Method for generative facial video compression using a facial feature transformer

Abstract

A method of video decoding includes receiving a bitstream and decoding one or more pictures using coded information of the bitstream. The decoding includes decoding, from the bitstream, a first facial feature of a first type of a face picture; converting the first facial feature to a second facial feature of a second type; and reconstructing the face picture based on the second facial feature.

Description

[0001] Cross-reference to related applications This disclosure claims priority to U.S. Provisional Application 63 / 737,795, filed December 22, 2024, which is incorporated herein by reference in its entirety. This disclosure also claims priority to U.S. Patent Application 19 / 379,516, filed November 4, 2025. Technical Field

[0002] This disclosure relates generally to video processing, and more specifically, to a facial feature converter for generative facial video compression and a method for generative facial video compression using said facial feature converter. Background Technology

[0003] Video consists of a set of still images (or "frames") that capture visual information. To reduce storage memory and transmission bandwidth, video can be compressed before storage or transmission and decompressed before display. This compression process is typically called encoding, while the decompression process is typically called decoding. There are many video coding formats that use standardized video coding techniques, the most common being based on prediction, transform, quantization, entropy coding, and loop filtering. Standardization organizations have developed video coding standards that specify particular video coding formats, such as the High Efficiency Video Coding (HEVC / H.265) standard, the Universal Video Coding (VVC / H.266) standard, and the AVS standard. As more advanced video coding technologies are incorporated into video standards, the coding efficiency of new video coding standards is also increasing. Summary of the Invention

[0004] Embodiments of this disclosure provide a video decoding method. The method includes: receiving a bitstream; and decoding one or more images using encoded information from the bitstream. The decoding includes decoding a first facial feature of a first type from the bitstream; converting the first facial feature into a second facial feature of a second type; and reconstructing the facial image based on the second facial feature.

[0005] Embodiments of this disclosure provide a facial feature converter. The facial feature converter includes a plurality of encoders, each encoder configured to encode a first facial feature of a first type into a uniformly embedded feature; and a plurality of decoders coupled to one or more of the encoders, each decoder configured to decode the uniformly embedded feature into a second facial feature of a second type, wherein the first type is different from the second type.

[0006] Embodiments of this disclosure provide a non-transitory computer-readable medium having a set of instructions stored thereon, the set of instructions being executable by one or more processors of a device to cause the device to perform the following operations: decoding a first facial feature of a first type from a bitstream of a facial image; converting the first facial feature into a second facial feature of a second type; and reconstructing the facial image based on the second facial feature. Attached Figure Description

[0007] Embodiments and aspects of this disclosure are illustrated in the following detailed description and accompanying drawings. The various features shown in the figures are not drawn to scale.

[0008] Figure 1 This is a schematic diagram illustrating an exemplary system for encoding image data according to some embodiments of the present disclosure.

[0009] Figure 2 This is a schematic diagram illustrating the architecture of a block-based video compression framework according to some embodiments of the present disclosure.

[0010] Figure 3 This is a schematic diagram illustrating the structure of an exemplary video sequence according to some embodiments of the present disclosure.

[0011] Figure 4A This is a schematic diagram illustrating an exemplary block-based encoding process according to some embodiments of the present disclosure.

[0012] Figure 4B This is a schematic diagram illustrating another exemplary block-based encoding process according to some embodiments of the present disclosure.

[0013] Figure 5A This is a schematic diagram illustrating an exemplary block-based decoding process according to some embodiments of the present disclosure.

[0014] Figure 5B This is a schematic diagram illustrating another exemplary block-based decoding process according to some embodiments of the present disclosure.

[0015] Figure 6 This is a schematic diagram illustrating an exemplary architecture of an end-to-end deep learning-based video compression framework according to some embodiments of the present disclosure.

[0016] Figure 7 This is a schematic diagram illustrating an exemplary architecture of a deep learning-based video generative compression framework according to some embodiments of the present disclosure.

[0017] Figure 8 This is a schematic diagram illustrating an exemplary encoder-decoder encoding framework with a compact feature size of 1×4×4 for speaking facial video according to some embodiments of the present disclosure.

[0018] Figure 9A This is a schematic diagram illustrating a general flowchart of a generative facial video compression (GFVC) system according to some embodiments of the present disclosure.

[0019] Figure 9B This is a schematic diagram illustrating facial representations in a generative facial video compression system according to some embodiments of the present disclosure.

[0020] Figure 10 An exemplary framework for generative facial video compression using a dense motion optical flow converter according to some embodiments of this disclosure is shown.

[0021] Figure 11 This is a schematic diagram illustrating an exemplary method for generative facial video compression using a facial feature converter according to some embodiments of the present disclosure.

[0022] Figure 12 This is a schematic diagram illustrating an exemplary facial feature converter 1200 according to some embodiments of the present disclosure.

[0023] Figure 13 This is a schematic diagram illustrating an exemplary method for converting facial features using a facial feature converter according to some embodiments of the present disclosure.

[0024] Figure 14 This is a schematic diagram illustrating an exemplary method for training a facial feature converter according to some embodiments of the present disclosure.

[0025] Figure 15A This is a schematic diagram illustrating an exemplary direct conversion scheme according to some embodiments of the present disclosure.

[0026] Figure 15B This is a schematic diagram illustrating an exemplary difference conversion scheme according to some embodiments of the present disclosure.

[0027] Figure 16 This is a schematic diagram illustrating another exemplary method for converting facial features using a facial feature converter according to some embodiments of the present disclosure.

[0028] Figure 17 This is a block diagram of an exemplary apparatus for encoding image data according to some embodiments of the present disclosure. Detailed Implementation

[0029] Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings, in which, unless otherwise stated, the same numerals in different figures represent the same or similar elements. The embodiments set forth in the following description of the exemplary embodiments do not represent all embodiments consistent with the invention. Rather, they are merely examples of apparatuses and methods consistent with the relevant aspects of the invention described in the appended claims. Specific aspects of this disclosure are described in more detail below. In the event of any conflict with terms and / or definitions incorporated by reference, the terms and definitions provided herein shall prevail.

[0030] The Joint Video Experts Group (JVET) of the ITU-T Video Coding Experts Group (ITU-T VCEG) and the ISO / IEC Moving Picture Experts Group (ISO / IEC MPEG) is currently developing a universal video coding standard (VVC / H.266). The VVC standard aims to double the compression efficiency of its predecessor, the High Efficiency Video Coding (HEVC / H.265) standard. In other words, VVC aims to achieve the same subjective quality as HEVC / H.265 using half the bandwidth.

[0031] To achieve the same subjective quality as HEVC / H.265 using half the bandwidth, JVET has been developing techniques other than HEVC using the Joint Exploratory Model (JEM) reference software. With coding techniques incorporated into JEM, JEM achieves significantly higher coding performance than HEVC.

[0032] The VVC standard has recently been finalized and continues to incorporate more coding technologies to provide better compression performance. VVC adopts the hybrid video coding system used in modern video compression standards such as HEVC, H.264 / AVC, MPEG2, and H.263.

[0033] Video is a set of still images (or "frames") arranged chronologically to store visual information. Video capture devices (e.g., cameras) can be used to capture and store those images chronologically, and video playback devices (e.g., televisions, computers, smartphones, tablets, video players, or any end-user terminal with a display) can be used to display these images chronologically. Furthermore, in some applications, video capture devices can transmit the captured video in real time to the video playback device (e.g., a computer with a display), for example, for video observation, conferencing, or live streaming.

[0034] To reduce the storage space and transmission bandwidth required for these applications, the video can be compressed before storage or transmission and decompressed before display. This compression and decompression can be implemented by software executed by a processor (e.g., a processor in a general-purpose computer) or by dedicated hardware. The module used for compression is typically called an "encoder," while the module used for decompression is typically called a "decoder." The encoder and decoder can be collectively referred to as a "codec." The encoder and decoder can be implemented as any of a variety of suitable hardware, software, or combinations thereof. For example, the hardware implementation of the encoder and decoder can include circuit systems such as one or more microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), discrete logic, or any combination thereof. The software implementation of the encoder and decoder can include program code, computer-executable instructions, firmware, or any suitable computer-implemented algorithm or process embedded in a computer-readable medium. Video compression and decompression can be implemented using various algorithms or standards, such as MPEG-1, MPEG-2, MPEG-4, H.26x series, etc. In some applications, the codec can decompress the video according to a first encoding standard and recompress the decompressed video using a second encoding standard. In this case, the codec can be called a "transcoder".

[0035] Video coding identifies and retains useful information for image reconstruction while ignoring information unimportant to the reconstruction. If the ignored, unimportant information cannot be fully reconstructed, such a coding process can be called "lossy." Otherwise, it can be called "lossless." Most coding processes are lossy, a trade-off between reducing required storage space and transmission bandwidth.

[0036] Useful information about an image being encoded (referred to as the "current image") includes changes relative to a reference image (e.g., a previously encoded and reconstructed image). These changes can include variations in pixel position, brightness, or color, with positional changes being of primary interest. The positional changes of a set of pixels representing an object can reflect the movement of that object between the reference image and the current image.

[0037] An image encoded without referencing another image (i.e., whose reference image is itself) is called an "I-image". If some or all blocks in an image (e.g., these blocks typically refer to different parts of a video image) are predicted using intra-frame prediction or inter-frame prediction with a reference image (e.g., one-way prediction), the image is called a "P-image". If at least one block in an image is predicted using two reference images (e.g., two-way prediction), the image is called a "B-image".

[0038] Figure 1 This is a block diagram illustrating a system 100 for encoding image data according to embodiments of the present disclosure. Image data may include images (also called “pictures” or “frames”), multiple images, or video. Images are still images. Multiple images may be spatially or temporally related or unrelated. Video consists of a set of images arranged in chronological order.

[0039] like Figure 1 As shown, system 100 includes a source device 120 that provides encoded video data for subsequent decoding by a target device 140. Consistent with the disclosed embodiments, both source device 120 and target device 140 can include any of a variety of devices, including desktop computers, laptops (e.g., notebook computers), servers, tablets, set-top boxes, mobile phones, vehicles, cameras, image sensors, robots, televisions, wearable devices (e.g., smartwatches or wearable cameras), display devices, digital media players, video game consoles, video streaming devices, etc. Source device 120 and target device 140 can be configured for wireless or wired communication.

[0040] refer to Figure 1 The source device 120 may include an image / video preprocessor 122, an image / video encoder 124, and an output interface 126. The target device 140 may include an input interface 142, an image / video decoder 144, and a machine vision application 146. The image / video encoder 124 encodes the input bitstream and outputs the encoded bitstream 162 via the output interface 126. The encoded bitstream 162 is transmitted through a communication medium 160 and received by the input interface 142. The image / video decoder 144 then decodes the encoded bitstream 162 to generate decoded data.

[0041] More specifically, the source device 120 may also include various devices (not shown) for providing source image data to be processed by the image / video encoder 124. Devices for providing source image data may include image / video acquisition devices, such as cameras, image / video archives or storage devices containing previously acquired images / videos, or image / video feed interfaces for receiving images / videos from image / video content providers.

[0042] Image / video encoder 124 and image / video decoder 144 can each be implemented as any of a variety of suitable encoder or decoder circuit systems, 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 combination thereof. When encoding or decoding is partially implemented in software, image / video encoder 124 or image / video decoder 144 may store instructions for software in a suitable, non-transitory computer-readable medium and execute the instructions in hardware using one or more processors to perform techniques consistent with this disclosure. Image / video encoder 124 or image / video decoder 144 may be included in one or more encoders or decoders, which may be integrated as part of a composite encoder / decoder (CODEC) in the respective device.

[0043] The image / video encoder 124 and image / video decoder 144 can operate according to any video coding standard, such as Advanced Video Coding (AVC), High Efficiency Video Coding (HEVC), or Universal Video Coding (VVC). AOMedia Video 1 (AV1), Joint Picture Experts Group (JPEG), Moving Picture Experts Group (MPEG), etc. Alternatively, the image / video encoder 124 and image / video decoder 144 can be custom devices that do not conform to existing standards. Although Figure 1 As not shown, but in some embodiments, both the image / video encoder 124 and the image / video decoder 144 may be integrated with the audio encoder and decoder, and may include suitable multiplexer-demultiplexer (MUX-DEMUX) units, or other hardware and software, to process the encoding of audio and video, including encoding both in a common data stream or encoding them separately in separate data streams.

[0044] Output interface 126 may include any type of medium or device capable of transmitting the encoded bit stream 162 from source device 120 to target device 140. For example, output interface 126 may include a transmitter or transceiver configured to transmit the encoded bit stream 162 directly from source device 120 to target device 140 in real time. The encoded bit stream 162 may be modulated according to communication standards such as wireless communication protocols and transmitted to target device 140.

[0045] Communication medium 160 may include transient media, such as wireless broadcasting or wired network transmission. For example, communication medium 160 may include radio frequency (RF) spectrum or one or more physical transmission lines (e.g., cables). Communication medium 160 may form part of a packet-based network, such as a local area network, a wide area network, or a global network such as the Internet. In some embodiments, communication medium 160 may include a router, a switch, a base station, or any other device that may be used to facilitate communication from source device 120 to target device 140. For example, a network server (not shown) may receive encoded bitstream 162 from source device 120 and provide encoded bitstream 162 to target device 140, for example, via network transmission.

[0046] The communication medium 160 may also be in the form of a storage medium (e.g., a non-transitory storage medium), such as a hard disk, flash drive, optical disk, digital video optical disk, Blu-ray disc, volatile or non-volatile memory, or any other suitable digital storage medium for storing encoded image data. In some embodiments, a computing device of a media production facility (e.g., an optical disk imprinting facility) may receive encoded image data from source device 120 and produce an optical disk containing the encoded video data.

[0047] Input interface 142 may include any type of medium or device capable of receiving information from communication medium 160. The received information includes an encoded bitstream 162. For example, input interface 142 may include a receiver or transceiver configured to receive the encoded bitstream 162 in real time.

[0048] System 100 can be configured to perform video encoding and decoding based on block-based video compression technology, deep learning-based video compression technology, speaking face video compression technology, etc.

[0049] The block-based video compression technique utilizes a block-based hybrid video coding framework to leverage spatial, temporal, and entropy redundancy in the video. This hybrid video coding framework includes motion compensation (e.g., intra / inter-frame prediction), transform (e.g., discrete cosine transform), quantization, and entropy coding. The block-based video compression technique can conform to various image / video coding standards, such as JPEG, JPEG2000, H.264 / MPEG4 Part 10, Audio Video Coding Standard (AVS), H.265 / HEVC, and Universal Video Coding (VVC).

[0050] Figure 2This is a schematic diagram illustrating a block-based video compression framework 200 according to some embodiments of the present disclosure. The block-based video compression framework 200 may include an encoder configured to generate a bitstream based on input video frames, and a decoder configured to reconstruct video frames based on the bitstream. For simplicity, Figure 2 Only the encoding end of the block-based video compression framework 200 is shown. It can be envisioned that the decoding end of the block-based video compression framework 200 is the reverse operation of the encoding end.

[0051] Specifically, such as Figure 2 As shown, the input frame x of the encoding end t A set of blocks, such as square regions, are divided into blocks of the same size (e.g., 8×8). The block-based video compression framework 200 includes the following steps.

[0052] The block-based video compression framework 200 performs motion estimation using a block-based motion estimation module 201. The motion estimation module 201 can estimate the motion of the current frame. Compared with the previously reconstructed frame The motion between them. Obtain the corresponding motion vector for each block. .

[0053] The block-based video compression framework 200 performs motion compensation using a motion compensation module 202. This is based on the motion vectors determined by the motion estimation module 201. The predicted frame is obtained by copying the corresponding pixels from the previously reconstructed frame to the current frame. Then, the original frame is obtained. With the predicted frame residuals between ,for = – .

[0054] The block-based video compression framework 200 performs transformation and quantization using a transform module 203 and a Q module 204, respectively. The residual is then processed by the Q module 204. Quantified as Before quantization, the transform module 203 uses a linear transform (e.g., DCT) to obtain better compression performance.

[0055] The block-based video compression framework 200 performs an inverse transform using an inverse transform module 205. The quantization result... It was used for inverse transformation to obtain the reconstructed residual. .

[0056] The block-based video compression framework 200 performs entropy coding using an entropy coding module 206. The motion vector is then encoded using the entropy coding method. and the quantification results All are encoded into one or more bit streams, and the one or more bit streams are sent to the decoder.

[0057] The block-based video compression framework 200 performs frame reconstruction using the reconstruction module 207. and The reconstructed frame is obtained by adding them together. ,Right now, = + The reconstructed frame was the first One frame is used for motion estimation.

[0058] The bitstream generated by the entropy coding module 206 can be decoded at the decoding end ( Figure 2 Decoding (not shown). Motion compensation, inverse quantization, and frame reconstruction can be performed to obtain the reconstructed frame. .

[0059] Combination Figure 3 , 4A Sections 4B, 5A, and 5B further describe the details of the block-based video compression framework 200. Specifically, Figure 3 The structure of an example video sequence 300 according to some embodiments of the present disclosure is shown. The video sequence 300 may be live video or video that has already been captured and archived. The video sequence 300 may be real video, computer-generated video (e.g., computer game video), or a combination thereof (e.g., real video with augmented reality effects). The video sequence 300 may originate from a video capture device (e.g., a camera), a video archive containing previously captured video (e.g., a video file stored on a storage device), or a video feed interface (e.g., a video broadcast transceiver) that receives video from a video content provider.

[0060] like Figure 3 As shown, video sequence 300 may include a series of images arranged chronologically along a timeline, including images 302, 304, 306, and 308. Images 302-306 are consecutive, and there are more images between images 306 and 308. Figure 3In this diagram, image 302 is an I-image, and its reference image is image 302 itself. Image 304 is a P-image, and its reference image is image 302, as indicated by the arrow. Image 306 is a B-image, and its reference images are images 304 and 308, as indicated by the arrow. In some embodiments, the reference image of an image (e.g., image 304) may not immediately precede or follow that image. For example, the reference image of image 304 may be an image preceding image 302. It should be noted that the reference images of images 302-306 are merely examples, and this disclosure does not limit the embodiments of the reference images to specific cases. Figure 3 The example shown.

[0061] Typically, due to the computational complexity of this task, video codecs do not encode or decode the entire image at once. Instead, they segment the image into multiple basic segments and encode or decode the image segment by segment. Such basic segments are referred to herein as basic processing units (“BPUs”). For example, Figure 3 Structure 310 illustrates an example structure of an image (e.g., any one of images 302-308) from video sequence 300. In structure 310, the image is divided into 4×4 basic processing units, the boundaries of which are shown as dashed lines. In some embodiments, the basic processing units may be referred to as “macroblocks” in some video coding standards (e.g., MPEG series, H.261, H.263, or H.264 / AVC), or as “coding tree units” (“CTUs”) in some other video coding standards (e.g., H.265 / HEVC, H.266 / VVC, or AVS). The basic processing units in the image may have variable sizes, such as 128×128, 64×64, 32×32, 16×16, 4×8, 16×32, or any arbitrary shape and size of pixels. The size and shape of the basic processing units for the image can be selected based on a balance between coding efficiency and the level of detail to be maintained in the basic processing units.

[0062] The basic processing unit can be a logical unit that may include a set of different types of video data stored in computer memory (e.g., in a video frame buffer). For example, a basic processing unit for a color image may include a luminance component (Y) representing non-color information, one or more chrominance components (e.g., Cb and Cr) representing color information, and associated syntax elements, wherein the size of the luminance and chrominance components may be the same as that of the basic processing unit. In some video coding standards (e.g., H.265 / HEVC, H.266 / VVC, or AVS), the luminance and chrominance components may be referred to as "code tree blocks" ("CTBs"). Any operation performed on the basic processing unit can be repeated on the luminance and chrominance components of the basic processing unit separately.

[0063] Video encoding involves multiple operational stages, examples of which are as follows: Figures 4A-4B and Figures 5A-5B As shown. For each stage, the size of the basic processing unit may still be too large to process, and therefore can be further divided into segments referred to herein as "basic processing subunits". In some embodiments, the basic processing subunit may be referred to as a "block" in some video coding standards (e.g., MPEG series, H.261, H.263, H.264 / AVC, or AVS), or as a "coding unit" ("CU") in some other video coding standards (e.g., H.265 / HEVC, H.266 / VVC, or AVS). The basic processing subunit may have the same or smaller size as the basic processing unit. Similar to the basic processing unit, the basic processing subunit is also a logical unit that may include a set of different types of video data (e.g., Y, Cb, Cr, and associated syntax elements) stored in computer memory (e.g., stored in a video frame buffer). Any operation performed on the basic processing subunit may be repeatedly performed on the luma and chroma components of the basic processing subunit separately. It should be noted that this division can be performed to deeper levels as needed for processing. It should also be noted that different schemes can be used to divide the basic processing units at different stages.

[0064] For example, in the pattern decision-making phase (examples of which are in...) Figure 4B As shown in the diagram, the encoder can decide which prediction mode (e.g., intra-image prediction or inter-image prediction) to use for the basic processing unit, which may be too large to make such a decision. The encoder can divide the basic processing unit into multiple basic processing subunits (e.g., CUs in H.265 / HEVC, H.266 / VVC, or AVS) and determine the prediction type for each individual basic processing subunit.

[0065] For another example, in the prediction phase (the example is in...) Figures 4A-4B As shown in the diagram, the encoder can perform prediction operations at the level of a basic processing subunit (e.g., a CU). However, in some cases, the basic processing subunit may still be too large to process. The encoder can further divide the basic processing subunit into smaller segments (e.g., referred to as "prediction blocks" or "PBs" in H.265 / HEVC, H.266 / VVC, or AVS), at which the prediction operations can be performed.

[0066] For another example, in the transformation phase (the example of which is in...) Figures 4A-4BAs shown, the encoder can perform transformation operations on residual basic processing subunits (e.g., CUs). However, in some cases, the basic processing subunits may still be too large to process. The encoder can further divide the basic processing subunits into smaller segments (e.g., referred to as "transform blocks" or "TBs" in H.265 / HEVC, H.266 / VVC, or AVS), at which the transformation operation can be performed. It should be noted that the partitioning scheme of the same basic processing subunit can differ between the prediction and transformation phases. For example, in H.265 / HEVC, H.266 / VVC, or AVS, the prediction blocks and transform blocks of the same CU can have different sizes and numbers.

[0067] exist Figure 3 In structure 310, the basic processing unit 312 is further divided into 3×3 basic processing sub-units, the boundaries of which are shown as dashed lines. Different schemes can be used to divide different basic processing units of the same image into basic processing sub-units.

[0068] In some implementations, to provide parallel processing and fault tolerance for video encoding and decoding, an image can be divided into multiple regions for processing, such that for a particular region of the image, the encoding or decoding process can be independent of information from any other region of the image. In other words, each region of the image can be processed independently. By doing so, the codec can process different regions of an image in parallel, thereby improving encoding efficiency. Furthermore, when data in one region is corrupted during processing or lost during network transmission, the codec can correctly encode or decode other regions of the same image without relying on the corrupted or lost data, thus providing fault tolerance. In some video coding standards, images can be divided into different types of regions. For example, H.265 / HEVC, H.266 / VVC, and AVS provide two types of regions: "slices" and "tiles." It should also be noted that different images in the video sequence 300 can have different segmentation schemes for dividing an image into multiple regions.

[0069] For example, in Figure 3 In the structure 310, the structure is divided into three regions 314, 316, and 318, the boundaries of which are shown as solid lines within the structure 310. Region 314 includes four basic processing units. Each of regions 316 and 318 includes six basic processing units. It should be noted that... Figure 3 The basic processing unit, basic processing subunit, and region of structure 310 described herein are merely examples, and this disclosure does not constitute a limitation on its embodiments.

[0070] Figure 4A A schematic diagram of an example encoding process 400A consistent with embodiments of this disclosure is shown. For example, the encoding process 400A may be performed by an encoder. Figure 4A As shown, the encoder can encode the video sequence 402 into a video bitstream 428 according to process 400A. Similar to... Figure 3 Video sequences 300 and 402 may include a set of images (referred to as "original images") arranged in chronological order. Similar to... Figure 3 In structure 310, the encoder can divide each raw image of video sequence 402 into multiple basic processing units, multiple basic processing sub-units, or multiple regions for processing. In some embodiments, the encoder can perform process 400A at the level of basic processing units for each raw image of video sequence 402. For example, the encoder can perform process 400A iteratively, wherein the encoder can encode one basic processing unit in one iteration of process 400A. In some embodiments, the encoder can perform process 400A in parallel for multiple regions (e.g., regions 314-318) of each raw image of video sequence 402.

[0071] exist Figure 4A In this process, the encoder feeds the basic processing unit (referred to as the "raw BPU") of the original image of video sequence 402 to prediction stage 404 to generate prediction data 406 and prediction BPU 408. The encoder can subtract prediction BPU 408 from the raw BPU to generate residual BPU 410. The encoder can feed residual BPU 410 to transform stage 412 and quantization stage 414 to generate quantization transform coefficients 416. The encoder can feed prediction data 406 and quantization transform coefficients 416 to binary encoding stage 426 to generate video bitstream 428. Components 402, 404, 406, 408, 410, 412, 414, 416, 426, and 428 can be referred to as the "forward path". During process 400A, after quantization stage 414, the encoder can feed quantization transform coefficients 416 to inverse quantization stage 418 and inverse transform stage 420 to generate reconstructed residual BPU 422. The encoder can add the reconstructed residual BPU 422 to the prediction BPU 408 to generate a prediction reference 424, which is used in the next iteration of process 400A during the prediction phase 404. Components 418, 420, 422, and 424 of process 400A can be referred to as a "reconstruction path". The reconstruction path can be used to ensure that both the encoder and the decoder use the same reference data for prediction.

[0072] The encoder can iteratively execute process 400A to encode each raw BPU of the raw image (in the forward path) and generate a prediction reference 424 for encoding the next raw BPU of the raw image (in the reconstruction path). After encoding all raw BPUs of the raw image, the encoder can continue to encode the next image in the video sequence 402.

[0073] Referring to process 400A, the encoder can receive a video sequence 402 generated by a video acquisition device (e.g., a camera). As used herein, the term "receive" can refer to any action of receiving, inputting, acquiring, retrieving, obtaining, reading, accessing, or otherwise inputting data.

[0074] In prediction phase 404, during the current iteration, the encoder may receive the original BPU and prediction reference 424, and perform prediction operations to generate prediction data 406 and prediction BPU 408. Prediction reference 424 can be generated from the reconstruction path of a previous iteration of process 400A. The purpose of prediction phase 404 is to reduce information redundancy by extracting prediction data 406 from prediction data 406 and prediction reference 424 that can be used to reconstruct the original BPU into prediction BPU 408.

[0075] Ideally, the predicted BPU 408 should be identical to the original BPU. However, due to non-ideal prediction and reconstruction operations, the predicted BPU 408 is typically slightly different from the original BPU. To record this difference, after generating the predicted BPU 408, the encoder can subtract it from the original BPU to generate the residual BPU 410. For example, the encoder can subtract the pixel value (e.g., grayscale or RGB value) of the predicted BPU 408 from the value of the corresponding pixel in the original BPU. Each pixel in the residual BPU 410 can have a residual value generated by this subtraction between the corresponding pixel in the original BPU and the predicted BPU 408. Compared to the original BPU, the predicted data 406 and the residual BPU 410 can have fewer bits, but they can be used to reconstruct the original BPU without significant quality degradation. Therefore, the original BPU is compressed.

[0076] To further compress the residual BPU 410, in the transform phase 412, the encoder can reduce the spatial redundancy of the residual BPU 410 by decomposing it into a set of two-dimensional “basic patterns,” each basic pattern being associated with “transform coefficients.” The basic patterns can have the same size (e.g., the size of the residual BPU 410). Each basic pattern can characterize a frequency-varying component of the residual BPU 410 (e.g., the frequency of brightness variation). No basic pattern can be reproduced by any combination of any other basic patterns (e.g., a linear combination). In other words, the decomposition decomposes the variation of the residual BPU 410 into the frequency domain. This decomposition is analogous to the discrete Fourier transform of a function, where the basic patterns are analogous to the basis functions of the discrete Fourier transform (e.g., trigonometric functions), and the transform coefficients are analogous to the coefficients associated with the basis functions.

[0077] Different transform algorithms can use different base modes. Various transform algorithms can be used in transform stage 412, such as discrete cosine transform, discrete sine transform, etc. The transform in transform stage 412 is reversible. That is, the encoder can recover the residual BPU 410 through the inverse operation of the transform (called the "inverse transform"). For example, to recover a pixel of the residual BPU 410, the inverse transform can be to multiply the value of the corresponding pixel in the base mode by the corresponding correlation coefficient and sum the products to produce a weighted sum. For video coding standards, both the encoder and decoder can use the same transform algorithm (and therefore the same base mode). Therefore, the encoder can only record the transform coefficients, and the decoder can reconstruct the residual BPU 410 based on the transform coefficients without receiving the base mode from the encoder. Compared to the residual BPU 410, the transform coefficients can have fewer bits, but they can be used to reconstruct the residual BPU 410 without significant quality degradation. Therefore, the residual BPU 410 is further compressed.

[0078] The encoder can further compress the transform coefficients in quantization phase 414. During the transform process, different fundamental modes can represent different frequencies of change (e.g., brightness change frequency). Because the human eye is generally better at identifying low-frequency changes, the encoder can ignore information about high-frequency changes without causing significant quality degradation in decoding. For example, in quantization phase 414, the encoder can generate quantized transform coefficients 416 by dividing each transform coefficient by an integer value (called a "quantization scale factor") and performing rounding on the quotient. After this operation, some transform coefficients of the high-frequency fundamental modes can be converted to zero, while the transform coefficients of the low-frequency fundamental modes can be converted to smaller integers. The encoder can ignore the zero-valued quantized transform coefficients 416, thereby further compressing the transform coefficients. The quantization process is also reversible, wherein the quantized transform coefficients 416 can be reconstructed into the transform coefficients in the inverse operation of quantization (called "inverse quantization").

[0079] Because the encoder ignores the remainder of such division in the rounding operation, the quantization stage 414 may be lossy. Typically, the quantization stage 414 may constitute the primary source of information loss in process 400A. The greater the information loss, the fewer bits the quantization transform coefficients 416 may require. To obtain different levels of information loss, the encoder can use different values ​​of the quantization parameters or any other parameters of the quantization process.

[0080] In the binary encoding stage 426, the encoder may use binary encoding techniques to encode the prediction data 406 and the quantization transform coefficients 416, such as entropy coding, variable-length coding, arithmetic coding, Huffman coding, context-adaptive binary arithmetic coding, or any other lossless or lossy compression algorithm. In some embodiments, in addition to the prediction data 406 and the quantization transform coefficients 416, the encoder may also encode other information in the binary encoding stage 426, such as the prediction mode used in the prediction stage 404, the parameters of the prediction operation, the transform type of the transform stage 412, the parameters of the quantization process (e.g., quantization parameters), encoder control parameters (e.g., bitrate control parameters), etc. The encoder may use the output data of the binary encoding stage 426 to generate a video bitstream 428. In some embodiments, the video bitstream 428 may be further packaged for network transmission.

[0081] Referring to the reconstruction path of process 400A, in the inverse quantization stage 418, the encoder can perform inverse quantization on the quantization transform coefficients 416 to generate reconstructed transform coefficients. In the inverse transform stage 420, the encoder can generate a reconstruction residual BPU 422 based on the reconstructed transform coefficients. The encoder can add the reconstruction residual BPU 422 to the prediction BPU 408 to generate a prediction reference 424 to be used in the next iteration of process 400A.

[0082] It should be noted that other variations of process 400A can be used to encode video sequence 402. In some embodiments, the stages of process 400A can be performed by the encoder in different orders. In some embodiments, one or more stages of process 400A can be combined into a single stage. In some embodiments, a single stage of process 400A can be divided into multiple stages. For example, transform stage 412 and quantization stage 414 can be combined into a single stage. In some embodiments, process 400A may include additional stages. In some embodiments, process 400A may be omitted. Figure 4A One or more stages in the process.

[0083] Figure 4B A schematic diagram of another example encoding process 400B consistent with embodiments of the present disclosure is shown. Process 400B can be derived from process 400A. For example, process 400B can be used by an encoder conforming to a hybrid video coding standard (e.g., H.26x series). Compared to process 400A, the forward path of process 400B additionally includes a mode decision stage 430, and divides the prediction stage 404 into a spatial prediction stage 4042 and a temporal prediction stage 4044. The reconstruction path of process 400B additionally includes a loop filtering stage 432 and a buffer 434.

[0084] Generally, prediction techniques can be categorized into two types: spatial prediction and temporal prediction. Spatial prediction (e.g., intra-image prediction or "intra-frame prediction") uses pixels from one or more encoded neighboring BPUs in the same image to predict the current BPU. That is, the prediction reference 424 in spatial prediction can include the neighboring BPUs. Spatial prediction can reduce the inherent spatial redundancy of the image. Temporal prediction (e.g., inter-image prediction or "inter-frame prediction") uses regions from one or more encoded images to predict the current BPU. That is, the prediction reference 424 in temporal prediction can include the encoded images. Temporal prediction can reduce the inherent temporal redundancy of the image.

[0085] Referring to process 400B, in the forward path, the encoder performs prediction operations in spatial prediction phase 4042 and temporal prediction phase 4044. For example, in spatial prediction phase 4042, the encoder may perform the intra-frame prediction. For a given original BPU of an image being encoded, prediction reference 424 may include one or more adjacent BPUs that have already been encoded (in the forward path) and reconstructed (in the reconstruction path) in the same image. The encoder can generate a predicted BPU 408 by extrapolating the adjacent BPUs. The extrapolation technique may include, for example, linear extrapolation or interpolation, polynomial extrapolation or interpolation, etc. In some embodiments, the encoder may perform the extrapolation at the pixel level, for example, extrapolating the corresponding pixel value for each pixel of the predicted BPU 408. The adjacent BPU used for extrapolation can be located from various directions relative to the original BPU, such as in the vertical direction (e.g., at the top of the original BPU), the horizontal direction (e.g., to the left of the original BPU), the diagonal direction (e.g., at the lower left, lower right, upper left, or upper right of the original BPU), or any direction defined in the video coding standard used. For the intra-frame prediction, the prediction data 406 may include, for example, the location (e.g., coordinates) of the adjacent BPU used, the size of the adjacent BPU used, the extrapolation parameters, the orientation of the adjacent BPU used relative to the original BPU, etc.

[0086] In another example, during the temporal prediction phase 4044, the encoder may perform the inter-frame prediction. For a given original BPU of the current image, the prediction reference 424 may include one or more images (referred to as "reference images") that have been encoded (in the forward path) and reconstructed (in the reconstruction path). In some embodiments, the reference images may be encoded and reconstructed on a BPU-by-BPU basis. For example, the encoder may add a reconstructed residual BPU 422 to the prediction BPU 408 to generate a reconstructed BPU. After all reconstructed BPUs of the same image have been generated, the encoder may generate a reconstructed image as the reference image. The encoder may perform a "motion estimation" operation to search for a matching region within a range (referred to as a "search window") of the reference image. The position of the search window in the reference image may be determined based on the position of the original BPU in the current image. For example, the search window may be centered at a position in the reference image that has the same coordinates as the original BPU in the current image and may extend outward by a predetermined distance. When the encoder identifies a region similar to the original BPU in the search window (e.g., using a pixel recursive algorithm, block matching algorithm, etc.), the encoder can determine such a region as a matching region. The matching region may have different specifications than the original BPU (e.g., less than, equal to, or greater than the original BPU, or have a different shape). Because the reference image and the current image are temporally separated in the timeline (e.g., as...), Figure 3 As shown), the matching region can be considered to have "moved" to the location of the original BPU over time. The encoder can record the direction and distance of this movement as a "motion vector". When using multiple reference images (e.g., Figure 3 When the encoder is used to search for a matching region for each reference image (306), it can determine the associated motion vector of the matching region. In some embodiments, the encoder can assign weights to the pixel values ​​of the matching region for each matching reference image.

[0087] The motion estimation can be used to identify various types of motion, such as translation, rotation, scaling, etc. For inter-frame prediction, the prediction data 406 may include, for example, the location (e.g., coordinates) of the matching region, the motion vector associated with the matching region, the number of reference images, the weights associated with the reference images, etc.

[0088] To generate the predicted BPU 408, the encoder can perform a "motion compensation" operation. This motion compensation can be used to reconstruct the predicted BPU 408 based on the predicted data 406 (e.g., motion vectors) and the predicted reference 424. For example, the encoder can move the matching region of the reference image according to the motion vectors, where the encoder can predict the original BPU of the current image. When using multiple reference images (e.g., ... Figure 3 When the encoder moves the matching regions of the plurality of reference images (306) according to the motion vector and average pixel value of each matching region, it can move the matching regions of the plurality of reference images. In some embodiments, if the encoder has already assigned weights to the pixel values ​​of the matching regions of each matching reference image, the encoder can perform a weighted summation of the pixel values ​​of the moved matching regions.

[0089] In some embodiments, the inter-frame prediction can be unidirectional or bidirectional. Unidirectional inter-frame prediction can use one or more reference images in the same temporal direction relative to the current image. For example, Figure 3 Image 304 in the image is a one-way inter-frame prediction image, wherein the reference image (e.g., image 302) precedes image 304. Two-way inter-frame prediction can use one or more reference images in two temporal directions relative to the current image. For example, Figure 3 Image 306 is a bidirectional inter-frame prediction image, wherein the reference images (e.g., images 304 and 308) are in two time directions relative to image 304.

[0090] Still referring to the forward path of process 400B, after spatial prediction stage 4042 and temporal prediction stage 4044, in mode decision stage 430, the encoder can select a prediction mode (e.g., one of intra-frame prediction or inter-frame prediction) for the current iteration of process 400B. For example, the encoder can perform rate-distortion optimization techniques, wherein the encoder selects a prediction mode based on the bit rate of a candidate prediction mode and the distortion of the reference image reconstructed under that candidate prediction mode to minimize the value of the cost function. Based on the selected prediction mode, the encoder can generate a corresponding prediction BPU 408 and prediction data 406.

[0091] In the reconstruction path of process 400B, if an intra-prediction mode has already been selected in the forward path, the encoder can directly feed the prediction reference 424 (e.g., the current BPU that has been encoded and reconstructed in the current image) to the spatial prediction stage 4042 for subsequent use (e.g., for extrapolation of the next BPU of the current image) after generating the prediction reference 424. The encoder can also feed the prediction reference 424 to the loop filtering stage 432, where the encoder can apply loop filtering to the prediction reference 424 to reduce or eliminate distortions (e.g., blockiness) introduced during the encoding of the prediction reference 424. The encoder can apply various loop filtering techniques in the loop filtering stage 432, such as deblocking, sample adaptive shifting, adaptive loop filtering, etc. The loop-filtered reference image can be stored in buffer 434 (or “decoded image buffer”) for subsequent use (e.g., as an inter-frame prediction reference image for future images of video sequence 402). The encoder may store one or more reference images in buffer 434 for use in the time prediction stage 4044. In some embodiments, the encoder may encode the parameters of the loop filter (e.g., the loop filter strength), as well as the quantization transform coefficients 416, the prediction data 406, and other information in the binary encoding stage 426.

[0092] Figure 5A A schematic diagram of an example decoding process 500A consistent with embodiments of this disclosure is shown. Process 500A may be corresponding to... Figure 4A The compression process 400A described above is followed by the decompression process. In some embodiments, process 500A may be similar to the reconstruction path described in process 400A. The decoder can decode the video bitstream 428 into video stream 504 according to process 500A. Video stream 504 may be very similar to video sequence 402. However, due to information loss during compression and decompression (e.g., Figures 4A-4B In the quantization stage 414, video stream 504 is typically different from video sequence 402. Similar to... Figures 4A-4B In processes 400A and 400B, the decoder can perform process 500A at the level of a basic processing unit (BPU) for each image encoded in the video bitstream 428. For example, the decoder can perform process 500A iteratively, wherein the decoder can decode one basic processing unit in one iteration of process 500A. In some embodiments, the decoder can perform process 500A in parallel for multiple regions (e.g., regions 314-318) of each image encoded in the video bitstream 428.

[0093] exist Figure 5AIn this process, the decoder may feed a portion of the video bitstream 428 associated with a basic processing unit (referred to as a "coded BPU") of the encoded image to a binary decoding stage 502. In binary decoding stage 502, the decoder may decode the portion into prediction data 406 and quantization transform coefficients 416. The decoder may feed the quantization transform coefficients 416 to an inverse quantization stage 418 and an inverse transform stage 420 to generate a reconstructed residual BPU 422. The decoder may feed the prediction data 406 to a prediction stage 404 to generate a prediction BPU 408. The decoder may add the reconstructed residual BPU 422 to the prediction BPU 408 to generate a prediction reference 424. In some embodiments, the prediction reference 424 may be stored in a buffer (e.g., a decoded image buffer in computer memory). The decoder may feed the prediction reference 424 to the prediction stage 404 for use in the next iteration of process 500A to perform a prediction operation.

[0094] The decoder can iteratively execute process 500A to decode each encoded BPU of the encoded image and generate a prediction reference 424 for encoding the next encoded BPU of the encoded image. After decoding all encoded BPUs of the encoded image, the decoder can output the image to video stream 504 for display and continue decoding the next encoded image in video bitstream 428.

[0095] In binary decoding stage 502, the decoder may perform the inverse operation of the binary encoding technique used by the encoder (e.g., entropy coding, variable-length coding, arithmetic coding, Huffman coding, context-adaptive binary arithmetic coding, or any other lossless compression algorithm). In some embodiments, in addition to the prediction data 406 and quantization transform coefficients 416, the decoder may also decode other information in binary decoding stage 502, such as prediction mode, parameters of the prediction operation, transform type, parameters of the quantization process (e.g., quantization parameters), encoder control parameters (e.g., bitrate control parameters), etc. In some embodiments, if the video bitstream 428 is transmitted over the network in the form of data packets, the decoder may unpack the video bitstream 428 before feeding it to binary decoding stage 502.

[0096] Figure 5BA schematic diagram of another example decoding process 500B consistent with embodiments of this disclosure is shown. Process 500B can be derived from process 500A. For example, process 500B can be used by a decoder conforming to a hybrid video coding standard (e.g., H.26x series). Compared to process 500A, process 500B further divides the prediction stage 404 into a spatial prediction stage 4042 and a temporal prediction stage 4044, and further includes a loop filtering stage 432 and a buffer 434.

[0097] In process 500B, for the encoded basic processing unit (referred to as the "current BPU") of the encoded image being decoded (referred to as the "current image"), the prediction data 406 decoded by the decoder from the binary decoding stage 502 can include various types of data depending on the prediction mode used by the encoder to encode the current BPU. For example, if the encoder uses intra-frame prediction to encode the current BPU, the prediction data 406 can include a prediction mode indicator (e.g., a flag value) indicating the intra-frame prediction, parameters of the intra-frame prediction operation, etc. The parameters of the intra-frame prediction operation can include, for example, the positions (e.g., coordinates) of one or more neighboring BPUs used as references, the size of the neighboring BPUs, extrapolation parameters, the orientation of the neighboring BPUs relative to the original BPU, etc. For another example, if the encoder uses inter-frame prediction to encode the current BPU, the prediction data 406 can include a prediction mode indicator (e.g., a flag value) indicating the inter-frame prediction, parameters of the inter-frame prediction operation, etc. The parameters of the inter-frame prediction operation may include, for example, the number of reference images associated with the current BPU, the weights associated with the reference images respectively, the positions (e.g., coordinates) of one or more matching regions in each reference image, and one or more motion vectors associated with the matching regions respectively.

[0098] Based on the prediction mode indicator, the decoder can determine whether to perform spatial prediction (e.g., intra-frame prediction) in the spatial prediction phase 4042 or temporal prediction (e.g., inter-frame prediction) in the temporal prediction phase 4044. Figure 4B The details of performing such spatial or temporal predictions are described herein and will not be repeated below. After performing such spatial or temporal predictions, the decoder can generate a prediction BPU 408. The decoder can add the prediction BPU 408 and the reconstruction residual BPU 422 to generate a prediction reference 424, as shown below. Figure 5A As described.

[0099] In process 500B, the decoder can feed prediction reference 424 to spatial prediction stage 4042 or temporal prediction stage 4044 for performing prediction operations in the next iteration of process 500B. For example, if the current BPU is decoded using the intra-frame prediction in spatial prediction stage 4042, the decoder can feed prediction reference 424 directly to spatial prediction stage 4042 for subsequent use (e.g., for extrapolation of the next BPU of the current image) after generating prediction reference 424 (e.g., the decoded current BPU). If the current BPU is decoded using inter-frame prediction in temporal prediction stage 4044, the decoder can feed prediction reference 424 to loop filtering stage 432 to reduce or eliminate distortion (e.g., blockiness) after generating prediction reference 424 (e.g., a reference image where all BPUs have been decoded). The decoder can be configured as follows: Figure 4B The method described herein applies loop filtering to prediction reference 424. The loop-filtered reference image may be stored in buffer 434 (e.g., a decoded image buffer in computer memory) for subsequent use (e.g., as an inter-frame prediction reference image for a future encoded image of video bitstream 428). The decoder may store one or more reference images in buffer 434 for use in the temporal prediction stage 4044. In some embodiments, the prediction data may further include parameters of the loop filtering (e.g., loop filtering strength). In some embodiments, the prediction data includes parameters of the loop filtering when the prediction mode indicator of prediction data 406 indicates that inter-frame prediction was used to encode the current BPU.

[0100] In addition to the block-based video compression techniques mentioned above, deep learning can also be used in video compression to achieve competitive performance compared to traditional compression schemes. For example, end-to-end image compression algorithms have shown better rate-distortion (RD) performance than JPEG, JPEG2000, and even HEVC due to end-to-end training and nonlinear transformations. Furthermore, video compression algorithms based on deep neural networks (DNNs), such as Deep Video Compression Models (DVC), can achieve promising RD performance. These schemes can be implemented without prior knowledge of the video content. For video conferencing / telephone applications, deep generative models, such as First Order Motion Models (FOMMs) and Face_vid2vid, can achieve promising performance at ultra-low bit rates. In particular, these models leverage the fact that variations in these videos are often present in human motion information, thus providing strong prior information that can be used for frame synthesis. These features are described by variations in human structure, such as landmarks or keypoints, and are further fed to animate the reference frames and generate human motion videos.

[0101] Deep learning-based algorithms can be used to replace or enhance some operations or functions of block-based video coding tools, including intra / inter-frame prediction, entropy coding, and loop filtering. End-to-end image / video compression algorithms can be used for joint optimization of the entire image / video compression framework, rather than designing just a specific module. For example, an end-to-end video coding scheme, namely the DVC scheme, can be used, which jointly optimizes all components of video compression. Furthermore, to address content adaptation and error propagation awareness issues, an online encoder update scheme can be used to improve video compression performance. Additionally, flexible feature coding (FVC) can be used to implement the end-to-end compression framework by developing all the main modules in the feature space. Recursive learning for video compression (RLVC) and HLVC can be used to leverage the temporal correlation between video frames based on recursive probabilistic models and weighted recursive quality enhancement networks. Four effective modules from learning-oriented video compression multi-frame prediction (M-LVC) can be used. However, similar to traditional video coding tools, these learning-based video compression methods target general natural scenes without specifically considering human content such as faces, torsos, or other body parts.

[0102] Figure 6This is a schematic diagram illustrating an exemplary architecture of an end-to-end deep learning-based video compression framework 600 according to some embodiments of the present disclosure. Framework 600 uses various deep learning models that jointly optimize components of video compression, such as motion estimation, motion compression, and residual compression. Specifically, the motion information is obtained and the current frame is reconstructed using learned optical flow (also known as dense motion optical flow) estimation. Then, two autoencoder-like neural networks are employed to compress the corresponding motion and residual information. Modules in framework 600 are jointly learned using a single loss function, whereby these modules cooperate by weighing the relationship between reducing the number of compressed bits and improving the quality of the decoded video. Figure 2 The block-based video compression framework 200 shown is... Figure 6 There is a one-to-one correspondence between the end-to-end deep learning-based video compression framework 600 shown. The relationship between them is described below, along with a brief overview of their differences. The end-to-end deep learning-based video compression framework 600 may include an encoder configured to generate a bitstream based on input video frames, and a decoder configured to reconstruct video frames based on the bitstream. For simplicity, Figure 6 Only the encoding end of the 600 video compression framework based on end-to-end deep learning is shown.

[0103] like Figure 6 As shown, framework 600 can perform motion estimation and compression. In the optical flow network module 601, a convolutional neural network (CNN) model can be used to estimate the optical flow, which is treated as motion information. Instead of directly encoding the raw optical flow values, an MV encoder-decoder network is used to compress and decode the optical flow values. First, the MV encoder network module 602 can be used to process the motion information. Encode the motion information. The encoded motion is characterized as The encoded motion representation can be further quantized by the Q module 603. Then, the corresponding reconstructed motion information can be decoded using the MV decoder network module 604. .

[0104] Frame 600 can also perform motion compensation. The motion compensation network, referred to as motion compensation network module 605, is designed to obtain the predicted frame based on the acquired optical flow. Then, the original frame is obtained. With the predicted frame residuals between ,for = – .

[0105] The Frame 600 can also perform transform, quantization, and inverse transform. This is achieved by using a highly nonlinear residual encoder-decoder network, such as... Figure 6 The residual encoder network module 606 shown replaces the linear transformation and converts the residual... Nonlinear mapping to representation Then, the Q module 607 will... Quantified as To construct an end-to-end training scheme, a quantization method is used. The quantized representation... The data is fed into a residual decoder network, called residual decoder network module 608, to obtain the reconstructed residual. .

[0106] Framework 600 can also perform entropy coding. During the testing phase, the quantized motion representation is performed by the bit rate estimation network module 609. and the residual characterization The data is encoded into bits and sent to the decoder. During the training phase, the CNN is used to estimate the required bit count overhead. and The probability distribution of each symbol in the array.

[0107] Furthermore, the loss of the frame 600 can be determined based on the original frame, the reconstructed frame, and the encoded frame. The loss determined here can also be used to optimize network connectivity within the frame 600 to achieve better performance.

[0108] Frame 600 can also perform frame reconstruction ( Figure 6 (not shown in the image), the frame reconstruction method is the same as the frame reconstruction method described in conjunction with frame 200.

[0109] The end-to-end deep learning-based video compression framework 600 can be used for facial video compression, for example, for generative video coding of speaking faces. For example, the end-to-end deep learning-based generative video coding of speaking faces can use generative models such as variational autoencoders (VAEs) and generative adversarial networks (GANs). The facial video compression can achieve potentially significant performance improvements. For example, X2Face can be used to control facial generation through image, audio, and pose encoding. Furthermore, realistic neural talking head models can be used through few-shot adversarial learning. For video-to-video synthesis tasks, Face-vid2vid can be used. Additionally, a scheme utilizing compact 3D keypoint representations to drive generative models to render the target frame can be used. Furthermore, a mobile-compatible video chat system based on FOMM can be used. VSBNet, which reconstructs the original frame from the keypoints using the adversarial learning, can also be used. Furthermore, an end-to-end speaking head video compression framework based on Compact Feature Learning (CFTE) can be used, designed for efficient speaking facial video compression in ultra-low bandwidth scenarios. The CFTE scheme utilizes the compact feature representation to compensate for temporal evolution and reconstructs target facial video frames in an end-to-end manner. Furthermore, under rate-distortion target supervision, the CFTE scheme can be integrated into the video coding framework. While these algorithms achieve frame reconstruction using a small number of facial parameters through the powerful rendering capabilities of deep generative models, certain head pose movements and facial expression movements still cannot be accurately rendered compared to the original motion video.

[0110] Figure 7 This is a schematic diagram illustrating an exemplary deep learning-based video generative compression framework 700 according to some embodiments of the present disclosure. Framework 700 is suitable for compressing and generating speaking facial videos. For example, framework 700 may be based on the first-order motion model (FOMM). The FOMM deforms a reference source frame to follow motion in driving video. While this approach is applicable to various types of video (e.g., moving images, cartoons), it can also be used for facial animation applications. The FOMM follows an encoder-decoder architecture, whose motion delivery components include the following steps.

[0111] First, an isovariant loss is used to learn a keypoint extractor (also known as a motion module) without explicit labels. This keypoint extractor computes two sets of ten learned keypoints for each of the source and driving frames. These learned keypoints are derived from a 64×64 channel-dimension feature map transformed by a Gaussian mapping function, thus each keypoint can represent feature information from different channels. It should be noted that each keypoint is represented by coordinates (x, y), which represent the most important information from the feature map.

[0112] Secondly, a dense motion network uses the markers and the source frames to generate a dense motion field and an occlusion map.

[0113] The encoder 710 then encodes the source frame using conventional image / video compression methods such as HEVC / VVC or JPEG / BPG. Here, VVC is used to compress the source frame.

[0114] In a subsequent stage, the generated feature map is distorted using the dense motion field (using a differentiable grid sampling operation) and then multiplied with the occlusion map.

[0115] Finally, the decoder 720 generates an image from the distorted image.

[0116] Figure 8 This is a schematic diagram illustrating an exemplary encoder-decoder encoding framework 800 with a compact feature size of 1×4×4 for speaking facial video according to some embodiments of the present disclosure. Figure 8 Another fundamental framework for the depth-based video generative compression scheme, namely CFTE, is provided, based on compact feature representation. This fundamental framework follows an encoder-decoder architecture that applies a context-based coding scheme.

[0117] At the encoder 810 end, the compression framework includes three modules: an encoder (also called a VVC encoding module) for compressing the keyframes, a feature extractor for extracting compact human features from other inter-frames, and a feature encoding module for compressing the inter-frame prediction residuals of the compact human features. First, the keyframes representing human texture are compressed using the VVC encoder. Each subsequent inter-frame is represented by a compact feature matrix of size 1×4×4 using the compact feature extractor. It should be noted that the size of the compact feature matrix is ​​not fixed, and the number of feature parameters can be increased or decreased according to specific bit consumption requirements. Then, inter-frame prediction and quantization are performed on these extracted features, and the residual entropy is encoded into the final bitstream.

[0118] At the decoder 820, this compression framework also includes three main modules: decoding for reconstructing the keyframes, reconstructing the compact features through entropy decoding and compensation, and generating the final video by utilizing the reconstructed features and the decoded keyframes. More specifically, during the generation of the final video, the decoded keyframes from the VVC bitstream are further characterized as features through compact feature extraction. Subsequently, given the features from the keyframes and inter-frames, the associated sparse motion field is computed, facilitating the generation of a pixel-level dense motion map (also known as dense motion optical flow) and an occlusion map. Finally, based on a deep generative model, the decoded keyframes, the pixel-level dense motion map, and the occlusion map with implicit motion field features are used to generate a final video with accurate appearance, pose, and expression.

[0119] While these video codecs have enabled applications such as video conferencing, chat, and live streaming, they cannot fully achieve efficient visual facial communication because they do not consider the statistical characteristics of facial visual signals. In addition to these hybrid video coding frameworks, model-based coding (MBC) is specifically designed to improve the efficiency of facial video compression by leveraging strong facial prior information.

[0120] Inspired by recent advances in deep generative models, particularly for Generative Adversarial Networks (GANs), the low-quality face reconstruction of early MBC techniques can be well addressed and improved. Specifically, learning-based face reproduction or animation models hold great promise for generative face video compression (GFVC). Figure 9A This is a schematic diagram illustrating a general flowchart of a generative facial video compression (GFVC) system according to some embodiments of the present disclosure. Figure 9A As shown, a conventional image / video codec 910 encodes and decodes the key reference frame 901 to obtain a decoded key reference frame 903. Subsequent inter-frame frames 902 are processed by a model-based codec 920. Specifically, at the encoding end, the subsequent inter-frame frames 902 are characterized using compact transmission symbols by an analysis model 921, and encoded into an encoded bitstream by parametric encoding 922. At the decoding end, the decoded key reference frame 903 and the facial representation parameters decoded by parametric decoding 923 are jointly fed into a synthesis model 924 to output a reconstructed inter-frame frame 904.

[0121] Figure 9B This is a schematic diagram illustrating facial representations in a generative facial video compression system according to some embodiments of the present disclosure. For example... Figure 9BAs shown, the animation model can efficiently represent the input facial frame 930 using different types of compact facial representation parameters (e.g., 2D landmarks, 2D keypoints, 3D keypoints, compact features, segmentation maps, and facial semantics), and relies on the powerful learning capabilities of deep generative models to reconstruct these facial frames, thereby greatly advancing the development of generative facial video compression technology. In this way, ultra-low bitrate and high-quality reconstructed facial video communication can be achieved.

[0122] While generative facial video compression schemes can achieve satisfactory rate-distortion (RD) performance, several drawbacks and challenges remain, limiting further performance improvements and practical applications. Specifically, each GFVC method employing different facial representations requires a pair of specially trained encoders and decoders. If the encoder and decoder do not match, facial video reconstruction cannot be successfully achieved. This incompatibility reduces the practicality of these methods, preventing their support by common image viewers on computers and mobile devices. Therefore, a converter is needed to be compatible with multiple facial representations from different GFVC encoders to adapt to a fixed decoder.

[0123] Embodiments of this disclosure provide a facial feature converter capable of converting between different types of facial features, and a method for generative facial video compression using the facial feature converter.

[0124] Figure 10 An exemplary framework for generative facial video compression using a facial feature converter system according to some embodiments of this disclosure is shown.

[0125] like Figure 10 As shown, at the encoding end, the face encoder 1010 first extracts facial features of a corresponding type (e.g., type A) from a facial image, such as 2D keypoints, 3D keypoints, and compact features, denoted as facial feature A. Then, the entropy encoder 1020 entropy-encodes the facial feature A into a bitstream and transmits it to the decoder. At the decoding end, the entropy decoder 1030 performs entropy decoding on the facial feature A. For a fixed decoder scenario, the decoder is configured to reconstruct a facial image based on facial features of type B (denoted as facial feature B). Type A and type B can be different. Then, the feature converter 1040 can be used to convert the received facial feature A of any type into facial feature B, while ensuring the reconstruction quality of the final generated face. Next, the face decoder 1050 can decode facial feature B to generate a facial image. A detailed description follows.

[0126] Figure 11This is a schematic diagram illustrating an exemplary method 1100 for generative facial video compression using the facial feature converter according to some embodiments of the present disclosure. For example, method 1100 may be... Figure 7 700, a deep learning-based video generative compression framework Figure 8 encoder-decoder encoding frame 800 or Figure 9A The image / video codec 910 and model-based codec 920 are implemented. (Reference) Figure 10 and Figure 11 Method 1100 includes steps 1102 to 1108.

[0127] In step 1102, a first facial feature is extracted from the input facial image, the first facial feature belonging to a first type. For example, the input facial image of the GFVC system is represented as follows: And the input facial image Facial feature A is encoded by facial encoder 1010. ,in, and These represent the features extracted by the face encoder and the face encoder, respectively.

[0128] In step 1104, the first facial feature is entropy encoded and transmitted to the decoder for entropy decoding to obtain the decoded first facial feature. For example, facial feature A is entropy encoded by entropy encoder 1020 and then decoded by entropy decoder 1030 to obtain the decoded facial feature. ,in, , These represent entropy encoding and entropy decoding, respectively. This indicates that the decoded facial features have been identified. In some embodiments, the entropy encoder 1020 may perform... Figure 4A and Figure 4B The binary encoding process 426 is shown, and the entropy decoder 1030 can execute it. Figure 5A and Figure 5B The binary decoding shown is 502.

[0129] In step 1106, the decoded first facial feature is converted into a second facial feature of a second type by a facial feature converter. The second facial feature belongs to a type specified by the facial feature converter, for example, type B. For example, the decoded facial feature A is converted by facial feature converter 1040 to obtain facial feature B. ,in This represents a facial feature converter. This represents the converted facial features obtained by the converter. In some embodiments, encoded information (e.g., indexes, parameters, or flags) of a specified type (i.e., a second type) is encoded, transmitted, and decoded from the bitstream.

[0130] In step 1108, based on the second facial feature, the facial decoder performs generative reconstruction of the facial image. For example, based on facial feature B, the facial decoder 1050 performs generative reconstruction of the facial image. ,in, This indicates that a facial image has been generated. This refers to the facial decoder.

[0131] To better support fixed decoder scenarios and maintain the versatility of the GFVC method, in some embodiments, a common facial feature converter can be used to achieve feature conversion between any two types. Figure 12 This is a schematic diagram illustrating an exemplary facial feature converter 1200 according to some embodiments of the present disclosure. For generative facial video compression systems supporting various types of facial features, multiple-input multiple-output networks can be used, and a unified embedding framework can be applied. For example... Figure 12 As shown, the facial feature converter 1200 includes multiple encoders 1210 (including encoders 1210a, 1210b, ..., 1210n) and multiple decoders 1220 (including decoders 1220a, 1220b, ..., 1220n), each corresponding to a different feature type. Each encoder 1210 is configured to convert a facial feature into a unified embedding representation; that is, for multiple features of different types, the corresponding encoder 1210 can encode the facial feature of the first type into facial features with the same dimension and in the same common space. Each decoder 1220 is configured to decode a unified embedding representation into a facial feature of the second type. Therefore, the facial feature of the first type is converted into the facial feature of the second type by the facial feature converter 1200. For example, encoder 1210a is configured to encode facial feature A into a unified embedding facial feature, encoder 1210b is configured to encode facial feature B into a unified embedding facial feature, and encoder 1210n is configured to encode facial feature N into a unified embedding facial feature. Decoder 1220a is configured to convert uniformly embedded facial features into facial feature A, decoder 1220b is configured to convert uniformly embedded facial features into facial feature B, and decoder 1220n is configured to convert uniformly embedded facial features into facial feature N.

[0132] In some embodiments, the plurality of encoders (e.g., encoders 1210a, 1210b, ... 1210n) and the plurality of decoders (e.g., 1220a, 1220b, ... 1220n) may include any combination of any number of the following devices: central processing unit (or “CPU”), graphics processing unit (or “GPU”), neural processing unit (“NPU”), microcontroller unit (“MCU”), optical processor, programmable logic controller, microcontroller, microprocessor, digital signal processor, intellectual property (IP) core, programmable logic array (PLA), programmable array logic (PAL), general-purpose array logic (GAL), complex programmable logic device (CPLD), field-programmable gate array (FPGA), system-on-chip (SoC), application-specific integrated circuit (ASIC), etc.

[0133] Figure 13 This is a schematic diagram illustrating an exemplary method 1300 for converting facial features using the facial feature converter according to some embodiments of the present disclosure. For example, method 1300 may be performed by facial feature converter 1200. Reference Figure 12 and Figure 13 Method 1300 includes steps 1302 and 1304.

[0134] In step 1302, a first type of facial feature is encoded by a feature encoder to obtain a uniformly embedded facial feature. In some embodiments, multiple feature encoders 1210 are used to encode different types of facial features respectively. The uniformly embedded facial feature is a facial feature with the same dimension and in the same common space. For example, different encoded facial features have the same dimension and are in the same common space. All facial features output from the feature encoder 1210 are mapped or projected to the same coordinate space. For example, facial features are spatially or semantically aligned and share a consistent reference for comparison, transformation, or reconstruction. Simultaneously, all facial features output from the feature encoder 1210 have the same number of elements (i.e., the same dimension).

[0135] In step 1304, the uniformly embedded facial features are decoded by a feature decoder to obtain a second type of second facial feature. For example, for each desired facial feature type, based on the indication of encoded information (e.g., indexes, parameters, or flags) decoded from the bitstream, a corresponding decoder can be selected from the plurality of feature decoders 1220, and the encoded facial features can be recovered to obtain the second type of second facial feature for generating a facial image. Therefore, method 1300 can use a facial feature converter 1200 to convert a first type of facial feature into a second type of facial feature.

[0136] In some embodiments, the first type of facial features are represented as features And the second type of facial features are represented as features The facial features The transformation can be described as: ,in, Indicates origin from features The unified embedding representation, Indicates the use of encoding features The encoder.

[0137] From the unified embedding to features The transformation can be described as follows: ,in Indicates the use of features to obtain The decoder. Therefore, facial features With facial features The transformation between them can be described as follows: ,in, Indicates from facial features To facial features The aforementioned conversion operation.

[0138] As shown in the embodiments above, this disclosure provides a scheme for generative facial video compression using a facial feature converter. This scheme can convert different decoded facial representations to support the fixed decoder and generate the facial video. The proposed scheme achieves the practicality of these existing GFVC methods and supports image decoders commonly used in computers and mobile phones.

[0139] In some embodiments, model supervision and loss functions may be used in the disclosed facial feature converter. To optimize the disclosed facial feature converter, training data is prepared by extracting facial features using different GFVC models to obtain different types of facial features, including 2D keypoints from a first-order motion model (FOMM), 3D keypoints from Face_vid2vid, compact features from Compact Features for Temporal Evolution (CFTE), facial semantics from Interactive Face Video Coding (IFVC), etc.

[0140] Figure 14 This is a schematic diagram illustrating an exemplary method 1400 for training the facial feature converter according to some embodiments of the present disclosure. For example, method 1400 may be performed by facial feature converter 1200. Reference Figure 14 Method 1400 includes steps 1402 to 1408.

[0141] In step 1402, a set of feasible transformations is constructed. For example, to ensure that the transformation between the original feature and the specified feature is both injective and surjective, feasible transformations are grouped in a "after n features" manner. Specifically, for supporting A system with available facial feature types, ranging from 1 to... Select the "after n features" interval. For the selected interval... The feasible transformation pair can be: }, Characterization with intervals The transformation pairs. The set of all feasible transformations can be represented as: .

[0142] In step 1404, a subset is randomly sampled from a feasible transformation set. For example, it can be drawn from the set K subsets are randomly sampled from the sample.

[0143] In step 1406, the loss between the transformed features and their corresponding ground truth labels is calculated. In some embodiments, the mean absolute error (also known as L1 loss) between the transformed features and their corresponding ground truth labels is calculated. For example, the loss function can be expressed as: The corresponding truth labels are obtained from the facial features using the conventional GFVC method without using the proposed converter.

[0144] In step 1408, the facial feature converter is trained based on the loss. For example, the encoder and decoder are optimized based on the loss to obtain the target loss.

[0145] During implementation, for features With features The conversion between the two can be obtained using the following formula to obtain the decoded facial features. : ,in, Indicates the use of the feature Encoder for encoding, Indicates the use of obtaining the feature The proposed facial feature converter supports the GFVC system, which can use any facial feature type at the encoding end and any fixed specified feature type at the decoding end.

[0146] This disclosure proposes to optimize the facial feature converter by grouping "after n" transformation pairs with injectivity and surjectivity, so that each independent facial encoder and decoder can be optimized in a balanced manner, and each possible transformation pair can be performed equally, to ensure the overall accuracy of the facial feature conversion and improve the broad applicability of the facial feature conversion in the GFVC system.

[0147] Although the above embodiments use the disclosed facial feature converter for generative facial video compression, it is conceivable that the disclosed facial feature converter can also be used for other learning-based compression schemes. For example, existing end-to-end compression algorithms still require a pair of specially trained encoders and decoders, which is not suitable for practical applications. Therefore, the disclosed facial feature converter can be effectively extended to end-to-end (E2E) encoding, where features from different E2E encoders can be efficiently transcoded to support image or video reconstruction with a fixed decoder.

[0148] Embodiments of this disclosure provide another facial feature converter, wherein feature difference conversion may be employed.

[0149] As described in the above embodiments, the facial feature converter directly converts different types of facial features. To further improve the interoperability of generative facial video coding, an optional conversion technique can be employed that converts the feature differences (rather than their original values) between two frames. Figure 15A This is a schematic diagram illustrating an exemplary direct conversion scheme 1500A according to some embodiments of the present disclosure. Figure 15B This is a schematic diagram illustrating an exemplary difference conversion scheme 1500B according to some embodiments of the present disclosure.

[0150] To make the first Taking the conversion of facial features A in a frame to B as an example, in the direct conversion scheme 1500A, such as... Figure 15A As shown, by facial feature converters (e.g., Figure 12 The facial feature converter 1200 shown converts facial features A Convert to features For more details on the direct conversion scheme, please refer to the above text. Figures 12 to 14 The description will not be repeated here.

[0151] refer to Figure 15B In the difference conversion scheme 1500B, for the facial feature converter, facial features in keyframes and feature differences between keyframes and inter-frames can be used.

[0152] Figure 16This is a schematic diagram illustrating another exemplary method 1600 for transforming facial features using a facial feature converter according to some embodiments of the present disclosure. For example, method 1600 may be performed by facial feature converter 1200. Reference Figure 15B and Figure 16 Method 1600 includes steps 1602 to 1616.

[0153] In step 1602, key facial features of keyframes in the video sequence are extracted by the face encoder. For example, the keyframes of the input facial video in the GFVC system are represented as follows: The keyframe The facial feature A is obtained by encoding with facial encoder A, and is represented as follows: ,in, and These represent the features extracted by the facial encoder and the facial encoder, respectively. The facial features... It belongs to the first type, such as type A.

[0154] In step 1604, the first facial features of the inter-frame are extracted. Similar to step 1602, the first facial features can be obtained. Input frame The facial features The facial features It belongs to the first type, such as type A.

[0155] In step 1606, the difference between the key facial feature and the first facial feature is calculated. For example, facial features With facial features The difference between them can be calculated as: .

[0156] In step 1608, the key facial features and the difference are entropy encoded, transmitted, and entropy decoded to obtain the decoded key facial features and the decoded difference. For example, ,in, , These represent entropy encoding and entropy decoding, respectively. In some embodiments, the difference is characterized by the encoded information in the bitstream.

[0157] In step 1610, the facial feature converter (e.g., Figure 12 The converter 1200 shown converts the decoded difference of the facial features into a corresponding second-type feature difference to obtain a converted difference of the second type—for example, type B. .

[0158] In step 1612, the decoded key facial features are recovered by the facial decoder to obtain the recovered key facial features of the second type. For example, the recovered key facial features of the second type are represented as follows: .

[0159] In step 1614, the second facial features are recovered by the face decoder based on the converted difference and the recovered key facial features. For example, keyframes are used. Feature B can be recovered using the following formula. Frame Feature B: .

[0160] In step 1616, the facial image is generatively reconstructed. For example, the facial image can be generatively reconstructed and obtained using the following formula: ,in, This indicates that a facial image has been generated. This refers to the facial decoder.

[0161] The embodiments described in this disclosure can be freely combined.

[0162] Figure 17 This is a block diagram of an exemplary apparatus 1700 for encoding image data according to some embodiments of the present disclosure. Apparatus 1700 can be used to perform the methods described above (e.g., Figure 11 Method 1100 in the middle Figure 13 Method 1300 in the middle Figure 14 Method 1400 and Figure 16 Method 1600 in the middle). For example Figure 17 As shown, device 1700 may include processor 1702. When processor 1702 executes the instructions described herein, device 1700 may become a dedicated machine for video encoding or decoding. Processor 1702 may be any type of circuit system capable of manipulating or processing information. For example, processor 1702 may include any number of central processing units (or “CPU”), graphics processing units (or “GPU”), neural processing units (“NPU”), microcontroller units (“MCU”), optical processors, programmable logic controllers, microcontrollers, microprocessors, digital signal processors, intellectual property (IP) cores, programmable logic arrays (PLAs), programmable array logic (PALs), general-purpose array logic (GALs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), system-on-a-chip (SoCs), application-specific integrated circuits (ASICs), and any combination thereof. In some embodiments, processor 1702 may also be a group of processors grouped into individual logic components. For example, such as Figure 17As shown, processor 1702 may include multiple processors, including processor 1702a, processor 1702b and processor 1702n.

[0163] The device 1700 may also include a memory 1704 configured to store data (e.g., instruction sets, computer code, intermediate data, etc.). For example, as Figure 17 As shown, the stored data may include program instructions (e.g., program instructions for implementing the methods described in this disclosure). Processor 1702 can access the program instructions and the data for processing (e.g., via bus 1710), and execute the program instructions to perform operations or manipulations on the data for processing. Memory 1704 may include a high-speed random access memory device or a non-volatile memory device. In some embodiments, memory 1704 may include any combination of any number of random access memories (RAM), read-only memories (ROM), optical discs, magnetic disks, hard disks, solid-state drives, flash drives, secure digital cards (SD cards), memory sticks, compact flash (CF) cards, etc. Memory 1704 may also be a group of memories grouped into a single logical component. Figure 17 (Not shown in the image).

[0164] Bus 1710 may be a communication device for transmitting data between components within device 1700, such as an internal bus (e.g., CPU-memory bus), an external bus (e.g., a Universal Serial Bus port, a Peripheral Component Interconnect Fast Port), etc.

[0165] For ease of explanation and to avoid ambiguity, processor 1702 and other data processing circuitry are collectively referred to as "data processing circuitry" in this disclosure. The data processing circuitry may be implemented entirely as hardware, or as a combination of software, hardware, or firmware. Furthermore, the data processing circuitry may be a single, independent module, or may be wholly or partially integrated into any other component of device 1700.

[0166] The device 1700 may also include a network interface 1706 to provide wired or wireless communication with a network (e.g., the Internet, intranet, local area network, mobile communication network, etc.). In some embodiments, the network interface 1706 may include any combination of any number of network interface controllers (NICs), radio frequency (RF) modules, transceivers, transceivers, modems, routers, gateways, wired network adapters, wireless network adapters, Bluetooth adapters, infrared adapters, near field communication (“NFC”) adapters, cellular network chips, etc.

[0167] In some embodiments, the device 1700 may further include a peripheral interface 1708 to provide connectivity to one or more peripheral devices. For example... Figure 17As shown, the peripheral devices may include, but are not limited to, cursor control devices (e.g., mouse, touchpad, or touchscreen), keyboards, displays (e.g., cathode ray tube displays, liquid crystal displays, or light-emitting diode displays), video input devices (e.g., cameras or input interfaces coupled to video archives), etc.

[0168] It should be noted that the video codec consistent with this disclosure can be implemented as any combination of any software or hardware modules in the apparatus 1700. For example, some or all stages of the disclosed methods can be implemented as one or more software modules of the apparatus 1700, such as program instructions that can be loaded into memory 1704. As another example, some or all stages of the disclosed methods can be implemented as one or more hardware modules of the apparatus 1700, such as dedicated data processing circuitry (e.g., FPGA, ASIC, NPU, etc.).

[0169] In some embodiments, a non-transitory computer-readable storage medium for storing the bitstream is also provided. The bitstream can be encoded and decoded using a facial feature converter according to the generative facial video compression method described above. For example, the bitstream may include encoded reference frames, encoded facial features, or encoded differences between facial features. The disclosed facial feature converter can be used to decode the encoded facial features and the encoded differences of a first type and convert them into facial features and differences of a second type, and can be based on the methods described above (e.g., methods 1100 and 1600). Figure 11 and Figure 16 Obtain the output video.

[0170] In some embodiments, a non-transitory computer-readable storage medium including an instruction set is also provided, and the instruction set can be executed by a device (e.g., the disclosed encoder and decoder) to perform the methods described above. Common forms of non-transitory media include, for example, floppy disks, flexible disks, hard disks, solid-state drives, magnetic tape or any other magnetic data storage media, CD-ROMs, any other optical data storage media, any physical media with a perforated pattern, RAM, PROMs and EPROMs, FLASH-EPROMs or any other flash memory, NVRAMs, caches, registers, any other memory chips or cassette tapes, and their networking versions. The device may include one or more processors (CPUs), input / output interfaces, network interfaces, and / or memory.

[0171] It should be noted that the embodiments described in this disclosure can be freely combined or used alone.

[0172] The embodiments may be further described using the following terms: 1. A video decoding method, comprising: Receive bitstream; and Decoding one or more images using the encoded information of the bitstream, wherein the decoding includes: Decode a first facial feature of a first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

[0173] 2. The method according to Clause 1, wherein converting the first facial feature into the second type of second facial feature further includes: Encode the first facial feature to obtain a uniformly embedded facial feature; and Decode the unified embedded facial features to obtain the second type of second facial features.

[0174] 3. The method according to Clause 1, wherein the first facial feature is a facial feature of an inter-frame frame, and converting the first facial feature into a second facial feature of the second type further includes: Obtain the restored facial features of the keyframe, wherein the restored facial features belong to the second type; Decode the encoded information, which represents the difference between the first facial features of the first type and the first facial features of the inter-frame frame; Convert the first facial feature difference into a second type of second facial feature difference; and Based on the difference between the recovered facial features and the second facial features of the keyframe, the second type of the second facial features of the inter-frame frame is generated.

[0175] 4. The method according to Clause 3, wherein converting the first facial feature difference into a second facial feature difference of the second type further comprises: Encode the first facial feature difference of the first type to obtain a uniformly embedded facial feature difference; and Decode the unified embedded facial feature difference to obtain the second type of the second facial feature difference.

[0176] 5. The method described in Clause 1 further includes: Decoding indicates the encoded information of the second type; and Based on the encoded information, the first facial feature is converted into the second facial feature of the second type.

[0177] 6. The method according to Clause 5, wherein the first type is different from the second type.

[0178] 7. The method according to Clause 6, wherein the first type and the second type are selected from: 2D keypoints in a first-order motion model (FOMM), 3D keypoints in Face_vid2vid, compact features in Compact Features for Temporal Evolution (CFTE), and facial semantics in Interactive Face Video Coding (IFVC).

[0179] 8. A facial feature converter, comprising: Multiple encoders, each configured to encode a first type of first facial feature into a uniform embedding feature; and Multiple decoders coupled to one or more of the encoders, each decoder being configured to decode the uniform embedding feature into a second type of second facial feature, wherein the first type is different from the second type.

[0180] 9. The facial feature converter according to Clause 8, wherein the plurality of encoders and the plurality of decoders are model-based encoders and decoders.

[0181] 10. The face feature converter according to Clause 9, wherein the model used for the model-based codec includes: a first-order motion model (FOMM), Face_vid2vid, compact features for temporal evolution (CFTE), and interactive face video coding (IFVC).

[0182] 11. The facial feature converter according to Clause 10, wherein the first type and the second type are selected from: 2D keypoints in the FOMM, 3D keypoints in the Face_vid2vid, compact features in the CFTE, and facial semantics in the IFVC.

[0183] 12. The facial feature converter according to Clause 9, wherein a loss of the facial feature converter is calculated, and the facial feature converter is trained based on the loss.

[0184] 13. A non-transitory computer-readable medium having a set of instructions stored thereon, the set of instructions being executable by one or more processors of a device to cause the device to perform the following operations: Decode the first facial features of the first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

[0185] 14. The non-transitory computer-readable medium according to Clause 13, wherein converting the first facial feature into the second facial feature of the second type further comprises: Encode the first facial feature to obtain a uniformly embedded facial feature; and Decode the unified embedded facial features to obtain the second type of second facial features.

[0186] 15. The non-transitory computer-readable medium according to Clause 13, wherein the first facial feature is a facial feature of an inter-frame frame, and converting the first facial feature into a second facial feature of the second type further comprises: Obtain the restored facial features of the keyframe, wherein the restored facial features belong to the second type; Decode the encoded information, which represents the difference between the first facial features of the first type and the first facial features of the inter-frame frame; Convert the first facial feature difference into a second type of second facial feature difference; and Based on the difference between the recovered facial features and the second facial features of the keyframe, the second type of the second facial features of the inter-frame frame is generated.

[0187] 16. The non-transitory computer-readable medium according to Clause 15, wherein converting the first facial feature difference into a second facial feature difference of the second type further comprises: Encode the first facial feature difference of the first type to obtain a uniformly embedded facial feature difference; and Decode the unified embedded facial feature difference to obtain the second type of the second facial feature difference.

[0188] 17. The non-transitory computer-readable medium according to Clause 13, wherein the operation further comprises: Decoding indicates the encoded information of the second type; and Based on the encoded information, the first facial feature is converted into the second facial feature of the second type.

[0189] 18. The non-transitory computer-readable medium as described in Clause 17, wherein the first type is different from the second type.

[0190] 19. The non-transitory computer-readable medium as described in Clause 18, wherein the first type and the second type are selected from: 2D keypoints in a first-order motion model (FOMM), 3D keypoints in Face_vid2vid, compact features in Compact Features for Temporal Evolution (CFTE), and facial semantics in Interactive Face Video Coding (IFVC).

[0191] It should be noted that the relational terms such as "first," "second," etc., used in this document are only used to distinguish one entity or operation from another, and do not require or imply any actual relationship or order between these entities or operations. Furthermore, the words "comprising," "having," "containing," and "including," as well as other similar forms, are intended to be identical in meaning and open-ended, because one or more items following any of these words do not imply an exhaustive list of such one or more items, nor do they imply limitation to only the listed one or more items.

[0192] As used herein, unless otherwise specified, the term "or" covers all possible combinations unless impractical. For example, if it is specified that a database may include A or B, then unless otherwise specified or impractical, the database may include A, or B, or A and B. As a second example, if it is specified that a database may include A, B, or C, then unless otherwise specified or impractical, the database may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0193] It should be understood that the above embodiments can be implemented in hardware, software (program code), or a combination of hardware and software. If implemented in software, it can be stored in the aforementioned computer-readable medium. When executed by a processor, the software can perform the disclosed methods. The computing units and other functional units described in this disclosure can be implemented in hardware, software, or a combination of hardware and software. Those skilled in the art should also understand that multiple modules / units described above can be combined into one module / unit, and each module / unit described above can be further divided into multiple sub-modules / sub-units.

[0194] In the foregoing specification, numerous specific details have been described with reference to embodiments, which may vary depending on the implementation. Certain adjustments and modifications may be made to the described embodiments. Other embodiments will be apparent to those skilled in the art upon consideration of the specification and practice of the art disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the appended claims. The sequence of steps shown in the figures is also to be considered for illustrative purposes only and is not intended to be limited to any particular sequence of steps. Therefore, those skilled in the art will understand that these steps may be performed in a different order while implementing the same method.

[0195] Exemplary embodiments have been disclosed in the accompanying drawings and description. However, many variations and modifications can be made to these embodiments. Therefore, although specific terminology has been used, it is used only in a general and descriptive sense and not for limiting purposes.

Claims

1. A video decoding method, comprising: Receive bit stream; as well as Decoding one or more images using the encoded information of the bitstream, wherein the decoding includes: Decode a first facial feature of a first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

2. The method according to claim 1, wherein, Converting the first facial feature into the second type of the second facial feature further includes: Encode the first facial feature to obtain a uniformly embedded facial feature; and Decode the unified embedded facial features to obtain the second type of second facial features.

3. The method according to claim 1, wherein, The first facial feature is a facial feature from an inter-frame frame, and converting the first facial feature into a second facial feature of the second type further includes: Obtain the restored facial features of the keyframe, wherein the restored facial features belong to the second type; Decode the encoded information, which represents the difference between the first facial features of the first type and the first facial features of the inter-frame frame; Convert the first facial feature difference into a second type of second facial feature difference; and Based on the difference between the recovered facial features and the second facial features of the keyframe, the second type of the second facial features of the inter-frame frame is generated.

4. The method according to claim 3, wherein, Converting the first facial feature difference to the second type of second facial feature difference further includes: Encode the first facial feature difference of the first type to obtain a uniformly embedded facial feature difference; and Decode the unified embedded facial feature difference to obtain the second type of the second facial feature difference.

5. The method according to claim 1, further comprising: Decoding indicates the encoded information of the second type; as well as Based on the encoded information, the first facial feature is converted into the second facial feature of the second type.

6. The method according to claim 5, wherein, The first type is different from the second type.

7. The method according to claim 6, wherein, The first type and the second type are selected from: 2D keypoints in the first-order motion model FOMM, 3D keypoints in Face_vid2vid, compact features in the compact features for temporal evolution CFTE, and facial semantics in interactive face video coding IFVC.

8. A facial feature converter, comprising: Multiple encoders, each configured to encode a first type of first facial feature into a uniform embedding feature; as well as Multiple decoders coupled to one or more of the encoders, each decoder being configured to decode the uniform embedding feature into a second type of second facial feature, wherein the first type is different from the second type.

9. The facial feature converter according to claim 8, wherein, The plurality of encoders and the plurality of decoders are model-based codecs.

10. The facial feature converter according to claim 9, wherein, The models used for the model-based codec include: a first-order motion model FOMM, Face_vid2vid, compact features for temporal evolution CFTE, and interactive face video coding IFVC.

11. A method for encoding a video sequence into a bitstream, comprising: Receive video sequences; Encode one or more images from the video sequence; as well as Send bit stream; The encoding includes: Extract a first facial feature from a facial image, wherein the first facial feature belongs to a first type; The first facial feature is encoded into a bitstream and transmitted to the decoder; The decoder end may include the following operations: Decode the first facial feature of the first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

12. A video decoding apparatus, comprising: The receiver is configured to receive bit streams; as well as A decoder is configured to decode one or more images using encoded information from the bitstream, wherein the decoding includes: Decode a first facial feature of a first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

13. An apparatus for encoding a video sequence into a bitstream, comprising: The receiver is configured to receive video sequences; An encoder is configured to encode one or more images of the video sequence; as well as The transmitter is configured to send a bit stream; The encoding includes: Extract a first facial feature from a facial image, wherein the first facial feature belongs to a first type; The first facial feature is encoded into a bitstream and transmitted to the decoder; The decoder end may include the following operations: Decode the first facial feature of the first type from the bitstream; Convert the first facial feature into a second type of second facial feature; and The facial image is reconstructed based on the second facial feature.

14. A non-transitory computer-readable medium having stored thereon an instruction set and a video bitstream, the instruction set being executable by one or more processors of a device to perform a method for decoding the video bitstream, the method comprising: Decode the first facial features of the first type from the bitstream; Convert the first facial feature into a second type of second facial feature; as well as The facial image is reconstructed based on the second facial feature.

15. The non-transitory computer-readable medium according to claim 14, wherein, Converting the first facial feature into the second type of the second facial feature further includes: Encode the first facial feature to obtain a uniformly embedded facial feature; and Decode the unified embedded facial features to obtain the second type of second facial features.

16. The non-transitory computer-readable medium according to claim 14, wherein, The first facial feature is a facial feature from an inter-frame frame, and converting the first facial feature into a second facial feature of the second type further includes: Obtain the restored facial features of the keyframe, wherein the restored facial features belong to the second type; Decode the encoded information, which represents the difference between the first facial features of the first type and the first facial features of the inter-frame frame; Convert the first facial feature difference into a second type of second facial feature difference; and Based on the difference between the recovered facial features and the second facial features of the keyframe, the second type of the second facial features of the inter-frame frame is generated.

17. The non-transitory computer-readable medium according to claim 16, wherein, Converting the first facial feature difference to the second type of second facial feature difference further includes: Encode the first facial feature difference of the first type to obtain a uniformly embedded facial feature difference; and Decode the unified embedded facial feature difference to obtain the second type of the second facial feature difference.

18. The non-transitory computer-readable medium according to claim 14, wherein, The operation also includes: Decoding indicates the encoded information of the second type; and Based on the encoded information, the first facial feature is converted into the second facial feature of the second type.

19. The non-transitory computer-readable medium according to claim 18, wherein, The first type is different from the second type.

20. The non-transitory computer-readable medium according to claim 19, wherein, The first type and the second type are selected from: 2D keypoints in the first-order motion model FOMM, 3D keypoints in Face_vid2vid, compact features in the compact features for temporal evolution CFTE, and facial semantics in interactive face video coding IFVC.

Images