Method, apparatus, and medium for video processing
By determining and filtering contextual information for video frames using a neural network-based model, the method improves video coding quality by mitigating error propagation and enhancing the coding process.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- DOUYIN VISION CO LTD
- Filing Date
- 2026-01-23
- Publication Date
- 2026-07-30
AI Technical Summary
Neural network-based video coding quality is expected to be further improved, and existing methods face challenges in mitigating error propagation during frame-by-frame coding.
A method for video processing that determines contextual information for a current frame from previous frames, filters this information using a neural network-based model, and performs conversion based on the filtered information to refine the coding process.
This approach mitigates error propagation and enhances coding quality by refining contextual information within the coding loop.
Smart Images

Figure CN2026074731_30072026_PF_FP_ABST
Abstract
Description
METHOD, APPARATUS, AND MEDIUM FOR VIDEO PROCESSINGFIELDS
[0001] Cases of the present disclosure relates generally to video processing techniques, and more particularly, to neural network-based video coding.BACKGROUND
[0002] The past decade has witnessed the rapid development of deep learning in a variety of areas, especially in computer vision and image processing. Neural network was invented originally with the interdisciplinary research of neuroscience and mathematics. It has shown strong capabilities in the context of non-linear transform and classification. Neural network-based image / video compression technology has gained significant progress during the past half decade. It is reported that the latest neural network-based image compression algorithm achieves comparable rate-distortion (R-D) performance with Versatile Video Coding (VVC) . With the performance of neural image compression continually being improved, neural network-based video compression has become an actively developing research area. However, coding quality of neural network-based video coding is generally expected to be further improved.SUMMARY
[0003] Implementations of the present disclosure provide a solution for video processing.
[0004] In a first aspect, a method for video processing is proposed. The method comprises: determining, for a conversion between a current frame of a video and a bitstream of the video with a neural network (NN) -based model, contextual information for the current frame from at least one frame coded before the current frame; filtering the contextual information for the current frame; and performing the conversion based on the filtered contextual information.
[0005] Based on the method in accordance with the first aspect of the present disclosure, the contextual information for the current frame is filtered and the filtered contextual information is used to performing the conversion. Compared with the conventional solution, the proposed method can advantageously refine the contextual information within the coding loop, and thus error propagation during frame-by-frame coding can be mitigated. Therefore, the coding quality can be improved.
[0006] In a second aspect, an apparatus for video processing is proposed. The apparatus comprises a processor and a non-transitory memory with instructions thereon. The instructions upon execution by the processor, cause the processor to perform a method in accordance with the first aspect of the present disclosure.
[0007] In a third aspect, a non-transitory computer-readable storage medium is proposed. The non-transitory computer-readable storage medium stores instructions that cause a processor to perform a method in accordance with the first aspect of the present disclosure.
[0008] In a fourth aspect, another non-transitory computer-readable recording medium is proposed. The non-transitory computer-readable recording medium stores a bitstream of a video which is generated by a method performed by an apparatus for video processing. The method comprises: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; and generating the bitstream based on the filtered contextual information with a neural network (NN) -based model.
[0009] In a fifth aspect, a method for storing a bitstream of a video is proposed. The method comprises: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; generating the bitstream based on the filtered contextual information with a neural network (NN) -based model; and storing the bitstream in a non-transitory computer-readable recording medium.
[0010] In a sixth aspect, there is provided a computer program for performing a method in accordance with the first aspect of the present disclosure when the computer program runs on a computer.
[0011] In a seventh aspect, there is provided a computer program product tangibly stored in a computer storage medium and comprising computer-executable instructions that, when executed by a device, cause the device to perform a method in accordance with the first aspect of the present disclosure.
[0012] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Through the following detailed description with reference to the accompanying drawings, the above and other objectives, features, and advantages of example cases of the present disclosure will become more apparent. In the example cases of the present disclosure, the same reference numerals usually refer to the same components.
[0014] Fig. 1A illustrates a block diagram that illustrates an example visual data coding system, in accordance with some cases of the present disclosure;
[0015] Fig. 1B illustrates a typical transform coding scheme;
[0016] Fig. 2 illustrates an image from the Kodak dataset and different representations of the image;
[0017] Fig. 3 illustrates a network architecture of an autoencoder implementing the hyperprior model;
[0018] Fig. 4 illustrates a block diagram of a combined model;
[0019] Fig. 5 illustrates an encoding process;
[0020] Fig. 6 illustrates a decoding process;
[0021] Figs. 7A and 7B illustrate an example structure of the multistage context model;
[0022] Fig. 8A illustrates an example of contextual filtering structure in accordance with some cases of the present disclosure,
[0023] Fig. 8B illustrates an example of reconstruction enhancement structure in accordance with some cases of the present disclosure;
[0024] Fig. 9 illustrates the overall framework of the video coding solution in accordance with some cases of the present disclosure;
[0025] Fig. 10 illustrates a flowchart of a method for video processing in accordance with some cases of the present disclosure;
[0026] Fig. 11A illustrates another example of contextual filtering structure in accordance with some cases of the present disclosure,
[0027] Fig. 11B illustrates another example of reconstruction enhancement structure in accordance with some cases of the present disclosure; and
[0028] Fig. 12 illustrates a block diagram of a computing device in which various cases of the present disclosure can be implemented.
[0029] Throughout the drawings, the same or similar reference numerals usually refer to the same or similar elements.DETAILED DESCRIPTION
[0030] Principle of the present disclosure will now be described with reference to some cases. It is to be understood that these cases are described only for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. The disclosure described herein can be implemented in various manners other than the ones described below.
[0031] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this disclosure belongs.
[0032] References in the present disclosure to “one case, ” “a case, ” “an example case, ” and the like indicate that the case described may include a particular feature, structure, or characteristic, but it is not necessary that every case includes the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same case. Further, when a particular feature, structure, or characteristic is described in connection with an example case, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other cases whether or not explicitly described.
[0033] It shall be understood that although the terms “first” and “second” etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of example cases. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.
[0034] The terminology used herein is for the purpose of describing particular cases only and is not intended to be limiting of example cases. As used herein, the singular forms “a” , “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” , “comprising” , “has” , “having” , “includes” and / or “including” , when used herein, specify the presence of stated features, elements, and / or components etc., but do not preclude the presence or addition of one or more other features, elements, components and / or combinations thereof. Example Environment
[0035] Fig. 1A is a block diagram that illustrates an example visual data coding system 100 that may utilize the techniques of this disclosure. As shown, the visual data coding system 100 may include a source device 110 and a destination device 120. The source device 110 can be also referred to as a visual data encoding device, and the destination device 120 can be also referred to as a visual data decoding device. In operation, the source device 110 can be configured to generate encoded visual data and the destination device 120 can be configured to decode the encoded visual data generated by the source device 110. The source device 110 may include a visual data source 112, a visual data encoder 114, and an input / output (I / O) interface 116.
[0036] The visual data source 112 may include a source such as a visual data capture device. Examples of the visual data capture device include, but are not limited to, an interface to receive visual data from a visual data provider, a computer graphics system for generating visual data, and / or a combination thereof.
[0037] The visual data may comprise one or more pictures of a video or one or more images. The visual data encoder 114 encodes the visual data from the visual data source 112 to generate a bitstream. The bitstream may include a sequence of bits that form a coded representation of the visual data. The bitstream may include coded pictures and associated visual data. The coded picture is a coded representation of a picture. The associated visual data may include sequence parameter sets, picture parameter sets, and other syntax structures. The I / O interface 116 may include a modulator / demodulator and / or a transmitter. The encoded visual data may be transmitted directly to destination device 120 via the I / O interface 116 through the network 130A. The encoded visual data may also be stored onto a storage medium / server 130B for access by destination device 120.
[0038] The destination device 120 may include an I / O interface 126, a visual data decoder 124, and a display device 122. The I / O interface 126 may include a receiver and / or a modem. The I / O interface 126 may acquire encoded visual data from the source device 110 or the storage medium / server 130B. The visual data decoder 124 may decode the encoded visual data. The display device 122 may display the decoded visual data to a user. The display device 122 may be integrated with the destination device 120, or may be external to the destination device 120 which is configured to interface with an external display device.
[0039] The visual data encoder 114 and the visual data decoder 124 may operate according to a visual data coding standard, such as video coding standard or still picture coding standard and other current and / or further standards.
[0040] Some exemplary cases of the present disclosure will be described in detailed hereinafter. It should be understood that section headings are used in the present document to facilitate ease of understanding and do not limit the cases disclosed in a section to only that section. Furthermore, while certain cases are described with reference to Versatile Video Coding or other specific visual data codecs, the disclosed techniques are applicable to other coding technologies also. Furthermore, while some cases describe coding steps in detail, it will be understood that corresponding steps decoding that undo the coding will be implemented by a decoder. Furthermore, the term visual data processing encompasses visual data coding or compression, visual data decoding or decompression and visual data transcoding in which visual data are represented from one compressed format into another compressed format or at a different compressed bitrate. 1. Brief Summary A neural network-based video compression method for better compression performance with the filtering enhancement technologies. In the present disclosure, contextual filtering and reconstruction enhancement is proposed to boost the coding performance. 2. Introduction The past decade has witnessed the rapid development of deep learning in a variety of areas, especially in computer vision and image processing. Inspired from the great success of deep learning technology to computer vision areas, many researchers have shifted their attention from conventional image / video compression techniques to neural image / video compression technologies. Neural network was invented originally with the interdisciplinary research of neuroscience and mathematics. It has shown strong capabilities in the context of non-linear transform and classification. Neural network-based image / video compression technology has gained significant progress during the past half decade. It is reported that the latest neural network-based image compression algorithm achieves comparable R-D performance with Versatile Video Coding (VVC) , the latest video coding standard developed by Joint Video Experts Team (JVET) with experts from MPEG and VCEG. With the performance of neural image compression continually being improved, neural network-based video compression has become an actively developing research area. However, neural network-based video coding still remains in its infancy due to the inherent difficulty of the problem. 2.1. Image / Video compression Image / video compression usually refers to the computing technology that compresses image / video into binary code to facilitate storage and transmission. The binary codes may or may not support losslessly reconstructing the original image / video, termed lossless compression and lossy compression. Most of the efforts are devoted to lossy compression since lossless reconstruction is not necessary in most scenarios. Usually the performance of image / video compression algorithms is evaluated from two aspects, i.e. compression ratio and reconstruction quality. Compression ratio is directly related to the number of binary codes, the less the better; Reconstruction quality is measured by comparing the reconstructed image / video with the original image / video, the higher the better. Image / video compression techniques can be divided into two branches, the classical video coding methods and the neural-network-based video compression methods. Classical video coding schemes adopt transform-based solutions, in which researchers have exploited statistical dependency in the latent variables (e.g., DCT or wavelet coefficients) by carefully hand-engineering entropy codes modeling the dependencies in the quantized regime. Neural network-based video compression is in two flavors, neural network-based coding tools and end-to-end neural network-based video compression. The former is embedded into existing classical video codecs as coding tools and only serves as part of the framework, while the latter is a separate framework developed based on neural networks without depending on classical video codecs. In the last three decades, a series of classical video coding standards have been developed to accommodate the increasing visual content. The international standardization organizations ISO / IEC has two expert groups namely Joint Photographic Experts Group (JPEG) and Moving Picture Experts Group (MPEG) , and ITU-T also has its own Video Coding Experts Group (VCEG) which is for standardization of image / video coding technology. The influential video coding standards published by these organizations include JPEG, JPEG 2000, H. 262, H. 264 / AVC and H. 265 / HEVC. After H. 265 / HEVC, the Joint Video Experts Team (JVET) formed by MPEG and VCEG has been working on a new video coding standard Versatile Video Coding (VVC) . The first version of VVC was released in July 2020. An average of 50%bitrate reduction is reported by VVC under the same visual quality compared with HEVC. Neural network-based image / video compression is not a new solution since there were a number of researchers working on neural network-based image coding. But the network architectures were relatively shallow, and the performance was not satisfactory. Benefit from the abundance of data and the support of powerful computing resources, neural network-based methods are better exploited in a variety of applications. At present, neural network-based image / video compression has shown promising improvements, confirmed its feasibility. Nevertheless, this technology is still far from mature and a lot of challenges need to be addressed. 2.2. Neural networks Neural networks, also known as artificial neural networks (ANN) , are the computational models used in machine learning technology which are usually composed of multiple processing layers and each layer is composed of multiple simple but non-linear basic computational units. One benefit of such deep networks is believed to be the capacity for processing data with multiple levels of abstraction and converting data into different kinds of representations. Note that these representations are not manually designed; instead, the deep network including the processing layers is learned from massive data using a general machine learning procedure. Deep learning eliminates the necessity of handcrafted representations, and thus is regarded useful especially for processing natively unstructured data, such as acoustic and visual signal, whilst processing such data has been a longstanding difficulty in the artificial intelligence field. 2.3. Neural networks for image compression Existing neural networks for image compression methods can be classified in two categories, i.e., pixel probability modeling and auto-encoder. The former one belongs to the predictive coding strategy, while the latter one is the transform-based solution. Sometimes, these two methods are combined together in literature. 2.3.1. Pixel Probability Modeling According to Shannon’s information theory, the optimal method for lossless coding can reach the minimal coding rate -log2 p (x) where p (x) is the probability of symbol x. A number of lossless coding methods were developed in literature and among them arithmetic coding is believed to be among the optimal ones. Given a probability distribution p (x) , arithmetic coding ensures that the coding rate to be as close as possible to its theoretical limit -log2 p (x) without considering the rounding error. Therefore, the remaining problem is to how to determine the probability, which is however very challenging for natural image / video due to the curse of dimensionality. Following the predictive coding strategy, one way to model p (x) is to predict pixel probabilities one by one in a raster scan order based on previous observations, where x is an image. p (x) = p (x1) p (x2|x1) …p (xi|x1, …, xi-1) …p (xm×n|x1, …, xm×n-1) (1) where m and n are the height and width of the image, respectively. The previous observation is also known as the context of the current pixel. When the image is large, it can be difficult to estimate the conditional probability, thereby a simplified method is to limit the range of its context. p (x) = p (x1) p (x2|x1) …p (xi|xi-k, …, xi-1) …p (xm×n|xm×n-k, …, xm×n-1) (2) where k is a pre-defined constant controlling the range of the context. It should be noted that the condition may also take the sample values of other color components into consideration. For example, when coding the RGB color component, R sample is dependent on previously coded pixels (including R / G / B samples) , the current G sample may be coded according to previously coded pixels and the current R sample, while for coding the current B sample, the previously coded pixels and the current R and G samples may also be taken into consideration. Neural networks were originally introduced for computer vision tasks and have been proven to be effective in regression and classification problems. Therefore, it has been proposed using neural networks to estimate the probability of p (xi) given its context x1, x2, …, xi-1. The pixel probability is proposed for binary images, i.e., xi∈ {-1, +1} . The neural autoregressive distribution estimator (NADE) is designed for pixel probability modeling, where is a feed-forward network with a single hidden layer. A similar work is presented, where the feed-forward network also has connections skipping the hidden layer, and the parameters are also shared. Experiments have been performed on the binarized MNIST dataset. NADE is extended to a real-valued model RNADE, where the probability p (xi|x1, …, xi-1) is derived with a mixture of Gaussians. Their feed-forward network also has a single hidden layer, but the hidden layer is with rescaling to avoid saturation and uses rectified linear unit (ReLU) instead of sigmoid. NADE and RNADE are improved by using reorganizing the order of the pixels and with deeper neural networks. Designing advanced neural networks plays an important role in improving pixel probability modeling. Multi-dimensional long short-term memory (LSTM) is proposed, which is working together with mixtures of conditional Gaussian scale mixtures for probability modeling. LSTM is a special kind of recurrent neural networks (RNNs) and is proven to be good at modeling sequential data. The spatial variant of LSTM is used for images later. Several different neural networks are studied, including RNNs and CNNs namely PixelRNN and PixelCNN, respectively. In PixelRNN, two variants of LSTM, called row LSTM and diagonal BiLSTM are proposed, where the latter is specifically designed for images. PixelRNN incorporates residual connections to help train deep neural networks with up to 12 layers. In PixelCNN, masked convolutions are used to suit for the shape of the context. Comparing with previous works, PixelRNN and PixelCNN are more dedicated to natural images: they consider pixels as discrete values (e.g., 0, 1, …, 255) and predict a multinomial distribution over the discrete values; they deal with color images in RGB color space; they work well on large-scale image dataset ImageNet. Gated PixelCNN is proposed to improve the PixelCNN and achieves comparable performance with PixelRNN but with much less complexity. PixelCNN++ is proposed with the following improvements upon PixelCNN: a discretized logistic mixture likelihood is used rather than a 256-way multinomial distribution; down-sampling is used to capture structures at multiple resolutions; additional short-cut connections are introduced to speed up training; dropout is adopted for regularization; RGB is combined for one pixel. PixelSNAIL is proposed, in which casual convolutions are combined with self-attention. Most of the above methods directly model the probability distribution in the pixel domain. Some researchers also attempt to model the probability distribution as a conditional one upon explicit or latent representations. That being said, we may estimate where h is the additional condition and p (x) =p (h) p (x|h) , meaning the modeling is split into an unconditional one and a conditional one. The additional condition can be image label information or high-level representations. 2.3.2. Auto-encoder Auto-encoder originates from the well-known work proposed by Hinton and Salakhutdinov. The method is trained for dimensionality reduction and consists of two parts: encoding and decoding. The encoding part converts the high-dimension input signal to low-dimension representations, typically with reduced spatial size but a greater number of channels. The decoding part attempts to recover the high-dimension input from the low-dimension representation. Auto-encoder enables automated learning of representations and eliminates the need of hand-crafted features, which is also believed to be one of the most important advantages of neural networks. Fig. 1 illustrates a typical transform coding scheme. The original image x is transformed by the analysis network ga to achieve the latent representation y. The latent representation y is quantized and compressed into bits. The number of bits R is used to measure the coding rate. The quantized latent representation is then inversely transformed by a synthesis network gs to obtain the reconstructed image The distortion is calculated in a perceptual space by transforming x and with the function gp. It is intuitive to apply auto-encoder network to lossy image compression. We only need to encode the learned latent representation from the well-trained neural networks. However, it is not trivial to adapt auto-encoder to image compression since the original auto-encoder is not optimized for compression thereby not efficient by directly using a trained auto-encoder. In addition, there exist other major challenges: First, the low-dimension representation should be quantized before being encoded, but the quantization is not differentiable, which is required in backpropagation while training the neural networks. Second, the objective under compression scenario is different since both the distortion and the rate need to be take into consideration. Estimating the rate is challenging. Third, a practical image coding scheme needs to support variable rate, scalability, encoding / decoding speed, interoperability. In response to these challenges, a number of researchers have been actively contributing to this area. The prototype auto-encoder for image compression is in Fig. 1B, which can be regarded as a transform coding strategy. The original image x is transformed with the analysis network y=ga (x) , where y is the latent representation which will be quantized and coded. The synthesis network will inversely transform the quantized latent representation back to obtain the reconstructed image The framework is trained with the rate-distortion loss function, i.e., where D is the distortion between x and R is the rate calculated or estimated from the quantized representation and λ is the Lagrange multiplier. It should be noted that D can be calculated in either pixel domain or perceptual domain. All existing research works follow this prototype and the difference might only be the network structure or loss function. In terms of network structure, RNNs and CNNs are the most widely used architectures. In the RNNs relevant category, a general framework was proposed for variable rate image compression using RNN. They use binary quantization to generate codes and do not consider rate during training. The framework indeed provides a scalable coding functionality, where RNN with convolutional and deconvolution layers is reported to perform decently. Then an improved version was proposed by upgrading the encoder with a neural network similar to PixelRNN to compress the binary codes. The performance is reportedly better than JPEG on Kodak image dataset using MS-SSIM evaluation metric. The RNN-based solution was further improved by introducing hidden-state priming. In addition, an SSIM-weighted loss function is also designed, and spatially adaptive bitrates mechanism is enabled. They achieve better results than BPG on Kodak image dataset using MS-SSIM as evaluation metric. A general framework was designed for rate-distortion optimized image compression. They use multiary quantization to generate integer codes and consider the rate during training, i.e. the loss is the joint rate-distortion cost, which can be MSE or others. They add random uniform noise to stimulate the quantization during training and use the differential entropy of the noisy codes as a proxy for the rate. They use generalized divisive normalization (GDN) as the network structure, which consists of a linear mapping followed by a nonlinear parametric normalization. The effectiveness of GDN on image coding is verified. An improved version was proposed, where they use 3 convolutional layers each followed by a down-sampling layer and a GDN layer as the forward transform. Accordingly, they use 3 layers of inverse GDN each followed by an up-sampling layer and convolution layer to stimulate the inverse transform. In addition, an arithmetic coding method is devised to compress the integer codes. The performance is reportedly better than JPEG and JPEG 2000 on Kodak dataset in terms of MSE. Furthermore, it was further improved by devising a scale hyper-prior into the auto-encoder. They transform the latent representation y with a subnet ha to z = ha (y) and z will be quantized and transmitted as side information. Accordingly, the inverse transform is implemented with a subnet hs attempting to decode from the quantized side information to the standard deviation of the quantized which will be further used during the arithmetic coding of On the Kodak image set, their method is slightly worse than BPG in terms of PSNR. The structures were further explored in the residue space by introducing an autoregressive model to estimate both the standard deviation and the mean. In the latest work Gaussian mixture model was used to further remove redundancy in the residue. The reported performance is on par with VVC on the Kodak image set using PSNR as evaluation metric. 2.3.3. Hyper Prior Model In the transform coding approach to image compression, the encoder subnetwork (section 2.3.2) transforms the image vector x using a parametric analysis transform into a latent representation y, which is then quantized to form Because is discrete-valued, it can be losslessly compressed using entropy coding techniques such as arithmetic coding and transmitted as a sequence of bits. Fig. 2 illustrates an image from the Kodak dataset and different representations of the image. As evident from the middle left and middle right image of Fig. 2, there are significant spatial dependencies among the elements of Notably, their scales (middle right image) appear to be coupled spatially. An additional set of random variables are introduced to capture the spatial dependencies and to further reduce the redundancies. In this case the image compression network is depicted in Fig. 3. In Fig. 3, the left hand of the models is the encoder ga and decoder gs (explained in section 2.3.2) . The right-hand side is the additional hyper encoder ha and hyper decoder hs networks that are used to obtain In this architecture the encoder subjects the input image x to ga, yielding the responses y with spatially varying standard deviations. The responses y are fed into ha, summarizing the distribution of standard deviations in z. z is then quantized compressed, and transmitted as side information. The encoder then uses the quantized vector to estimate σ, the spatial distribution of standard deviations, and uses it to compress and transmit the quantized image representation The decoder first recovers from the compressed signal. It then uses hs to obtain σ, which provides it with the correct probability estimates to successfully recover as well. It then feeds into gs to obtain the reconstructed image. When the hyper encoder and hyper decoder are added to the image compression network, the spatial redundancies of the quantized latent are reduced. The rightmost image in Fig. 2 correspond to the quantized latent when hyper encoder / decoder are used. Compared to middle right image, the spatial redundancies are significantly reduced, as the samples of the quantized latent are less correlated. As shown in Fig. 2, an image from the Kodak dataset is shown on the left. Visualization of the latent representation y of that image is shown on the middle left. In addition, standard deviations σ of the latent is shown on the middle right, and latents y after the hyper prior (hyper encoder and decoder) network is introduced on the right. Fig. 3 illustrates a network architecture of an autoencoder implementing the hyperprior model. The left side shows an image autoencoder network, the right side corresponds to the hyperprior subnetwork. The analysis and synthesis transforms are denoted as ga and gs, respectively. Q represents quantization, and AE, AD represent arithmetic encoder and arithmetic decoder, respectively. The hyperprior model consists of two subnetworks, hyper encoder (denoted with ha) and hyper decoder (denoted with hs) . The hyper prior model generates a quantized hyper latent which comprises information about the probability distribution of the samples of the quantized latent is included in the bitsteam and transmitted to the receiver (decoder) along with 2.3.4. Context Model Although the hyperprior model improves the modelling of the probability distribution of the quantized latent additional improvement can be obtained by utilizing an autoregressive model that predicts quantized latents from their causal context (Context Model) . The term auto-regressive means that the output of a process is later used as input to it. For example the context model subnetwork generates one sample of a latent, which is later used as input to obtain the next sample. Fig. 4 is a schematic diagram illustrating an example combined model configured to jointly optimize a context model along with a hyperprior and the autoencoder. The following Table 1 illustrates meaning of different symbols. Table 1 –Illustration of symbols A joint architecture was used where both hyperprior model subnetwork (hyper encoder and hyper decoder) and a context model subnetwork are utilized. The hyperprior and the context model are combined to learn a probabilistic model over quantized latents which is then used for entropy coding. As depicted in Fig. 4, the outputs of context subnetwork and hyper decoder subnetwork are combined by the subnetwork called Entropy Parameters, which generates the mean μ and scale (or variance) σ parameters for a Gaussian probability model. The gaussian probability model is then used to encode the samples of the quantized latents into bitstream with the help of the arithmetic encoder (AE) module. In the decoder the gaussian probability model is utilized to obtain the quantized latents from the bitstream by arithmetic decoder (AD) module. As shown in Fig. 4, the combined model jointly optimizes an autoregressive component that estimates the probability distributions of latents from their causal context (Context Model) along with a hyperprior and the underlying autoencoder. Real-valued latent representations are quantized (Q) to create quantized latents and quantized hyper-latents which are compressed into a bitstream using an arithmetic encoder (AE) and decompressed by an arithmetic decoder (AD) . The highlighted region corresponds to the components that are executed by the receiver (i.e. a decoder) to recover an image from a compressed bitstream. Typically, the latent samples are modeled as gaussian distribution or gaussian mixture models (not limited to) . According to Fig. 4, the context model and hyper prior are jointly used to estimate the probability distribution of the latent samples. Since a gaussian distribution can be defined by a mean and a variance (aka sigma or scale) , the joint model is used to estimate the mean and variance (denoted as μ and σ) . 2.3.5. The encoding process using joint auto-regressive hyper prior model Fig. 4 corresponds to the state-of-the-art compression method. In this section and the next, the encoding and decoding processes will be described separately. Fig. 5 depicts the encoding process. The input image is first processed with an encoder subnetwork. The encoder transforms the input image into a transformed representation called latent, denoted by y. y is then input to a quantizer block, denoted by Q, to obtained the quantized latent is then converted to a bitstream (bits1) using an arithmetic encoding module (denoted AE) . The arithmetic encoding block converts each sample of the into a bitstream (bits1) one by one, in a sequential order. The modules hyper encoder, context, hyper decoder, and entropy parameters subnetworks are used to estimate the probability distributions of the samples of the quantized latent The latent y is input to hyper encoder, which outputs the hyper latent (denoted by z) . The hyper latent is then quantized and a second bitstream (bits2) is generated using arithmetic encoding (AE) module. The factorized entropy module generates the probability distribution, that is used to encode the quantized hyper latent into bitstream. The quantized hyper latent includes information about the probability distribution of the quantized latent The Entropy Parameters subnetwork generates the probability distribution estimations, that are used to encode the quantized latent The information that is generated by the Entropy Parameters typically include a mean μ and scale (or variance) σ parameters, that are together used to obtain a gaussian probability distribution. A gaussian distribution of a random variable x is defined as wherein the parameter μ is the mean or expectation of the distribution (and also its median and mode) , while the parameter σ is its standard deviation (or variance, or scale) . In order to define a gaussian distribution, the mean and the variance need to be determined. The entropy parameters module is used to estimate the mean and the variance values. The subnetwork hyper decoder generates part of the information that is used by the entropy parameters subnetwork, the other part of the information is generated by the autoregressive module called context module. The context module generates information about the probability distribution of a sample of the quantized latent, using the samples that are already encoded by the arithmetic encoding (AE) module. The quantized latent is typically a matrix composed of many samples. The samples can be indicated using indices, such as [i, j, k] or [i, j] depending on the dimensions of the matrix The samples [i, j] are encoded by AE one by one, typically using a raster scan order. In a raster scan order the rows of a matrix are processed from top to bottom, wherein the samples in a row are processed from left to right. In such a scenario (wherein the raster scan order is used by the AE to encode the samples into bitstream) , the context module generates the information pertaining to a sample [i, j] , using the samples encoded before, in raster scan order. The information generated by the context module and the hyper decoder are combined by the entropy parameters module to generate the probability distributions that are used to encode the quantized latent into bitstream (bits1) . Finally, the first and the second bitstream are transmitted to the decoder as result of the encoding process. It is noted that the other names can be used for the modules described above. In the above description, all of the elements in Fig. 5 are collectively called encoder. The analysis transform that converts the input image into latent representation is also called an encoder (or auto-encoder) . 2.3.6. The decoding process using joint auto-regressive hyper prior model Fig. 6 depicts the state-of-the-art decoding process. In the decoding process, the decoder first receives the first bitstream (bits1) and the second bitstream (bits2) that are generated by a corresponding encoder. The bits2 is first decoded by the arithmetic decoding (AD) module by utilizing the probability distributions generated by the factorized entropy subnetwork. The factorized entropy module typically generates the probability distributions using a predetermined template, for example using predetermined mean and variance values in the case of gaussian distribution. The output of the arithmetic decoding process of the bits2 is which is the quantized hyper latent. The AD process reverts to AE process that was applied in the encoder. The processes of AE and AD are lossless, meaning that the quantized hyper latent that was generated by the encoder can be reconstructed at the decoder without any change. After obtaining of it is processed by the hyper decoder, whose output is fed to entropy parameters module. The three subnetworks, context, hyper decoder and entropy parameters that are employed in the decoder are identical to the ones in the encoder. Therefore, the exact same probability distributions can be obtained in the decoder (as in encoder) , which is essential for reconstructing the quantized latent without any loss. As a result, the identical version of the quantized latent that was obtained in the encoder can be obtained in the decoder. After the probability distributions (e.g. the mean and variance parameters) are obtained by the entropy parameters subnetwork, the arithmetic decoding module decodes the samples of the quantized latent one by one from the bitstream bits1. From a practical standpoint, autoregressive model (the context model) is inherently serial, and therefore cannot be sped up using techniques such as parallelization. Finally, the fully reconstructed quantized latent is input to the synthesis transform (denoted as decoder in Fig. 6) module to obtain the reconstructed image. In the above description, all of the elements in Fig. 6 are collectively called decoder. The synthesis transform that converts the quantized latent into reconstructed image is also called a decoder (or auto-decoder) . 2.4 Multistage Context model for entropy probability modeling In section 2.3.4, pixel CNN is employed as the context model to further build the relationship between different models, which provides better coding performance through the will designed context modeling. However, it limits the throughput of the probability prediction, because there is a strong dependence on the probability modeling of the elements, it needs to be decoded serially one by one, which might be a nightmare for high resolution images. To solve these issues, a multistage context model is proposed as the replacement of the context model, it provides better parallelism in GPU calculation and greatly reduces the decoding time of high-resolution images. Specifically, the Multistage Context model divided the latent information into 2 groups in channel dimension, as shown in Fig. 7A. Fig. 7A and 7B illustrate an example structure of the multistage context model. After splitting the channel dimension, for latent of each group, 4-stage context modeling will be applied to build context information for latent elements. The shape of the context modeling is provided in Fig. 7B. Benefit from the design of the multistage context model, the times of the context prediction process is no longer dependent on the resolution of the input image, it comes to be a fixed number. In this design, it only need to perform the context prediction process 8 times, so that the decoding time is much faster than the original autoregressive context model. 2.5. Neural networks for video compression Similar to conventional video coding technologies, neural image compression serves as the foundation of intra compression in neural network-based video compression, thus development of neural network-based video compression technology comes later than neural network-based image compression but needs far more efforts to solve the challenges due to its complexity. Starting from 2017, a few researchers have been working on neural network-based video compression schemes. Compared with image compression, video compression needs efficient methods to remove inter-picture redundancy. Inter-picture prediction is then a crucial step in these works. Motion estimation and compensation is widely adopted but is not implemented by trained neural networks until recently. Studies on neural network-based video compression can be divided into two categories according to the targeted scenarios: random access and the low-latency. In random access case, it requires the decoding can be started from any point of the sequence, typically divides the entire sequence into multiple individual segments and each segment can be decoded independently. In low-latency case, it aims at reducing decoding time thereby usually merely temporally previous frames can be used as reference frames to decode subsequent frames. 2.5.1. Low-latency The early work first splits the video sequence frames into blocks and each block will choose one from two available modes, either intra coding or inter coding. If intra coding is selected, there is an associated auto-encoder to compress the block. If inter coding is selected, motion estimation and compensation are performed with tradition methods and a trained neural network will be used for residue compression. The outputs of auto-encoders are directly quantized and coded by the Huffman method. Another neural network-based video coding scheme with PixelMotionCNN was proposed. The frames are compressed in the temporal order, and each frame is split into blocks which are compressed in the raster scan order. Each frame will firstly be extrapolated with the preceding two reconstructed frames. When a block is to be compressed, the extrapolated frame along with the context of the current block are fed into the PixelMotionCNN to derive a latent representation. Then the residues are compressed by the variable rate image scheme. This scheme performs on par with H. 264. Another end-to-end neural network-based video compression framework is then proposed, in which all the modules are implemented with neural networks. The scheme accepts current frame and the prior reconstructed frame as inputs and optical flow will be derived with a pre-trained neural network as the motion information. The motion information will be warped with the reference frame followed by a neural network generating the motion compensated frame. The residues and the motion information are compressed with two separate neural auto-encoders. The whole framework is trained with a single rate-distortion loss function. It achieves better performance than H. 264. An advanced neural network-based video compression scheme is proposed. It inherits and extends traditional video coding schemes with neural networks with the following major features: 1) using only one auto-encoder to compress motion information and residues; 2) motion compensation with multiple frames and multiple optical flows; 3) an on-line state is learned and propagated through the following frames over time. This scheme achieves better performance in MS-SSIM than HEVC reference software. An extended end-to-end neural network-based video compression framework is proposed afterwards. In this solution, multiple frames are used as references. It is thereby able to provide more accurate prediction of current frame by using multiple reference frames and associated motion information. In addition, motion field prediction is deployed to remove motion redundancy along temporal channel. Postprocessing networks are also introduced in this work to remove reconstruction artifacts from previous processes. The performance is better than H. 265 by a noticeable margin in terms of both PSNR and MS-SSIM. The scale-space flow is then proposed to replace commonly used optical flow by adding a scale parameter. It is reportedly achieving better performance than H. 264. A multi-resolution representation for optical flows is proposed. Concretely, the motion estimation network produces multiple optical flows with different resolutions and let the network to learn which one to choose under the loss function. The performance is better than H. 265. 2.5.2. Random access A frame interpolation based method was initially designed. The key frames are first compressed with a neural image compressor and the remaining frames are compressed in a hierarchical order. They perform motion compensation in the perceptual domain, i.e. deriving the feature maps at multiple spatial scales of the original frame and using motion to warp the feature maps, which will be used for the image compressor. The method is reportedly on par with H. 264. Another interpolation-based video compression is then proposed, wherein the interpolation model combines motion information compression and image synthesis, and the same auto-encoder is used for image and residual. Afterwards, a neural network-based video compression method based on variational auto-encoders with a deterministic encoder is proposed. Concretely, the model consists of an auto-encoder and an auto-regressive prior. Different from previous methods, this method accepts a group of pictures (GOP) as inputs and incorporates a 3D autoregressive prior by taking into account of the temporal correlation while coding the laten representations. It provides comparative performance as H. 265. 2.5.3. Deep Contextual Video Compression The learning-based video compression model consists of an independent intra codec, a contextual or encoder EC, a contextual decoder DC, a motion encoder, a motion decoder, and an optical flow net. The intra codec is used to compress the key frame (I frame) in each GOP (group of images) . When compress the P frames xt, optical flow net is used to estimate the flow between previous decoded frame and current P frame xt. Flow map is employed as the motion vector. At encoder, a motion codec is used to compress the flow map and the reconstructed flow map is employed for consistency between encoding and decoding. The decoded frame is warped to context Ct based on the decoded optical flow When compressing P frame xt, xt is concatenated with context Ct and the result is fed to the contextual codec. The concatenation between xt and Ct is to let the network learn conditional coding. At decoder, the key frame or I frame is first decoded using the independent intra codec decoder. The optical flow is decoded using the motion codec decoder. The decoded frame is warped to context Ct based on the decoded optical flow. The inter codec is used to decompress the learned conditional coding information and it is concatenated with the context Ct to obtain the reconstructed frame The overall process is as follows: 2.5.4. Multi-Scale Deep Contextual Video Compression The compression of frame xt is taken as an example. Multi-scale features are first extracted from propagated feature The propagated feature is the feature extracted from previous frames. Motion vector is adopted to warp the multi-scale features to multi-scale local contexts The is concatenated to the current frame xt, is concatenated the to the mid-feature and the is concatenated to mid-feature The concatenation lets the network learn how to conduct conditional coding by itself. When decoding, multi-scale contexts are also concatenated to recover the frame. The overall process can be formulated as: where EC and ED are the contextual encoder and the contextual decoder. 2.6. Preliminaries Almost all the natural image / video is in digital format. A grayscale digital image can be represented by where is the set of values of a pixel, m is the image height and n is the image width. For example, is a common setting and in this case thus the pixel can be represented by an 8-bit integer. An uncompressed grayscale digital image has 8 bits-per-pixel (bpp) , while compressed bits are definitely less. A color image is typically represented in multiple channels to record the color information. For example, in the RGB color space an image can be denoted by with three separate channels storing Red, Green and Blue information. Similar to the 8-bit grayscale image, an uncompressed 8-bit RGB image has 24 bpp. Digital images / videos can be represented in different color spaces. The neural network-based video compression schemes are mostly developed in RGB color space while the traditional codecs typically use YUV color space to represent the video sequences. In YUV color space, an image is decomposed into three channels, namely Y, Cb and Cr, where Y is the luminance component and Cb / Cr are the chroma components. The benefits come from that Cb and Cr are typically down sampled to achieve pre-compression since human vision system is less sensitive to chroma components. A color video sequence is composed of multiple color images, called frames, to record scenes at different timestamps. For example, in the RGB color space, a color video can be denoted by X = {x0, x1, …, xt, …, xT-1} where T is the number of frames in this video sequence, If m = 1080, n = 1920, and the video has 50 frames-per-second (fps) , then the data rate of this uncompressed video is 1920×1080×8×3×50 = 2,488,320,000 bits-per-second (bps) , about 2.32 Gbps, which needs a lot storage thereby definitely needs to be compressed before transmission over the internet. Usually the lossless methods can achieve compression ratio of about 1.5 to 3 for natural images, which is clearly below requirement. Therefore, lossy compression is developed to achieve further compression ratio, but at the cost of incurred distortion. The distortion can be measured by calculating the average squared difference between the original image and the reconstructed image, i.e., mean-squared-error (MSE) . For a grayscale image, MSE can be calculated with the following equation. Accordingly, the quality of the reconstructed image compared with the original image can be measured by peak signal-to-noise ratio (PSNR) : where is the maximal value in e.g., 255 for 8-bit grayscale images. There are other quality evaluation metrics such as structural similarity (SSIM) and multi-scale SSIM (MS-SSIM) . To compare different lossless compression schemes, it is sufficient to compare either the compression ratio given the resulting rate or vice versa. However, to compare different lossy compression methods, it has to take into account both the rate and reconstructed quality. For example, to calculate the relative rates at several different quality levels, and then to average the rates, is a commonly adopted method; the average relative rate is known as Bjontegaard’s delta-rate (BD-rate) . There are other important aspects to evaluate image / video coding schemes, including encoding / decoding complexity, scalability, robustness, and so on. 3. Problems Even though significant progress has been made in conditional based NVC, several potential improvements remain unexplored. For example, how to effectively integrate advanced filtering techniques into conditional video coding remains an open question. Filtering, particularly filtering inside the coding loop, is a promising method to reduce propagation error accumulation during long prediction chains and theoretically offers great potential. However, effectively utilizing these techniques requires addressing challenges such as balancing the trade-off between rate and distortion. In addition, the potential contributions of out-of-loop enhancement to conditional NVC merit further investigation. 4. Detailed solutions 4.1. Details of the solution The detailed solutions below should be considered as examples to explain general concepts. These solutions should not be interpreted in a narrow way. Furthermore, these solutions can be combined in any manner. Figs. 8A and 8B illustrate an example of contextual filtering and reconstruction enhancement structure, respectively. 1. In one example, contextual filtering will be applied prior to the motion compensation process to enhance the quality of of Ct. a) In one example, neural network will be employed to enhance the contextual filtering process, as it shown in the Fig. 8A. i. In one example, residual blocks are utilized in the contextual filtering network. 1. In one example, each residual block consists 1 / 2 / 3 / 4 / 5 convolution layers. 2. In one example, a single convolution layer is applied before residual blocks. 3. In one example, a single convolution layer is applied after residual blocks. 4. In one example, as depicted in Fig. 8A, multiple residual blocks are concatenated. Let M denote the total number of residual blocks used in the contextual filtering network, where M can be one of the following values: 8, 10, 16, 24, or 32. 5. In one example, the number of filters N in the convolutional layers can be set to one of the following values: 16, 32, 64, or 96. ii. In one example, transformer block is utilized in the contextual filtering network. iii. In one example, an activation function is utilized in the contextual filtering network. iv. In one example, in addition to Ct, rate index / lambda value will be fed into the contextual filtering. v. In one example, in addition to Ct, reference frames will be fed into the contextual filtering. vi. In one example, in addition to Ct, pre-coded frames will be fed into the contextual filtering. vii. In one example, in addition to Ct, intra frames might be fed into the contextual filtering. b) In one example, to train contextual filtering network, training loss is employed, let Rt denotes the rate of t-th frame, Dt denotes the distortion of t-th frame, the training loss can be calculated as: i. In one example, Dt might be mean square error. ii. In one example, Dt might be ms-ssim. iii. In one example, a multi-stage training strategy is employed to train the contextual filtering network: the network initially trains on single-frame sequences and then progressively extends to sequences of two frames, three frames, seven frames, and finally, long-sequence training. 2. Reconstruction enhancement will be employed to further improve the coding performance beyond the coding loop by employing a reconstruction enhancement network. a) In one example, neural network will be employed to enhance the reconstruction enhancement process. i. In one example, residual block is utilized in the reconstruction enhancement network, as it shown in the Fig. 8B. 1. In one example, each residual block consists of 1 / 2 / 3 / 4 / 5 convolution layers. 2. In one example, a single convolution layer is applied before residual blocks. 3. In one example, a single convolution layer is applied after residual blocks. 4. In one example, as depicted in Fig. 8B, multiple residual blocks are concatenated. Let M denote the total number of residual blocks used in the contextual filtering network, where M can be one of the following values: 8, 10, 16, 24, or 32. 5. In one example, the number of filters N in the convolutional layers can be set to one of the following values: 16, 32, 64, or 96. ii. In one example, transformer block is utilized in the reconstruction enhancement network. iii. In one example, activation function is utilized in the reconstruction enhancement network. iv. In one example, besides rate index / lambda value will be fed into the reconstruction enhancement to boost the coding performance. v. In one example, besides reference frames and reconstructed optical flow will be fed into the reconstruction enhancement. vi. In one example, besides contextual information will be fed into the reconstruction enhancement. vii. In one example, besides intra frames will be fed into the reconstruction enhancement. b) In one example, to train contextual filtering network, training loss is employed, let and denote the reconstructed frame and enhance frame, the training loss can be calculated as: i. In one example, D might be mean square error. ii. In one example, D might be ms-ssim. 3. Encoder decision will be employed to determine whether to enable the proposed contextual filtering and reconstruction enhancement. a) In one example, contextual counter con, which records the number of frames that enable contextual filtering within a context refresh period will be utilized. i. In one example, the con can be 0 / 1 / 2 / 3 / 4, or other integer number. ii. In one example, the con is predefined values. iii. In one example, the con can be calculated on the encoder based on initial frames. b) In one example, a maximum quality counter mqc is employed in conjunction with con. c) In one example, when con is lower than mqc, contextual filtering will be enabled if it results in quality improvement. d) In one example, contextual counter con will reset to zero at the beginning of context refresh. e) In one example, when con is larger than mqc, a progressive rate loss strategy is employed to determine whether to enable contextual filtering, let r1 denote the rate without contextual filtering, r2 represents the rate with contextual filtering, t stands the current frame index, tfl denotes the total frame length, and pf denotes the progressive factor, the loss is defined as follows: i. In one example, the contextual filtering will be enabled when Lpl is lower than 0. ii. In one example, the pf can be 0.12 / 0.14 / 0.16 / 0.18 / 0.20, or other positive float number. iii. In one example, the pf can be calculated and transmitted on the encoder based on initial frames. f) In one example, a flag is transmitted in the bitstream to indicate whether contextual filtering is enabled for each frame. g) In one example, the encoder makes a decision on whether to enable reconstruction enhancement. Specifically, the encoder evaluates whether applying reconstruction enhancement improves the quality of the current frame and determines whether to enable or disable reconstruction enhancement for that frame. 4.2. Benefit of the solutions According to the present disclosure, several filtering technologies will be utilized based on the top of a well-trained end-to-end compression model, so that the main codec does not need to be retrained. The proposed filters can further boost the coding performance based on the top of the current model. 5. Example Implementations Fig. 9 demonstrate the overall framework of the proposed method. xt, and denote the t-th frame, the t-th reconstructed frame, and the t-th enhanced frame, respectively. Ct and represent the t-th contextual frame and the t-th enhanced contextual frame. Modules in green boxes presents the module in existing works, while blue and orange boxes represent the proposed in-loop contextual filtering Fcon and out-of-loop reconstruction enhancement Frec respectively. Based on the filtering network, we proposed corresponding encoder decision mechanisms, which can be seen in the Algorithm 1 in the following Table 2. Table 2 -Encoding procedure with coding decision
[0041] More details of the cases of the present disclosure will be described below which are related to neural network-based video coding. Inspired by the success of neural networks, several studies have explored enhancing filtering techniques in traditional video compression by leveraging deep learning. In an existing design, a learning based loop filter is introduced to reduce compression artifacts in traditional codecs. Instead of directly subtracting the prediction frame from the current frame to remove temporal redundancy, conditional coding tends to maintain high-dimensional contextual feature information (a.k.a., contextual information) . This information is used as the conditional input in the transformation module to better utilize temporal correlations in the coding process. While filtering has significantly improved traditional coding, its integration into the conditional coding framework remains relatively unexplored.
[0042] To solve the above problems and some other problems not mentioned, visual data processing solutions as described below are disclosed. The cases of the present disclosure should be considered as examples to explain the general concepts and should not be interpreted in a narrow way. Furthermore, these cases can be applied individually or combined in any manner.
[0043] Fig. 10 illustrates a flowchart of a method 1000 for video processing in accordance with some cases of the present disclosure. The method 1000 may be implemented during a conversion between a current frame of a video and a bitstream of the video with a neural network (NN) -based model. As used herein, an NN-based model may be a model based on neural network technologies. For example, an NN-based model may specify sequence of neural network modules (also called architecture) and model parameters. The neural network module may comprise a set of neural network layers. Each neural network layer specifies a tensor operation which receives and outputs tensor, and each layer has trainable parameters. It should be understood that the possible implementations of the NN-based model described here are merely illustrative and therefore should not be construed as limiting the present disclosure in any way.
[0044] As shown in Fig. 10, the method 1000 starts at 1002 where contextual information for the current frame is determined from at least one frame coded before the current frame. By way of example rather than limitation, the current frame is the t-th frame of the video, and the contextual information for the current frame comprises contextual information ct-1 from the previous frame.
[0045] At 1004, the contextual information for the current frame is filtered. For example, the NN-based model may comprise a first NN-based module. The first NN-based module may be configured to filter the contextual information for the current frame, and thus the first NN-based module may also be referred to as “contextual filtering network” , “contextual filtering structure” , or “contextual filtering” for short. This will be descried in detail below.
[0046] At 1006, the conversion is performed based on the filtered contextual information. For example, a motion compensation module in the NN-based model generated the predicted conditional information for the current frame based on the filtered contextual information, and the conversion is performed based on the predicted conditional information. As such, the proposed filtering operation belongs in-loop filtering.
[0047] In some cases, the conversion may include encoding the current frame into the bitstream. In addition, the bitstream may be stored in a non-transitory computer-readable recording medium. Alternatively or additionally, the conversion may include decoding the current frame from the bitstream. It should be understood that the above illustrations and / or examples are described merely for purpose of description. The scope of the present disclosure is not limited in this respect.
[0048] In view of the above, the contextual information for the current frame is filtered and the filtered contextual information is used to performing the conversion. Compared with the conventional solution, the proposed method can advantageously refine the contextual information within the coding loop, and thus error propagation during frame-by-frame coding can be mitigated. Therefore, the coding quality can be improved.
[0049] In some cases, the first NN-based module may comprise at least one residual block, at least one transformer block, at least one activation function, and / or the like. For example, the activation function may be one of the following: Rectified Linear Unit (ReLU) , leaky ReLU, Gaussian Error Linear Unit (GELU) , Exponential Linear Unit (ELU) , Scaled Exponential Linear Unit (SELU) , or the like.
[0050] In some cases, each of the at least one residual block may comprise a predetermined number (such as 1, 2, 3, 4, 5 or the like) of convolution layers. In one example case, the first NN-based module may comprise a single convolutional layer preceding the at least one residual block, as shown in Fig. 8A. Additionally or alternatively, the first NN-based module may comprise a single convolutional layer following the at least one residual block, as shown in Fig. 8A. In Figs. 8A-8B and 11A-11B, convolution parameters are denoted as: number of filters × height of kernel × width of kernel / stride. The number of filters may be equivalent to the number of channels.
[0051] In some cases, the at least one residual block may comprise a plurality of residual blocks, and the plurality of residual blocks may be concatenated. In the example shown in Fig. 8A, the contextual filtering structure includes M residual blocks. The number (i.e., M) of the plurality of residual blocks may be 8, 10, 16, 24, 32, or the like. It should be understood that the specific values recited herein are intended to be examples rather than limiting the scope of the present disclosure.
[0052] In some cases, each of the at least one residual block may comprise a stack of a convolution layer, an activation function, and a convolution layer, as shown in Fig. 11A. Additionally or alternatively, the number of channels (i.e., N in Figs. 8A and 11A) in convolution layers of the first NN-based module may be 16, 32, 64, 96, or the like.
[0053] In some cases, an input to the first NN-based module may comprise a bitrate, a coefficient (e.g., λ) for rate-distortion optimization, at least one reference frame of the current frame, at least one frame coded before the current frame, at least one intra-coded frame of the video, and / or the like. Therefore, the filtering of the contextual information may be performed further based on these additional input.
[0054] In some cases, a training loss for training the first NN-based module with a training video may be determined based on a bitrate of respective frames of the training video and a distortion of respective frames of the training video. For example, the distortion may be measured based on mean square error (MSE) , multi-scale structural similarity index measure (MS-SSIM) , or the like.
[0055] By way of example rather than limitation, the training loss may be determined as follows: where Lcf represents the training loss, Rt represents a bitrate of t-th frame of the training video, Dt represents a distortion of the t-th frame of the training video, λt represents a weighting factor, and T represents the number of frames for determining the training loss. In one example case, the number of frames T may be fixed during the training process. For example, the number of frames T may be equal to the number of frames of a training video, or a predetermined number.
[0056] In another example case, the number of frames T may increase progressively in a training process of the first NN-based module. For example, the first NN-based module may be initially trained on single-frame sequences and then progressively extends to sequences of two frames, three frames, seven frames, and finally, long-sequence training. This may also be referred to as “multi-stage training strategy” .
[0057] In some cases, at 1006, the current frame may be reconstructed based on the filtered contextual information, and the method 1000 further may comprise: adjusting the reconstructed current frame, so as to enhance the reconstructed current frame. As such, the reconstruction enhancement is an out-of-loop operation. This design ensures that the enhanced frame does not interfere with subsequent coding operations, thereby enabling stable optimization and consistent improvements in reconstructed frame quality.
[0058] In some cases, the NN-based model may comprise a second NN-based module. The second NN-based module may be configured to adjust the reconstructed current frame so as to enhance the reconstructed current frame, and thus the second NN-based module may also be referred to as “reconstruction enhancement network” , “reconstruction enhancement structure” , or “reconstruction enhancement” for short.
[0059] In some cases, the second NN-based module may comprise at least one residual block, at least one transformer block, at least one activation function, and / or the like. For example, the activation function may be one of the following: Rectified Linear Unit (ReLU) , leaky ReLU, Gaussian Error Linear Unit (GELU) , Exponential Linear Unit (ELU) , Scaled Exponential Linear Unit (SELU) , or the like.
[0060] In some cases, each of the at least one residual block may comprise a predetermined number (such as 1, 2, 3, 4, 5 or the like) of convolution layers. In one example case, the second NN-based module may comprise a single convolutional layer preceding the at least one residual block, as shown in Fig. 8B. Additionally or alternatively, the second NN-based module may comprise a single convolutional layer following the at least one residual block, as shown in Fig. 8B.
[0061] In some cases, the at least one residual block may comprise a plurality of residual blocks, and the plurality of residual blocks may be concatenated. In the example shown in Fig. 8B, the contextual filtering structure includes M residual blocks. The number (i.e., M) of the plurality of residual blocks may be 8, 10, 16, 24, 32, or the like. It should be understood that the specific values recited herein are intended to be examples rather than limiting the scope of the present disclosure.
[0062] In some cases, each of the at least one residual block may comprise a stack of a convolution layer, an activation function, and a convolution layer, as shown in Fig. 11B. Additionally or alternatively, the number of channels (i.e., N in Figs. 8B and 11B) in convolution layers of the second NN-based module may be 16, 32, 64, 96, or the like.
[0063] In some cases, an input to the second NN-based module may comprise a bitrate, a coefficient (e.g., λ) for rate-distortion optimization, at least one reference frame of the current frame, a reconstructed optical flow of the current frame, contextual information for the current frame, at least one intra-coded frame of the video, and / or the like. Therefore, the adjustment of the reconstructed current frame may be performed further based on these additional input.
[0064] In some cases, a training loss for training the second NN-based module with a training video may be determined based on a distortion between a reconstructed frame of the training video and a result of adjusting the reconstructed frame with the second NN-based module. For example, the distortion may be measured based on MSE, MS-SSIM, or the like.
[0065] In some cases, first information regarding whether to filter the contextual information for the current frame may be determined at an encoder. In addition, the bitstream may comprise an indication (such as a syntax element, a flag, or the like) indicating the first information. Alternatively, the first information may also be determined at a decoder.
[0066] In one example case, the first information may be determined based on a contextual counter and a threshold. The threshold may also be referred to as maximum quality counter mqc. The contextual counter indicates the number of frames, for which contextual information is filtered, before the current frame within a context refresh period. In other words, the contextual counter records the number of frames, to which contextual filtering is applied, before the current frame within a context refresh period. For example, both the contextual counter and contextual information may be reset at beginning of the context refresh period. By way of example, the contextual counter may be reset to a predetermined number, such as 0 or the like. The contextual information may be reset to an initial state, such as a zero tensor or the like.
[0067] In some cases, if the contextual counter is smaller than the threshold and filtering the contextual information for the current frame improves a quality of reconstructed current frame, it may be determined that the contextual information for the current frame is filtered, i.e., contextual filtering is applied to the contextual information for the current frame. if the contextual counter is smaller than the threshold and filtering the contextual information for the current frame does not improve a quality of reconstructed current frame, it may be determined that the contextual information for the current frame is not filtered, i.e., contextual filtering is not applied to the contextual information for the current frame. For example the quality of reconstructed current frame may be measured based on MSE, MS-SSIM, Peak Signal-to-Noise Ratio (PSNR) , or the like.
[0068] In addition, if the contextual counter is larger than the threshold, the first information may be determined based on a progressive rate loss for the current frame. In one example case, if the contextual counter is equal to the threshold and filtering the contextual information for the current frame improves a quality of reconstructed current frame, it may be determined that the contextual information for the current frame is filtered. Alternatively, if the contextual counter is equal to the threshold, the first information may be determined based on the progressive rate loss for the current frame.
[0069] The progressive rate loss may be determined based on at least one of the following: a first bitrate for the current frame without filtering the contextual information for the current frame, a second bitrate for the current frame with the contextual information for the current frame being filtered, a frame index of the current frame, or a total number of frames of the video. By way of example rather than limitation, the progressive rate loss may be determined as follows: where Lpl represents the progressive rate loss, r1 represents the first bitrate, r2 represents the second bitrate, t represents the frame index of the current frame, tfl represents the total number of frames of the video, and pf represents a weighting factor. In one example case, the weighting factor pf may be a positive float number, such as 0.12, 0.14, 0.16, 0.18, 0.20 or the like. In another example case, the weighting factor pf may be indicated in the bitstream.
[0070] In some cases, if the progressive rate loss is smaller than a predetermined value (such as 0 or the like) , it may be determined that the contextual information for the current frame is filtered. If the progressive rate loss is larger than the predetermined value, it may be determined that the contextual information for the current frame is not filtered. In one example case, if the progressive rate loss is equal to the predetermined value, it may be determined that the contextual information for the current frame is filtered. Alternatively, if the progressive rate loss is equal to the predetermined value, it may be determined that the contextual information for the current frame is not filtered.
[0071] In some cases, second information regarding whether to adjust the reconstructed current frame may be determined at an encoder. In addition, the bitstream may comprise an indication (such as a syntax element, a flag, or the like) indicating the second information. Alternatively, the second information may also be determined at a decoder.
[0072] In some cases, the second information may be determined by evaluating whether adjusting the reconstructed current frame improves a quality of the reconstructed current frame. For example, if adjusting the reconstructed current frame improves a quality of the reconstructed current frame, it may be determined that the reconstructed current frame is adjusted. if adjusting the reconstructed current frame does not improve the quality of the reconstructed current frame, it may be determined that the reconstructed current frame is not adjusted. By way of example rather than limitation, the quality of reconstructed current frame may be measured based on MSE, MS-SSIM, PSNR, or the like.
[0073] While the various steps in the flowchart 1000 are presented and described sequentially, one of ordinary skill in the art will appreciate that some or all of the steps may be executed in different orders and some or all of the steps may be executed in parallel. Further, in various cases, one or more of the steps described below may be omitted, repeated, and / or performed in a different order. Accordingly, the specific arrangement of steps shown in Fig. 10 should not be construed as limiting the scope of the case.
[0074] For purpose of illustration, the proposed solution will be discussed in more detail based on a conditional-based neural video compression (NVC) framework. Let xt denote the t-th frame to be encoded. The motion estimation network gme, parameterized by θme, is first employed to estimate the optical flow ft between xt and the previously encoded frame This optical flow ft is subsequently compressed through a series of operations, including a parametric analysis transformation gma (parameterized by θma) , quantization Q, and a parametric synthesis transformation gms (parameterized by θms) , resulting in the reconstructed optical flow ft. Next, the motion compensation network gmc generates the predicted conditional information pt. Specifically, gmc takes the encoded frame the compressed flow and the contextual information ct-1 from the previous frame as inputs. These inputs are processed within the motion compensation network through warping and feature extraction, guided by the parameters θmc. The frame codec is then applied to compress the current frame’s information. This involves a parametric conditional analysis transformation gfa (parameterized by θfa) , followed by quantization Q, and a parametric conditional synthesis transformation gfs (parameterized by θfs) . The contextual information of the current frame, ct, is further processed through a single convolutional layer gconv, parameterized by θc, to produce the reconstructed frame In summary, the entire coding process of conditional-based NVC can be summarized as follows: ct = gfs (Q (gfa (xt, pt; θfa) ) , pt; θfs) , (9)
[0075] Based on the structure of the conditional-based NVC, the above-described contextual filtering Fcon and reconstruction enhancement Frec are introduced to further boost the coding performance, as illustrated in Fig. 9.
[0076] The solution of in-loop contextual filtering will be discussed at first. To address the issue of error propagation in long video sequences, the context refresh is introduced in the coding process, which periodically updates the contextual information to mitigate accumulated errors. Rather than directly utilizing contextual information ct-1 in motion compensation, the contextual information ct-1 is first refined before usage. When contextual filtering is enabled, the motion compensation process in Equation (8) can be reformulated as follows: where gcf represents the contextual filtering network, parameterized by θcf. Unlike existing approaches that directly utilize ct-1, the proposed method incorporates the refined contextual information into the motion compensation process, thereby reducing error propagation and enhancing temporal consistency.
[0077] The training objective of contextual filtering is to enhance the quality of the current frame while minimizing bit rate consumption. This objective can be optimized using the Lagrange multiplier method, formulated as: where R is the rate of the features to be transmitted, and D is the distortion loss, measured as mean squared error (MSE) . A hierarchical quality optimization and long-sequence training is employed to mitigate error propagation. To streamline the optimization process, the above-mentioned multi-stage training strategy may be used.
[0078] The proposed contextual filtering structure can advantageously improve coding performance. In addition, when contextual filtering is applied, it serves as a substitute for context refresh, resulting in higher quality and a lower bit rate, not only for the current frame but also for subsequent frames. Furthermore, contextual filtering improves the overall quality with only a marginal increase in bit rate. Given the rate-distortion trade-off in the current frame, enabling contextual filtering remains beneficial for enhancing coding efficiency.
[0079] Instead of directly outputting the current coded frame out-of-loop reconstruction enhancement further improves coding performance beyond the coding loop by employing a reconstruction enhancement network gre, formulated as:
[0080] Since the enhanced frame is only utilized to improve the quality of the current frame and does not influence the subsequent coding process, the optimization objective of reconstruction enhancement is straightforward, given by: where D represents the distortion loss, computed as mean squared error (MSE) in our method. A training augmentation strategy that incorporates random frame selection and variable-rate training may be employed. During each training iteration, randomly sample a rate point and a frame at different temporal positions, allowing the network to learn how distortions evolve over time and across different compression rates, thereby optimizing enhancement quality under diverse coding conditions.
[0081] Figs. 11A and 11B illustrate the architecture of the contextual filtering and reconstruction enhancement modules, which share a unified design. The input is first projected into the feature space via a 3×3 convolutional layer with N filters. This is followed by M residual blocks, each consisting of two 3×3 convolutional layers with intermediate ReLU activations, to refine the feature representations. Finally, a concluding 3×3 convolutional layer, symmetric to the initial embedding layer, maps the features back to the target space, producing either the enhanced contextual representation or the improved reconstructed frame. It should be understood that the possible implementations of the contextual filtering and reconstruction enhancement modules described here are merely illustrative and therefore should not be construed as limiting the present disclosure in any way. For example, the architecture of the contextual filtering module may also be different from the reconstruction enhancement module.
[0082] The design of these modules is guided by three key principles: simplicity, to ensure computational efficiency; scalability, to support diverse application scenarios; and stability, during both training and inference. To achieve these goals, the widely used residual structure is used as the fundamental building block, avoiding more complex architectures such as transformers or attention modules, which typically incur higher computational overhead.
[0083] To enable smooth transitions between the pixel / feature domains and the target representation space, shallow convolutional layers are introduced before and after the residual stack. This simple yet effective design facilitates better representation learning while maintaining efficiency. Moreover, the model’s complexity can be flexibly adjusted by tuning the hyperparameters M and N, making it well-suited for deployment under varying system constraints.
[0084] To ensure that contextual filtering and reconstruction enhancement contribute positively to the overall compression performance, an adaptive coding decision mechanism is proposed. The detailed procedure is presented in Algorithm 1 in the above Table 2.
[0085] While contextual filtering improves reconstruction quality, it may also increase the bitstream size, potentially leading to suboptimal rate-distortion (R-D) trade-offs. To address this, the proposed strategy dynamically determines whether to enable filtering on a per-frame basis, aiming to balance distortion and rate across the entire sequence. This decision process is guided by two observations as follows: (i) Intra-Period Reference Dependencies. Following the context refresh scheme, the encoding sequence is divided into fixed-length periods (e.g., 32 frames or the like) , with each period starting with a context refresh. Within a period, early frames act as reference points for subsequent frames. Enhancing these early frames-even at the cost of higher bitrates-often leads to improved overall performance due to the propagation of higher-quality references. (ii) Inter-Period Global Dependencies. Beyond a single refresh period, earlier frames in the entire sequence influence a larger number of subsequent frames. As encoding proceeds sequentially, the quality of earlier frames has a compounding effect on later predictions. Therefore, investing bitrate in these frames can yield long-term benefits, while frames near the end of the sequence are less impactful and may not warrant additional filtering.
[0086] Based on these insights, there is provided an adaptive decision strategy that jointly considers both local and global reference relationships. Specifically, the adaptive decision strategy introduces a contextual counter con that tracks the number of frames using contextual filtering within a refresh period. A maximum quality counter mqc is used in conjunction with con: if con<MQc, filtering is enabled for frames that yield any quality gain. This guarantees that at least mqc frames benefit from filtering within each period, provided such filtering is beneficial.
[0087] For the remaining frames, a progressive rate-loss strategy is employed to determine whether filtering should be applied. The decision criterion is defined as: where r1 and r2 are the bitrates without and with contextual filtering, respectively, t is the index of the current frame, tfl is the total frame length, and pf is a progressive factor that controls the maximum allowable rate increase. The contextual filtering is enabled only if Lpl<0.
[0088] In addition, adaptive decision-making is also applied to the reconstruction enhancement module. Since this module does not affect the bitstream or subsequent frames, its application is determined solely by whether it improves the current frame’s reconstruction quality.
[0089] To sum up, there is provided a contextual filtering approach that enhances coding performance by refining contextual information within the coding loop. Additionally, a reconstruction enhancement module is introduced to improve reconstruction quality further. To ensure stable performance, there is provided an adaptive coding decision mechanism that dynamically determines when to apply these modules, preventing potential degradation while maintaining optimal rate-distortion trade-offs.
[0090] According to further cases of the present disclosure, a non-transitory computer-readable recording medium is provided. The non-transitory computer-readable recording medium stores a bitstream of a video which is generated by a method performed by an apparatus for video processing. The method comprises: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; and generating the bitstream based on the filtered contextual information with a neural network (NN) -based model.
[0091] According to still further cases of the present disclosure, a method for storing bitstream of a video is provided. The method comprises: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; generating the bitstream based on the filtered contextual information with a neural network (NN) -based model; and storing the bitstream in a non-transitory computer-readable recording medium.
[0092] Implementations of the present disclosure can be described in view of the following clauses, the features of which can be combined in any reasonable manner.
[0093] Clause 1. A method for video processing, comprising: determining, for a conversion between a current frame of a video and a bitstream of the video with a neural network (NN) -based model, contextual information for the current frame from at least one frame coded before the current frame; filtering the contextual information for the current frame; and performing the conversion based on the filtered contextual information.
[0094] Clause 2. The method of clause 1, wherein the NN-based model comprises a first NN-based module, and the first NN-based module is configured to filter the contextual information for the current frame.
[0095] Clause 3. The method of clause 2, wherein the first NN-based module comprises at least one of the following: at least one residual block, at least one transformer block, or at least one activation function.
[0096] Clause 4. The method of clause 3, wherein each of the at least one residual block comprises a predetermined number of convolution layers.
[0097] Clause 5. The method of any of clauses 3-4, wherein the first NN-based module comprises a single convolutional layer preceding the at least one residual block.
[0098] Clause 6. The method of any of clauses 3-5, wherein the first NN-based module comprises a single convolutional layer following the at least one residual block.
[0099] Clause 7. The method of any of clauses 3-6, wherein the at least one residual block comprises a plurality of residual blocks, and the plurality of residual blocks are concatenated.
[0100] Clause 8. The method of clause 7, wherein the number of the plurality of residual blocks is one of the following: 8, 10, 16, 24, or 32.
[0101] Clause 9. The method of any of clauses 3-8, wherein each of the at least one residual block comprises a stack of a convolution layer, an activation function, and a convolution layer.
[0102] Clause 10. The method of any of clauses 2-9, wherein the number of channels in convolution layers of the first NN-based module is one of the following: 16, 32, 64, or 96.
[0103] Clause 11. The method of any of clauses 2-10, wherein an input to the first NN-based module comprises at least one of the following: a bitrate, a coefficient for rate-distortion optimization, at least one reference frame of the current frame, at least one frame coded before the current frame, or at least one intra-coded frame of the video.
[0104] Clause 12. The method of any of clauses 2-11, wherein a training loss for training the first NN-based module with a training video is determined based on a bitrate of respective frames of the training video and a distortion of respective frames of the training video.
[0105] Clause 13. The method of clause 12, wherein the distortion is measured based on mean square error (MSE) or multi-scale structural similarity index measure (MS-SSIM) .
[0106] Clause 14. The method of any of clauses 12-13, wherein the training loss is determined as follows: wherein Lcf represents the training loss, Rt represents a bitrate of t-th frame of the training video, Dt represents a distortion of the t-th frame of the training video, λt represents a weighting factor, and T represents the number of frames for determining the training loss.
[0107] Clause 15. The method of clause 14, wherein the number of frames T increases progressively in a training process of the first NN-based module.
[0108] Clause 16. The method of any of clauses 1-15, wherein performing the conversion comprises: reconstructing the current frame based on the filtered contextual information, and the method further comprises: adjusting the reconstructed current frame.
[0109] Clause 17. The method of clause 16, wherein the NN-based model comprises a second NN-based module, and the second NN-based module is configured to adjust the reconstructed current frame.
[0110] Clause 18. The method of clause 17, wherein the second NN-based module comprises at least one of the following: at least one residual block, at least one transformer block, or at least one activation function.
[0111] Clause 19. The method of clause 18, wherein each of the at least one residual block comprises a predetermined number of convolution layers.
[0112] Clause 20. The method of any of clauses 18-19, wherein the second NN-based module comprises a single convolutional layer preceding the at least one residual block.
[0113] Clause 21. The method of any of clauses 18-20, wherein the second NN-based module comprises a single convolutional layer following the at least one residual block.
[0114] Clause 22. The method of any of clauses 18-21, wherein the at least one residual block comprises a plurality of residual blocks, and the plurality of residual blocks are concatenated.
[0115] Clause 23. The method of clause 22, wherein the number of the plurality of residual blocks is one of the following: 8, 10, 16, 24, or 32.
[0116] Clause 24. The method of any of clauses 18-23, wherein each of the at least one residual block comprises a stack of a convolution layer, an activation function, and a convolution layer.
[0117] Clause 25. The method of any of clauses 17-24, wherein the number of channels in convolution layers of the second NN-based module is one of the following: 16, 32, 64, or 96.
[0118] Clause 26. The method of any of clauses 17-25, wherein an input to the second NN-based module comprises at least one of the following: a bitrate, a coefficient for rate-distortion optimization, at least one reference frame of the current frame, a reconstructed optical flow of the current frame, contextual information for the current frame, or at least one intra-coded frame of the video.
[0119] Clause 27. The method of any of clauses 17-26, wherein a training loss for training the second NN-based module with a training video is determined based on a distortion between a reconstructed frame of the training video and a result of adjusting the reconstructed frame with the second NN-based module.
[0120] Clause 28. The method of clause 27, wherein the distortion is measured based on MSE or MS-SSIM.
[0121] Clause 29. The method of any of clauses 1-28, wherein first information regarding whether to filter the contextual information for the current frame is determined at an encoder.
[0122] Clause 30. The method of clause 29, wherein the first information is determined based on a contextual counter and a threshold, the contextual counter indicating the number of frames, for which contextual information is filtered, before the current frame within a context refresh period.
[0123] Clause 31. The method of clause 30, wherein the contextual counter and contextual information are reset at beginning of the context refresh period.
[0124] Clause 32. The method of any of clauses 30-31, wherein in accordance with that the contextual counter is smaller than the threshold and filtering the contextual information for the current frame improves a quality of reconstructed current frame, it is determined that the contextual information for the current frame is filtered.
[0125] Clause 33. The method of any of clauses 30-32, wherein in accordance with that the contextual counter is larger than the threshold, the first information is determined based on a progressive rate loss for the current frame, the progressive rate loss being determined based on at least one of the following: a first bitrate for the current frame without filtering the contextual information for the current frame, a second bitrate for the current frame with the contextual information for the current frame being filtered, a frame index of the current frame, or a total number of frames of the video.
[0126] Clause 34. The method of clause 33, wherein the progressive rate loss is determined as follows: wherein Lpl represents the progressive rate loss, r1 represents the first bitrate, r2 represents the second bitrate, t represents the frame index of the current frame, tfl represents the total number of frames of the video, and pf represents a weighting factor.
[0127] Clause 35. The method of clause 34, wherein the weighting factor pf is a positive float number.
[0128] Clause 36. The method of any of clauses 34-35, wherein the weighting factor pf is indicated in the bitstream.
[0129] Clause 37. The method of any of clauses 33-36, wherein in accordance with that the progressive rate loss is smaller than a predetermined value, it is determined that the contextual information for the current frame is filtered.
[0130] Clause 38. The method of clause 37, wherein the predetermined value is 0.
[0131] Clause 39. The method of any of clauses 30-38, wherein the bitstream comprises an indication indicating the first information.
[0132] Clause 40. The method of any of clauses 16-39, wherein second information regarding whether to adjust the reconstructed current frame is determined at an encoder.
[0133] Clause 41. The method of clause 40, wherein the second information is determined by evaluating whether adjusting the reconstructed current frame improves a quality of the reconstructed current frame.
[0134] Clause 42. The method of any of clauses 1-41, wherein the conversion comprises encoding the current video block into the bitstream.
[0135] Clause 43. The method of any of clauses 1-41, wherein the conversion comprises decoding the current video block from the bitstream.
[0136] Clause 44. The method of any of clauses 1-41, wherein the conversion comprises generating the bitstream from the video, and the method further comprises storing the bitstream in a non-transitory computer-readable recording medium.
[0137] Clause 45. An apparatus for video processing comprising a processor and a non-transitory memory with instructions thereon, wherein the instructions upon execution by the processor, cause the processor to perform a method in accordance with any of clauses 1-44.
[0138] Clause 46. A non-transitory computer-readable storage medium storing instructions that cause a processor to perform a method in accordance with any of clauses 1-44.
[0139] Clause 47. A non-transitory computer-readable recording medium storing a bitstream of a video which is generated by a method performed by an apparatus for video processing, wherein the method comprises: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; and generating the bitstream based on the filtered contextual information with a neural network (NN) -based model.
[0140] Clause 48. A method for storing a bitstream of a video, comprising: determining contextual information for a current frame of the video from at least one frame coded before the current frame; filtering the contextual information for the current frame; generating the bitstream based on the filtered contextual information with a neural network (NN) -based model; and storing the bitstream in a non-transitory computer-readable recording medium.
[0141] Clause 49. A computer program for performing a method in accordance with any of clauses 1-44 when the computer program runs on a computer.
[0142] Clause 50. A computer program product tangibly stored in a computer storage medium and comprising computer-executable instructions that, when executed by a device, cause the device to perform a method in accordance with any of clauses 1-44. Example Device
[0143] Fig. 12 illustrates a block diagram of a computing device 1200 in which various cases of the present disclosure can be implemented. The computing device 1200 may be implemented as or included in the source device 110 (or the video encoder 114 or 200) or the destination device 120 (or the video decoder 124 or 300) .
[0144] It would be appreciated that the computing device 1200 shown in Fig. 12 is merely for purpose of illustration, without suggesting any limitation to the functions and scopes of the cases of the present disclosure in any manner.
[0145] As shown in Fig. 12, the computing device 1200 includes a general-purpose computing device 1200. The computing device 1200 may at least comprise one or more processors or processing units 1210, a memory 1220, a storage unit 1230, one or more communication units 1240, one or more input devices 1250, and one or more output devices 1260.
[0146] In some cases, the computing device 1200 may be implemented as any user terminal or server terminal having the computing capability. The server terminal may be a server, a large-scale computing device or the like that is provided by a service provider. The user terminal may for example be any type of mobile terminal, fixed terminal, or portable terminal, including a mobile phone, station, unit, device, multimedia computer, multimedia tablet, Internet node, communicator, desktop computer, laptop computer, notebook computer, netbook computer, tablet computer, personal communication system (PCS) device, personal navigation device, personal digital assistant (PDA) , audio / video player, digital camera / video camera, positioning device, television receiver, radio broadcast receiver, E-book device, gaming device, or any combination thereof, including the accessories and peripherals of these devices, or any combination thereof. It would be contemplated that the computing device 1200 can support any type of interface to a user (such as “wearable” circuitry and the like) .
[0147] The processing unit 1210 may be a physical or virtual processor and can implement various processes based on programs stored in the memory 1220. In a multi-processor system, multiple processing units execute computer executable instructions in parallel so as to improve the parallel processing capability of the computing device 1200. The processing unit 1210 may also be referred to as a central processing unit (CPU) , a microprocessor, a controller or a microcontroller.
[0148] The computing device 1200 typically includes various computer storage medium. Such medium can be any medium accessible by the computing device 1200, including, but not limited to, volatile and non-volatile medium, or detachable and non-detachable medium. The memory 1220 can be a volatile memory (for example, a register, cache, Random Access Memory (RAM) ) , a non-volatile memory (such as a Read-Only Memory (ROM) , Electrically Erasable Programmable Read-Only Memory (EEPROM) , or a flash memory) , or any combination thereof. The storage unit 1230 may be any detachable or non-detachable medium and may include a machine-readable medium such as a memory, flash memory drive, magnetic disk or another other media, which can be used for storing information and / or data and can be accessed in the computing device 1200.
[0149] The computing device 1200 may further include additional detachable / non-detachable, volatile / non-volatile memory medium. Although not shown in Fig. 12, it is possible to provide a magnetic disk drive for reading from and / or writing into a detachable and non-volatile magnetic disk and an optical disk drive for reading from and / or writing into a detachable non-volatile optical disk. In such cases, each drive may be connected to a bus (not shown) via one or more data medium interfaces.
[0150] The communication unit 1240 communicates with a further computing device via the communication medium. In addition, the functions of the components in the computing device 1200 can be implemented by a single computing cluster or multiple computing machines that can communicate via communication connections. Therefore, the computing device 1200 can operate in a networked environment using a logical connection with one or more other servers, networked personal computers (PCs) or further general network nodes.
[0151] The input device 1250 may be one or more of a variety of input devices, such as a mouse, keyboard, tracking ball, voice-input device, and the like. The output device 1260 may be one or more of a variety of output devices, such as a display, loudspeaker, printer, and the like. By means of the communication unit 1240, the computing device 1200 can further communicate with one or more external devices (not shown) such as the storage devices and display device, with one or more devices enabling the user to interact with the computing device 1200, or any devices (such as a network card, a modem and the like) enabling the computing device 1200 to communicate with one or more other computing devices, if required. Such communication can be performed via input / output (I / O) interfaces (not shown) .
[0152] In some cases, instead of being integrated in a single device, some or all components of the computing device 1200 may also be arranged in cloud computing architecture. In the cloud computing architecture, the components may be provided remotely and work together to implement the functionalities described in the present disclosure. In some cases, cloud computing provides computing, software, data access and storage service, which will not require end users to be aware of the physical locations or configurations of the systems or hardware providing these services. In various cases, the cloud computing provides the services via a wide area network (such as Internet) using suitable protocols. For example, a cloud computing provider provides applications over the wide area network, which can be accessed through a web browser or any other computing components. The software or components of the cloud computing architecture and corresponding data may be stored on a server at a remote position. The computing resources in the cloud computing environment may be merged or distributed at locations in a remote data center. Cloud computing infrastructures may provide the services through a shared data center, though they behave as a single access point for the users. Therefore, the cloud computing architectures may be used to provide the components and functionalities described herein from a service provider at a remote location. Alternatively, they may be provided from a conventional server or installed directly or otherwise on a client device.
[0153] The computing device 1200 may be used to implement video encoding / decoding in cases of the present disclosure. The memory 1220 may include one or more video coding modules 1225 having one or more program instructions. These modules are accessible and executable by the processing unit 1210 to perform the functionalities of the various cases described herein.
[0154] In the example cases of performing video encoding, the input device 1250 may receive video data as an input 1270 to be encoded. The video data may be processed, for example, by the video coding module 1225, to generate an encoded bitstream. The encoded bitstream may be provided via the output device 1260 as an output 1280.
[0155] In the example cases of performing video decoding, the input device 1250 may receive an encoded bitstream as the input 1270. The encoded bitstream may be processed, for example, by the video coding module 1225, to generate decoded video data. The decoded video data may be provided via the output device 1260 as the output 1280.
[0156] While this disclosure has been particularly shown and described with references to example cases thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present application as defined by the appended claims. Such variations are intended to be covered by the scope of this present application. As such, the foregoing description of cases of the present application is not intended to be limiting.
Claims
1.A method for video processing, comprising:determining, for a conversion between a current frame of a video and a bitstream of the video with a neural network (NN) -based model, contextual information for the current frame from at least one frame coded before the current frame;filtering the contextual information for the current frame; andperforming the conversion based on the filtered contextual information.2.The method of claim 1, wherein the NN-based model comprises a first NN-based module, and the first NN-based module is configured to filter the contextual information for the current frame.3.The method of claim 2, wherein the first NN-based module comprises at least one of the following:at least one residual block,at least one transformer block, orat least one activation function.4.The method of claim 3, wherein each of the at least one residual block comprises a predetermined number of convolution layers.5.The method of any of claims 3-4, wherein the first NN-based module comprises a single convolutional layer preceding the at least one residual block.6.The method of any of claims 3-5, wherein the first NN-based module comprises a single convolutional layer following the at least one residual block.7.The method of any of claims 3-6, wherein the at least one residual block comprises a plurality of residual blocks, and the plurality of residual blocks are concatenated.8.The method of claim 7, wherein the number of the plurality of residual blocks is one of the following: 8, 10, 16, 24, or 32.9.The method of any of claims 3-8, wherein each of the at least one residual block comprises a stack of a convolution layer, an activation function, and a convolution layer.10.The method of any of claims 2-9, wherein the number of channels in convolution layers of the first NN-based module is one of the following: 16, 32, 64, or 96.11.The method of any of claims 2-10, wherein an input to the first NN-based module comprises at least one of the following:a bitrate,a coefficient for rate-distortion optimization,at least one reference frame of the current frame,at least one frame coded before the current frame, orat least one intra-coded frame of the video.12.The method of any of claims 2-11, wherein a training loss for training the first NN-based module with a training video is determined based on a bitrate of respective frames of the training video and a distortion of respective frames of the training video.13.The method of claim 12, wherein the distortion is measured based on mean square error (MSE) or multi-scale structural similarity index measure (MS-SSIM) .14.The method of any of claims 12-13, wherein the training loss is determined as follows: wherein Lcf represents the training loss, Rt represents a bitrate of t-th frame of the training video, Dt represents a distortion of the t-th frame of the training video, λt represents a weighting factor, and T represents the number of frames for determining the training loss.15.The method of claim 14, wherein the number of frames T increases progressively in a training process of the first NN-based module.16.The method of any of claims 1-15, wherein performing the conversion comprises: reconstructing the current frame based on the filtered contextual information, andthe method further comprises: adjusting the reconstructed current frame.17.The method of claim 16, wherein the NN-based model comprises a second NN-based module, and the second NN-based module is configured to adjust the reconstructed current frame.18.The method of claim 17, wherein the second NN-based module comprises at least one of the following:at least one residual block,at least one transformer block, orat least one activation function.19.The method of claim 18, wherein each of the at least one residual block comprises a predetermined number of convolution layers.20.The method of any of claims 18-19, wherein the second NN-based module comprises a single convolutional layer preceding the at least one residual block.21.The method of any of claims 18-20, wherein the second NN-based module comprises a single convolutional layer following the at least one residual block.22.The method of any of claims 18-21, wherein the at least one residual block comprises a plurality of residual blocks, and the plurality of residual blocks are concatenated.23.The method of claim 22, wherein the number of the plurality of residual blocks is one of the following: 8, 10, 16, 24, or 32.24.The method of any of claims 18-23, wherein each of the at least one residual block comprises a stack of a convolution layer, an activation function, and a convolution layer.25.The method of any of claims 17-24, wherein the number of channels in convolution layers of the second NN-based module is one of the following: 16, 32, 64, or 96.26.The method of any of claims 17-25, wherein an input to the second NN-based module comprises at least one of the following:a bitrate,a coefficient for rate-distortion optimization,at least one reference frame of the current frame,a reconstructed optical flow of the current frame,contextual information for the current frame, orat least one intra-coded frame of the video.27.The method of any of claims 17-26, wherein a training loss for training the second NN-based module with a training video is determined based on a distortion between a reconstructed frame of the training video and a result of adjusting the reconstructed frame with the second NN-based module.28.The method of claim 27, wherein the distortion is measured based on MSE or MS-SSIM.29.The method of any of claims 1-28, wherein first information regarding whether to filter the contextual information for the current frame is determined at an encoder.30.The method of claim 29, wherein the first information is determined based on a contextual counter and a threshold, the contextual counter indicating the number of frames, for which contextual information is filtered, before the current frame within a context refresh period.31.The method of claim 30, wherein the contextual counter and contextual information are reset at beginning of the context refresh period.32.The method of any of claims 30-31, wherein in accordance with that the contextual counter is smaller than the threshold and filtering the contextual information for the current frame improves a quality of reconstructed current frame, it is determined that the contextual information for the current frame is filtered.33.The method of any of claims 30-32, wherein in accordance with that the contextual counter is larger than the threshold, the first information is determined based on a progressive rate loss for the current frame, the progressive rate loss being determined based on at least one of the following:a first bitrate for the current frame without filtering the contextual information for the current frame,a second bitrate for the current frame with the contextual information for the current frame being filtered,a frame index of the current frame, ora total number of frames of the video.34.The method of claim 33, wherein the progressive rate loss is determined as follows: wherein Lpl represents the progressive rate loss, r1 represents the first bitrate, r2 represents the second bitrate, t represents the frame index of the current frame, tfl represents the total number of frames of the video, and pf represents a weighting factor.35.The method of claim 34, wherein the weighting factor pf is a positive float number.36.The method of any of claims 34-35, wherein the weighting factor pf is indicated in the bitstream.37.The method of any of claims 33-36, wherein in accordance with that the progressive rate loss is smaller than a predetermined value, it is determined that the contextual information for the current frame is filtered.38.The method of claim 37, wherein the predetermined value is 0.39.The method of any of claims 30-38, wherein the bitstream comprises an indication indicating the first information.40.The method of any of claims 16-39, wherein second information regarding whether to adjust the reconstructed current frame is determined at an encoder.41.The method of claim 40, wherein the second information is determined by evaluating whether adjusting the reconstructed current frame improves a quality of the reconstructed current frame.42.The method of any of claims 1-41, wherein the conversion comprises encoding the current video block into the bitstream.43.The method of any of claims 1-41, wherein the conversion comprises decoding the current video block from the bitstream.44.The method of any of claims 1-41, wherein the conversion comprises generating the bitstream from the video, andthe method further comprises storing the bitstream in a non-transitory computer-readable recording medium.45.An apparatus for video processing comprising a processor and a non-transitory memory with instructions thereon, wherein the instructions upon execution by the processor, cause the processor to perform a method in accordance with any of claims 1-44.46.A non-transitory computer-readable storage medium storing instructions that cause a processor to perform a method in accordance with any of claims 1-44.47.A non-transitory computer-readable recording medium storing a bitstream of a video which is generated by a method performed by an apparatus for video processing, wherein the method comprises:determining contextual information for a current frame of the video from at least one frame coded before the current frame;filtering the contextual information for the current frame; andgenerating the bitstream based on the filtered contextual information with a neural network (NN) -based model.48.A method for storing a bitstream of a video, comprising:determining contextual information for a current frame of the video from at least one frame coded before the current frame;filtering the contextual information for the current frame;generating the bitstream based on the filtered contextual information with a neural network (NN) -based model; andstoring the bitstream in a non-transitory computer-readable recording medium.49.A computer program for performing a method in accordance with any of claims 1-44 when the computer program runs on a computer.50.A computer program product tangibly stored in a computer storage medium and comprising computer-executable instructions that, when executed by a device, cause the device to perform a method in accordance with any of claims 1-44.