An image processing device and method that enable suppression of an increase in the amount of coding of a scaling list. The image processing device sets a coefficient located at the beginning of a quantization matrix by adding a replacement difference coefficient that is a difference between a replacement coefficient used to replace a coefficient located at the beginning of the quantization matrix and the coefficient located at the beginning of the quantization matrix to the coefficient located at the beginning of the quantization matrix; up-converts the set quantization matrix; and dequantizes quantized data using an up-converted quantization matrix in which a coefficient located at the beginning of the up-converted quantization matrix has been replaced with the replacement coefficient. The device and method can be applied to an image processing device.

Systems, apparatuses, and methods for utilizing different modulation coding schemes (MCSs) for different components of a video stream are disclosed. A system includes a transmitter sending a video stream over a wireless link to a receiver. The transmitter splits the video stream into low, medium, and high quality components, and then the transmitter modulates the different components using different MCS's. For example, the transmitter modulates the low quality component using a lower, robust MCS level to increase the likelihood that this component is received. Also, the medium quality component is modulated using a medium MCS level and the high frequency component is modulated using a higher MCS level. If only the low quality component is received by the receiver, then the receiver reconstructs and displays a low quality video frame from this component, which avoids a glitch in the display of the video stream.

Systems, apparatuses, and methods for utilizing different modulation coding schemes (MCSs) for different components of a video stream are disclosed. A system includes a transmitter sending a video stream over a wireless link to a receiver. The transmitter splits the video stream into low, medium, and high quality components, and then the transmitter modulates the different components using different MCS's. For example, the transmitter modulates the low quality component using a lower, robust MCS level to increase the likelihood that this component is received. Also, the medium quality component is modulated using a medium MCS level and the high frequency component is modulated using a higher MCS level. If only the low quality component is received by the receiver, then the receiver reconstructs and displays a low quality video frame from this component, which avoids a glitch in the display of the video stream.

A method of machine learning model development includes building an autoencoder including an encoder trained to map an input into a latent representation, and a decoder trained to map the latent representation to a reconstruction of the input. The method includes building an artificial neural network classifier including the encoder, and a classification layer partially trained to perform a classification in which a class to which the input belongs is predicted based on the latent representation. Neural network inversion is applied to the classification layer to find inverted latent representations within a decision boundary between classes in which a result of the classification is ambiguous, and inverted inputs are obtained from the inverted latent representations. Each inverted input is labeled with a class that is its ground truth, and thereby producing added training data for the classification, and the classification layer is further trained using the added training data.

A method of machine learning model development includes building an autoencoder including an encoder trained to map an input into a latent representation, and a decoder trained to map the latent representation to a reconstruction of the input. The method includes building an artificial neural network classifier including the encoder, and a classification layer partially trained to perform a classification in which a class to which the input belongs is predicted based on the latent representation. Neural network inversion is applied to the classification layer to find inverted latent representations within a decision boundary between classes in which a result of the classification is ambiguous, and inverted inputs are obtained from the inverted latent representations. Each inverted input is labeled with a class that is its ground truth, and thereby producing added training data for the classification, and the classification layer is further trained using the added training data.

Video processing methods and apparatuses for processing a current block in a current picture include receiving input data of the current block, determining a reference picture, determining whether picture sizes of the current and reference pictures are different, determining whether horizontal wraparound motion compensation is enabled for predicting the current block, performing motion compensation for the current block to obtain a reference block from the reference picture, and encoding or decoding the current block according to the reference block. Horizontal wraparound motion compensation is disabled when the picture sizes of the current and reference pictures are different.

Video processing methods and apparatuses for processing a current block in a current picture include receiving input data of the current block, determining a reference picture, determining whether picture sizes of the current and reference pictures are different, determining whether horizontal wraparound motion compensation is enabled for predicting the current block, performing motion compensation for the current block to obtain a reference block from the reference picture, and encoding or decoding the current block according to the reference block. Horizontal wraparound motion compensation is disabled when the picture sizes of the current and reference pictures are different.

The present invention is concerned with methods, encoders, decoders and data streams for coding pictures, and in particular a consecutive sequence of pictures, Some embodiments may exploit the so-called Gradual Decoder Refresh—GDR—coding scheme for coding the pictures. Some embodiments may suggest Scalable Coding and Gradual Decoder Refresh improvements.

The present invention is concerned with methods, encoders, decoders and data streams for coding pictures, and in particular a consecutive sequence of pictures, Some embodiments may exploit the so-called Gradual Decoder Refresh—GDR—coding scheme for coding the pictures. Some embodiments may suggest Scalable Coding and Gradual Decoder Refresh improvements.

An image decoding method performed by a decoding device according to the present document comprises the steps of: obtaining image information; and updating a decoded picture buffer (DPB) on the basis of the image information, wherein the step of obtaining the image information comprises the steps of: obtaining an output layer set (OLS) DPB parameter index for a target OLS; and obtaining DPB parameter information relating to the target OLS on the basis of the OLS DPB parameter index.