 2C or 4C-7C, wherein the CNN comprises a plurality of convolution layers, and the method further comprises: determining a plurality of values of Alpha for the LeakyReLU activation function to be applied to output values of two or more different convolution layers of the plurality of convolution layers; and encoding in a bitstream that includes an encoded representation of the video data one or more syntax elements that provide information needed by a video decoder to derive the values of Alpha.

[0235] Aspect 12C. The method of aspect 11C, wherein the one or more syntax elements directly specify the values of Alpha.

[0236] Aspect 13C. The method of aspect 11C, wherein the one or more syntax elements specify an index of Alpha within a predefined set.

[0237] Aspect 14C: The method of any of aspects 1C-13C, wherein the method comprises decoding the video data, decoding the video data comprising reconstructing a block of the video data and applying a filter based on the CNN to the reconstructed block.

[0238] Aspect 15C: The method of any of aspects 1B-14C, wherein the method comprises encoding the video data, and encoding the video data comprises reconstructing a block of the video data and applying the CNN-based filter to the reconstructed block.

[0239] Aspect 16C: A device for encoding or decoding video data comprising: a memory for storing the video data; and one or more processors implemented in circuitry, the one or more processors configured to: reconstruct a block of the video data; and apply a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0240] Aspect 17C: The device of aspect 16C, wherein the LeakyReLU activation function is defined as:

[0241]

[0242] where y is an output value of a CNN convolution layer of the CNN-based filter, and Alpha is a fixed parameter.

[0243] Aspect 18C: The device of aspect 17C, wherein the CNN comprises a plurality of convolution layers and the value of Alpha is the same for each of the convolution layers.

[0244] Aspect 19C: The device of aspect 17C, wherein the CNN comprises a plurality of convolution layers, and the value of Alpha is different for two or more of the convolution layers.

[0245] Aspect 20C: The device of aspects 17C-19C, wherein the one or more processors are further configured to determine the value of Alpha based on one or more syntax elements signaled in a bitstream that comprises an encoded representation of the video data.

[0246] Aspect 21C: The device of aspect 20C, wherein the value of Alpha is signaled in the bitstream.

[0247] Aspect 22C: The device of aspect 20C, wherein an index of Alpha in a predefined set is signaled in the bitstream, and wherein determining the value of Alpha comprises determining the value of Alpha that corresponds to the index of Alpha in the predefined set.

[0248] Aspect 23C: The device of any of aspects 17C-18C or 20C-22C, wherein the CNN- based filter includes a plurality of convolution layers and the one or more processors are further configured to: determine a single value of Alpha for a LeakyReLU activation function to be applied to output values of the convolution layers; and encode, in a bitstream that includes an encoded representation of the video data, one or more syntax elements that provide information needed by a video decoder to derive the value of Alpha.

[0249] Aspect 24C: The device of any of aspects 17C or 18C-22C, wherein the CNN includes a plurality of convolution layers and the one or more processors are further configured to: determine a plurality of values of Alpha for a LeakyReLU activation function to be applied to output values of two or more different convolution layers of the plurality of convolution layers; and encode, in a bitstream that includes an encoded representation of the video data, one or more syntax elements that provide information needed by a video decoder to derive the values of Alpha.

[0250] Aspect 25C: The device of any of aspects 16C-24C, further comprising: a display configured to display decoded video data.

[0251] Aspect 26C: The device of any of aspects 16C-25C, wherein the device comprises one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.

[0252] Aspect 27C: The device of any of aspects 16C-26C, wherein the device comprises a video decoder.

[0253] Aspect 28C: The device of any of aspects 16C-27C, wherein the device comprises a video encoder.

[0254] Aspect 29C: A device for coding video data comprising: means for reconstructing a block of the video data; and means for applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0255] Aspect 29C: A device for coding video data comprising: means for reconstructing a block of the video data; and means for applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0256] Aspect 30C: A computer-readable storage medium having stored thereon instructions that, when executed, cause one or more processors to: reconstruct a block of video data; and apply a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN-based filter uses a LeakyReLU activation function.

[0257] Aspect 30C: The computer-readable storage medium of Aspect 26C, comprising means for performing the method of any of claims 1C-15C.

[0258] It is recognized that, in light of the examples, certain acts or events of any of the techniques described herein can be performed in a different sequence, can be added, modified or omitted (e.g., not all described acts or events are necessary for a practical practice of the techniques), and some acts or events can be performed concurrently, such as through multi-threaded processing, interrupt processing, or multiple processors, rather than sequentially.

[0259] In one or more examples, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media can include computer-readable storage media, which corresponds to a tangible medium such as data storage media, or communication media including any medium that facilitates transfer of a computer program from one place to another, e.g., according to a communication protocol. In this manner, computer- readable media generally can correspond to (1) tangible computer-readable storage media which is non-transitory or (2) a communication medium such as a signal or carrier wave. Data storage media can be any available media that can be accessed by one or more computers or one or more processors to retrieve instructions, code and / or data structures for implementation of the techniques described in this disclosure. A computer program product can include a computer-readable medium.

[0260] By way of example, and not limitation, such computer-readable storage media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage, or other magnetic storage devices, flash memory, or any other storage medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. It should be understood, however, that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transitory media, but are instead directed to non-transitory, tangible storage media. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0261] 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 circuitry. Accordingly, the terms "processor" and "processing circuitry," as used herein can refer to any of the foregoing structures or any other structure suitable for implementation of the techniques described herein. In addition, in some aspects, the functionality described herein can be provided within dedicated hardware and / or software modules configured for encoding and decoding, or incorporated in a combined codec. Also, the techniques could be fully implemented in one or more circuits or logic elements.

[0262] The techniques of this disclosure can be implemented in a wide variety of devices or apparatuses, including a wireless handset, an integrated circuit (IC) or a set of ICs (e.g., a chip set). Various components, modules, or units are described herein to emphasize functionality that can be provided by devices configured to perform the techniques described herein. But, the various components, modules, or units can not necessarily be implemented in hardware separate from other components, modules, or units. Rather, such units can be combined in a codec hardware unit or provided by a set of inter-operating hardware units, including one or more processors as described above, along with appropriate software and / or firmware.

[0263] Various examples have been described. These and other examples are within the scope of the following claims.

Claims

1. A method of decoding video data, the method comprising: obtaining one or more syntax elements from a bitstream that comprises an encoded representation of the video data; reconstructing a block of the video data; applying a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN comprises a plurality of convolutional layers, the CNN-based filter applies a Leaky Rectified Linear Unit (LeakyReLU) activation function to output values of the convolutional layers, and for each convolutional layer of the plurality of convolutional layers, the LeakyReLU activation function applied to the output values of the convolutional layer is defined as: where y is an output value of a convolutional layer and Alpha is a fixed parameter, and for at least one convolutional layer of the plurality of convolutional layers, a value of Alpha used in the LeakyReLU activation function applied to output values of the at least one convolutional layer is determined based on the one or more syntax elements.

2. The method of claim 1, wherein, the same value of Alpha is used in the LeakyReLU activation function applied to output values of each of the convolutional layers.

3. The method of claim 1, wherein, values of Alpha used in the LeakyReLU activation function applied to output values of two or more of the convolutional layers are different.

4. The method of claim 1, wherein, the one or more syntax elements directly specify the value of Alpha.

5. The method of claim 1, wherein obtaining the one or more syntax elements comprises obtaining an index of Alpha within a predefined set signaled in the bitstream, and wherein determining the value of Alpha comprises determining the value of Alpha in the predefined set corresponding to the index of Alpha.

6. The method of claim 1, wherein, the method further comprises: determining a plurality of values of Alpha based on the one or more syntax elements; and applying the CNN-based filter to the reconstructed block comprises applying the LeakyReLU activation function to output values of two or more different convolutional layers of the plurality of convolutional layers using different values of Alpha in the plurality of values of Alpha.

7. The method of claim 6, wherein, the one or more syntax elements comprise a plurality of syntax elements that directly specify the plurality of values of Alpha.

8. The method of claim 6, wherein, the one or more syntax elements comprise a plurality of syntax elements that specify a plurality of indices of Alpha within a predefined set.

9. A device of decoding video data, the device comprising: a memory for storing the video data; and one or more processors implemented in circuitry, the one or more processors configured to: obtain one or more syntax elements from a bitstream that comprises an encoded representation of the video data; reconstruct a block of the video data; apply a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN comprises a plurality of convolutional layers, the CNN-based filter applies a Leaky Rectified Linear Unit (LeakyReLU) activation function to output values of the convolutional layers, and for each convolutional layer of the plurality of convolutional layers, the LeakyReLU activation function applied to the output values of the convolutional layer is defined as: where y is an output value of a convolutional layer and Alpha is a fixed parameter, and for at least one convolutional layer of the plurality of convolutional layers, a value of Alpha used in the LeakyReLU activation function applied to output values of the at least one convolutional layer is determined based on the one or more syntax elements. for each of the plurality of convolutional layers, the LeakyReLU activation function applied to the output values of the convolutional layer is defined as: where y is an output value of a convolutional layer and Alpha is a fixed parameter, and for at least one of the plurality of convolutional layers, a value of Alpha used in the LeakyReLU activation function applied to output values of the at least one convolutional layer is determined based on the one or more syntax elements.

10. The apparatus of claim 9, wherein, a same value of Alpha is used in the LeakyReLU activation function applied to output values of each of the convolutional layers.

11. The apparatus of claim 9, wherein, values of Alpha used in the LeakyReLU activation function applied to output values of two or more of the convolutional layers are different.

12. The apparatus of claim 9, wherein, the one or more syntax elements directly specify the value of Alpha.

13. The device of claim 9, wherein as part of obtaining the one or more syntax elements, the one or more processors are configured to obtain an index of Alpha within a predefined set signaled in the bitstream, and wherein determining the value of Alpha includes determining the value of Alpha in the predefined set corresponding to the index of Alpha.

14. The apparatus of claim 9, wherein, the one or more processors are further configured to: determine a plurality of values of Alpha based on the one or more syntax elements; and as part of applying the CNN-based filter to the reconstructed block, the one or more processors are further configured to apply the LeakyReLU activation function to output values of two or more different convolutional layers of the plurality of convolutional layers using different values of Alpha from the plurality of values of Alpha.

15. The apparatus of claim 14, wherein, the one or more syntax elements include a plurality of syntax elements that directly specify the plurality of values of Alpha.

16. The apparatus of claim 14, wherein, the one or more syntax elements include a plurality of syntax elements that specify a plurality of indices of Alpha within a predefined set.

17. The apparatus of claim 9, further comprising: a display configured to display decoded video data.

18. The apparatus of claim 9, wherein, the device includes one or more of a camera, a computer, a mobile device, a broadcast receiver device, or a set-top box.

19. A non-transitory computer-readable storage medium having instructions stored thereon that, when executed, cause one or more processors to: obtain one or more syntax elements from a bitstream that includes an encoded representation of video data; reconstruct a block of video data; apply a convolutional neural network (CNN)-based filter to the reconstructed block, wherein the CNN includes a plurality of convolutional layers, the CNN-based filter applies a leaky rectified linear unit (LeakyReLU) activation function to output values of the convolutional layers, and for each of the plurality of convolutional layers, the LeakyReLU activation function applied to the output values of the convolutional layer is defined as: where y is an output value of a convolutional layer and Alpha is a fixed parameter, and for at least one of the plurality of convolutional layers, a value of Alpha used in the LeakyReLU activation function applied to output values of the at least one convolutional layer is determined based on the one or more syntax elements. For at least one of the plurality of convolutional layers, determining a value of Alpha used in the LeakyReLU activation function applied to output values of the at least one convolutional layer based on the one or more syntax elements.

20. The non-transitory computer-readable storage medium of claim 19, wherein, using a same value of Alpha in the LeakyReLU activation function applied to output values of each of the convolutional layers.

Images