73 results about "Block transform" patented technology

Some aspects of the disclosure provide a method of video decoding. In some examples, values of one or more target quantization coefficients in a target region of quantization coefficients of a current block are obtained. A sub-block transform (SBT) mode of the current block is derived based on the values of the one or more target quantization coefficients in the target region. The current block is reconstructed based on the SBT mode of the current block. Apparatus and non-transitory computer-readable storage medium counterpart embodiments are also contemplated.

Some embodiments of a method can include obtaining a point cloud, which can include a first set of information describing geometry of the point cloud and a second set of information describing attributes of the point cloud; performing geometry encoding of the first set of information to generate a geometry bitstream; performing a two-stage attribute compression process to generate an attribute bitstream, wherein a first stage of the two-stage attribute compression process includes performing a block transform on each node of a set of nodes of the point cloud, and wherein a second stage of the two-stage attribute compression process includes performing hierarchical encoding on the set of nodes of the point cloud, and outputting an output bitstream including the geometry bitstream and the attribute bitstream.

A method and apparatus are provided for video encoding and decoding. Regarding a method for encoding, the method includes encoding a coding unit using a sub-block transform including a plurality of transform units. Encoding the coding unit includes defining a residual block for a first transform unit of the plurality of transform units. The first transform unit extends across a medial portion of the coding unit to at least one of a first pair of opposed edges of the coding unit while being spaced by transform blocks of second and third transform units of the plurality of transform units from a second pair of opposed edges of the coding unit. The method also includes causing storage and / or transmission of motion information, a prediction block identified by the motion information to be associated with the coding unit, and information regarding the residual block for the first transform unit.

An example method of video decoding includes receiving a video bitstream that includes multiple blocks, including a current block. The method also includes identifying a partial region of the current block, where residual data for the current block outside of the partial region is zero, and reconstructing the current block by applying a subblock transform to the partial region of the current block. Instructions for the example method may be stored in a computer system or storage medium.

There is a problem in that implicit MTS performance is lost in a case that the implict MTS is combined with secondary transform. The present invention provides an image decoding apparatus that can more preferably apply transform by MTS and secondary transform. A video decoding apparatus includes: a second transformer configured to apply transform using a transform matrix to the transform coefficient to modify the transform coefficient in a case that secondary transform is enabled; a first transformer configured to apply separate transform including vertical transform and horizontal transform to the transform coefficient; and an implicit transform configuration unit configured to disable implicit transform in a case that the secondary transform is enabled, an intra subpartition mode is not used, and subblock transform is not used, and configured to derive a horizontal transform type according to a width of a target TU and derive a vertical transform type according to a height of the target TU in a case that the implicit transform is enabled. The first transformer performs transform according to the vertical transform type, and transform according to the horizontal transform type.

The invention provides a video encoding and decoding method and a related device. The video encoding and decoding method comprises: receiving input data related to a current block in a current picture, the input data comprising residual data of the current block for encoding at an encoder end or transform coefficients of the current block for decoding at a decoder end, the prediction data of the current block is generated by applying intra block copying or intra template matching prediction; applying a target transform pattern to the current block to derive a final transform coefficient at the encoder side or to derive reconstructed residual data at the decoder side, where the target transform pattern includes a sub-block transform or a partial transform, and wherein the sub-block transform divides the current block into a plurality of sub-blocks and applies transform processing to one or more sub-blocks that are target portions of the current block, or the partial transform applies transform processing only to the target portions of the current block; and providing the final transform coefficient at the encoder side or the reconstructed residual data at the decoder side. According to the video coding and decoding method and the related device, the coding and decoding efficiency can be improved.

The invention provides a multi-head attention processing method, a related device and a medium, and the method comprises the steps: dividing an input matrix into a first number of sub-blocks, and carrying out the first linear transformation through a first number of processing units, and obtaining a first number of first transformed sub-blocks; performing sub-block transformation on a first number of first transformed sub-blocks corresponding to the first number of processing units to obtain a second transformed sub-block corresponding to each processing head, and inputting the second transformed sub-block into a processing head corresponding to the second transformed sub-block for attention calculation to obtain a first attention result sub-block; and performing sub-block inverse transformation on the first attention result sub-blocks corresponding to the processing heads to obtain second attention result sub-blocks corresponding to the processing units, and generating a multi-head attention result matrix based on the second attention result sub-blocks. The multi-head attention processing efficiency can be improved, and the multi-head attention processing method and device can be applied to scenes such as artificial intelligence assistants and machine translation.

The present application belongs to the field of computer vision and image processing, aiming at the double bottleneck of "semantic overfitting" and "space-frequency feature mislocation" existing in the existing deep fake detection, the present application proposes a deep fake face detection method based on dual domain. The present application constructs a dual domain complementary architecture of dual domain deep fake detection network DDSF-Net based on semantic suppression and space-frequency alignment, the micro mark difference module in the network significantly enhances the residual subtle fake artifacts between pixels by eliminating redundant face attributes. And the space-frequency mapping module uses the block transformation strategy to adaptively mine the abnormal distribution of frequency domain, realizes the multi-dimensional enhancement of the fake features. At the same time, through the complementary fusion of dual domain features, the detection accuracy and generalization are effectively balanced. Extensive experiments on multiple benchmark datasets show that the performance of the method in the same dataset and cross dataset scenarios is significantly better than that of the existing advanced technology.

The present invention relates to a device (e.g., a decoder for video decoding) that can determine that a sub-block transform (SBT) is enabled for a video block. The device can determine whether a secondary transform (e.g., a low-frequency non-separable transform (LFNST) or a non-separable primary transform (NSPT)) is enabled for the video block. The device can determine a transform function to decode the video block based on determining whether the secondary transform is enabled for the video block. For example, a first type of transform function (e.g., discrete cosine transform (DCT) II) can be used to decode the video block if the LFNST or NSPT is enabled for the video block. A second type of transform function can be used to decode the video block if the secondary transform is disabled for the video block. The device can decode the video block using the transform function.

A decoder for decoding a one-dimensional digital waveform signal from a data stream by use of block-wise transform decoding, the decoder configured to decode a sequence of transform coefficients of a predetermined block of the one-dimensional digital waveform signal from the data stream by deriving a position information from the data stream, locating, using the position information, a position in the sequence of transform coefficients; and decoding one or more first transform coefficients of the sequence of transform coefficients from the data stream, which precede the position in the sequence of transform coefficients, and attributing a predetermined value to one or more second transform coefficients of the sequence of transform coefficients, which follow the position in the sequence of transform coefficients. Further disclosed is an encoder, a system, methods, a data stream and a computer program.

A method of encoding images into a bitstream, the method comprises determining a residual subblock comprising an area of a residual block for a block to be encoded, and encoding transform coefficients of a transform block to be encoded based on the residual subblock area. A corresponding method comprises decoding images from a bitstream. The methods may comprise obtaining a prediction information portion of a block, comprising a plurality of samples for the block and determining the residual subblock based on the prediction samples. The methods may comprise inferring the position of the residual subblock. The process may be an Advanced Subblock Transform (ASBT) process. The method may comprise selectively encoding an orientation of the transform block in the bitstream, and wherein the orientation of the transform block is context-adaptive binary arithmetic coding (CABAC) coded; wherein at least one CABAC context is associated with the orientation of the transform block. [Fig. 14]

The present application relates to the technical field of radar sensor, and relates to a millimeter wave radar echo signal processing method and system. When the surrounding environment is detected by a millimeter wave radar, a transmitting array generates and transmits FMCW radar wave detection pulses, a receiving array receives echo signals of the radar wave detection pulses reflected by targets in the environment, and the echo signals enter a receiving signal processor. The received signal pulses are converted into radar digital baseband signal sequences by a radio frequency front end module, a radar echo digital baseband intermediate signal sequence is obtained through first step digital baseband signal processing, the radar echo digital baseband intermediate signal sequence is transmitted to hardware for performing second step digital baseband processing through a high-speed data interface, the second step digital baseband processing is performed, CFAR and AoA detection are finally performed, and information of the detected target is obtained. Compared with a traditional scheme, the data amount of inter-chip transmission is greatly reduced, so that the baseband data scale that can be processed by the radar system is increased, and the radar detection capability is improved.

An example method of video decoding includes receiving a video bitstream that includes multiple blocks, including a current block. The method also includes identifying a partial region of the current block, where residual data for the current block outside of the partial region is zero, and reconstructing the current block by applying a subblock transform to the partial region of the current block. Instructions for the example method may be stored in a computer system or storage medium.

This disclosure provides a method, apparatus, storage medium, and electronic device for quantizing feature data in a model. The method includes: acquiring feature data to be quantized from a target model deployed on a target device, wherein the feature data includes at least two data blocks; performing an inter-block transformation on the feature data using a pre-constructed inter-block orthogonal transformation matrix to obtain inter-block transformed data; performing an intra-block transformation on each data block included in the inter-block transformed data using a pre-constructed intra-block orthogonal transformation matrix to obtain intra-block transformed data; and quantizing the intra-block transformed data to obtain quantized feature data. This disclosure can evenly distribute the energy of the feature data across all blocks, eliminating the influence of high-variance blocks on the quantization sharing index, and redistributing the intra-block data so that the intra-block data is evenly mapped to each codeword interval of the codebook, solving the codebook collapse problem and thus significantly improving quantization accuracy.

Embodiments in the present specification provide a method and device for processing video signal including steps of: obtaining a sub-block transform (SBT) flag indicating whether a SBT is applied, wherein the SBT represents a transform applied to a transform unit corresponding to one of subblocks split from a coding unit; determining a variable value related to a transform target region based on the SBT flag and a size value of the transform unit, wherein if the SBT flag equals to 1 and the size value is greater than a threshold value, the variable value is determined as the threshold value; obtaining position information of a last significant coefficient according to a scan order within the transform unit, wherein the position information of the last significant coefficient is coded based on the variable value; and performing an inverse-transform based on the position information of the last significant coefficient.

A mechanism for processing video data is disclosed. The mechanism includes determining to apply a sub-block transform (SBT) to an intra-predicted coding unit (CU). A conversion is performed between a visual media data and a bitstream based on the intra-predicted CU. For example, a CU can be split into subblocks for intra prediction and then the SBT can be applied to the residual of each subblock.

The application discloses a video coding method, device and equipment, and belongs to the technical field of audio and video. The method comprises the following steps: decoding a target coding unit to obtain a quantization coefficient matrix corresponding to the target coding unit; determining first reference information according to quantization coefficients in the quantization coefficient matrix; obtaining the value of a transform flag corresponding to the first reference information, wherein the transform flag refers to a flag of a sub-block transform position, and the sub-block transform position refers to the position of a sub-block in the coding unit which needs to be transformed and quantized; and determining the sub-block transform position of the target coding unit according to the obtained value of the transform flag. The application implicitly indicates the flag of the sub-block transform position in the target coding unit through the quantization coefficients in the quantization coefficient matrix corresponding to the coding unit, avoids explicit coding of the sub-block transform position, reduces the number of bits occupied by a video code stream, and improves the video coding efficiency.

A method of initial residual hypothesis derivation for joint sign prediction of transform coefficients of a residue block. According to this method, an initial residual hypothesis is derived by applying inverse transformation to de-quantized transform coefficients with the set of sign-predicted transform coefficients set to zero or by applying the inverse transformation to the de-quantized transform coefficients corresponding to non-sign-predicted transform coefficients only to form a first part of the initial residual hypothesis, and adding a second part by summing up products of normalized template functions and absolute values of respective sign-predicted transform coefficients for the set of sign-predicted transform coefficients. The normalized template functions represent approximated entry values of inverse transformation matrixes along top and left block boundaries for corresponding transform coefficient indexes associated with the respective sign-predicted transform coefficients. The initial residual hypothesis is used to derive sign prediction for the sign-predicted transform coefficients.

Embodiments of the present disclosure relate to a method and apparatus for video coding, device and computer readable medium. A method for video decoding includes determining a maximum transform size associated with a video sequence based on a first high level syntax element of the video sequence, the maximum transform size corresponding to a maximum transform unit area. When a sub-block transform (SBT) mode is enabled, a maximum block size of the SBT mode is allowed to be constrained by the maximum transform size. Wherein the maximum block size of the SBT mode is allowed to be constrained by the maximum transform size includes allowing a width and a height of a maximum coding unit (CU) of the SBT mode to be derived as a minimum value between the maximum transform size indicated by the first high level syntax element and a width and a height of a maximum CU of the SBT mode allowed to be indicated in a second high level syntax element.

This invention provides a radar signal modulation recognition method based on temporal modeling and cross-variable fusion, comprising: preprocessing the radar IQ signal to be identified into blocks to obtain block tensors; constructing a hybrid neural network model, which includes a block transform temporal encoder, a modern temporal convolutional network cross-variable fusion module, and a classification head. The block transform temporal encoder is used to capture long-range temporal dependencies, and the modern temporal convolutional network cross-variable fusion module is used to perform cross-variable fusion on the extracted temporal feature tensors; inputting the block tensors into the trained hybrid neural network model for processing using the block transform temporal encoder, the modern temporal convolutional network cross-variable fusion module, and the classification head, and outputting radar signal modulation recognition results. This improves the model's adaptability and generalization ability; compensates for the shortcomings of local segmentation methods in global modeling capabilities; and achieves joint feature extraction of multi-channel radar signals.

This application discloses an image processing method, device, system, electronic equipment, and storage medium, applied at the encoding end. The method includes extracting a one-dimensional feature vector from an original image block; transforming the original image block into a multidimensional feature map based on the one-dimensional feature vector; quantizing and encoding the one-dimensional feature vector to generate a first code stream; discretely encoding the multidimensional feature map to generate a second code stream, thereby efficiently compressing the spatial-independent vector and the multidimensional feature map; and sending the first and second code streams to the decoding end. Since the encoding stream comprises two layers each representing different types of image information, image reconstruction from the two-layer code streams maintains information integrity even at low bit rates, thus improving visual effects and experience.

A method for video processing includes determining, based on a first indication, whether a sub-block residual coding scheme is applied to residual of a current video block in a transform skip mode, the sub-block residual coding scheme splitting the residual of the current video block into multiple sub-blocks and a subset of the multiple sub-blocks have non-zero coefficients; determining, based on a second indication, a specific split pattern to be applied to the residual of the current video block, in response to the sub-block residual coding scheme being applied to the residual of the current video block; deriving, based on a third indication, the subset of the multiple sub-blocks which have non-zero coefficients; and performing a conversion on the residue of the current video block based on the determined subset of sub-blocks having non-zero coefficients.

The present document relates an audio encoding and decoding system (referred to as an audio codec system). In some embodiments, a method of audio signal encoding comprises: receiving an input audio signal; transforming a sequence of samples of the input audio signal into a block of transform coefficients, the transform coefficients indicative of the spectral energy of the block; estimating a spectral envelope of the block from the transform coefficients; adjusting the transform coefficients using the spectral envelope and a spectral shaper, the spectral shaper including one or more parameters indicative of a fundamental frequency of a multi-sinusoidal signal model, where the fundamental frequency corresponds to a time domain delay; and entropy coding the adjusted transform coefficients.

This application discloses an image processing method, device, system, electronic equipment, and storage medium, applied at the encoding end. The method includes extracting a one-dimensional feature vector from an original image block; transforming the original image block into a multidimensional feature map based on the one-dimensional feature vector; quantizing and encoding the one-dimensional feature vector to generate a first code stream; discretely encoding the multidimensional feature map to generate a second code stream, thereby efficiently compressing the spatial-independent vector and the multidimensional feature map; and sending the first and second code streams to the decoding end. Since the encoding stream comprises two layers each representing different types of image information, image reconstruction from the two-layer code streams maintains information integrity even at low bit rates, thus improving visual effects and experience.

According to one aspect of the present disclosure, a method of decoding is provided. The method may include, in response to a sub-block transform (SBT) with corner partitioning or center partitioning being enabled for a current block, determining, by a processor, whether a non-separable transform is enabled for the current block. The method may include, in response to the non-separable transform being enabled, determining, by the processor, the non-separable transform from among a plurality of non-separable transforms based on a size of a SBT partition. The method may include decoding, by the processor, the current block based on the non-separable transform.

A method and apparatus for joint sign prediction of transform coefficients of residual blocks in a video coding system are disclosed. At the encoder side, a transform coefficient region or an index value range is determined according to coding context associated with the current block. A set of signs associated with a set of selected transform coefficients are determined for joint sign prediction are determined according to the transform coefficient region or the index value range. Joint sign prediction for the set of signs is determined by selecting a hypothesis from a group of hypotheses for the set of signs that achieves a minimum cost. The sign prediction is then used for coding the set of signs. A corresponding method and apparatus for the decoder side is also disclosed.