17 results about "Quantization matrix" patented technology

A system and method for energy-efficient approximate digital JPEG and MJPEG-compression. The system includes a controller unit to control a processing loop for processing image blocks based on a comparison of a current image block to a previous image block. The system includes a quantization unit configured to quantize the frequency domain representation using an approximate quantization process and a quantization (Q) matrix. The quantization unit is configured to: identify, a nearest power of two value for each element of the quantization matrix; generate an updated Q matrix by assigning each element of the quantization matrix with the identified nearest power of two value; and shift each element of the updated Q matrix by a number of bits to generate a quantized frequency domain representation. The number of bits corresponds to the identified nearest power of two for the corresponding element of the updated Q matrix.

The present application provides an image decoding apparatus and method, an image encoding apparatus and method, and a storage medium. A decoding unit decodes data corresponding to a first array of quantized coefficients including an N x M array of quantized coefficients corresponding to a first block and data corresponding to a second array of quantized coefficients including an N x M array of quantized coefficients corresponding to a second block. An inverse quantization unit derives a first array of orthogonal transform coefficients from the first array of quantized coefficients using at least a first quantization matrix of elements of the N x M array and derives a second array of orthogonal transform coefficients from the second array of quantized coefficients using at least a second quantization matrix of elements of the N x M array. An inverse orthogonal transform unit inverse orthogonally transforms the first array of orthogonal transform coefficients to generate a first prediction residual of pixels of a P x Q array and inverse orthogonally transforms the second array of orthogonal transform coefficients to generate a second prediction residual of pixels of the N x M array.

The present application provides an image decoding apparatus and method, an image encoding apparatus and method, and a storage medium. A decoding unit decodes data corresponding to a first array of quantized coefficients including an N x M array of quantized coefficients corresponding to a first block and data corresponding to a second array of quantized coefficients including an N x M array of quantized coefficients corresponding to a second block. An inverse quantization unit derives a first array of orthogonal transform coefficients from the first array of quantized coefficients using at least a first quantization matrix of elements of the N x M array and derives a second array of orthogonal transform coefficients from the second array of quantized coefficients using at least a second quantization matrix of elements of the N x M array. An inverse orthogonal transform unit inverse orthogonally transforms the first array of orthogonal transform coefficients to generate a first prediction residual of pixels of a P x Q array and inverse orthogonally transforms the second array of orthogonal transform coefficients to generate a second prediction residual of pixels of the N x M array.

Disclosed herein is a video decoding method including determining one or more adaptation parameter sets including a quantization matrix set including a plurality of quantization matrices, determining an adaptation parameter set including a quantization matrix set applied to a current picture or a current slice from among the one or more adaptation parameter sets, dequantizing transform coefficients of a current block of a current picture or a current slice based on the quantization matrix set of the determined adaptation parameter set, and reconstructing the current block based on the dequantized transform coefficients, wherein the adaptation parameter set includes coding information applied to one or more pictures or slices.

Provided is an image encoding device that encodes image data in units of blocks, wherein a quantization means quantizes a transform coefficient, which is generated by using only a second transform means, by using a quantization matrix to generate a quantized transform coefficient, and quantizes a transform coefficient generated by using both first transform means and the second transform means without using the quantization matrix to derive a quantized transform coefficient.

The encoding device includes: a predictor configured to generate, for each component, a prediction block corresponding to an encoding-target block; a residual generator configured to generate, for each component, a prediction residual representing a difference between the encoding-target block and the prediction block; a mode selector configured to select one mode either an individual encoding mode performing a transform process and a quantization process on a prediction residual of the first component and a prediction residual of the second component for each single component, or a joint encoding mode performing a transform process and a quantization process on a joint prediction residual generated from the prediction residual of the first component and the prediction residual of the second component; a quantization controller configured to determine a quantization matrix to be applied in the quantization process based on the mode selected by the mode selector.

A method for signaling a scaling matrix for transform factor quantization is provided. A decoder receives data from a bitstream to be decoded as a current picture of video, and the current picture is decoded using a plurality of scaling matrices. The decoder receives a reference index offset for a first scaling matrix of the plurality of scaling matrices. The decoder applies the reference index offset to a first index identifying the first scaling matrix to derive a second index identifying the second scaling matrix of the plurality of scaling matrices. The second scaling matrix has been previously reconstructed. The decoder reconstructs the first scaling matrix by referencing the second scaling matrix. The decoder inversely quantizes the transform factors of a transform block of the current picture using the plurality of scaling matrices and reconstructs the current picture using the inversely quantized transform factors.

The present application provides an image decoding apparatus and method, an image encoding apparatus and method, and a storage medium. A decoding unit decodes data corresponding to a first array of quantized coefficients including an N x M array of quantized coefficients corresponding to a first block and data corresponding to a second array of quantized coefficients including an N x M array of quantized coefficients corresponding to a second block. An inverse quantization unit derives a first array of orthogonal transform coefficients from the first array of quantized coefficients using at least a first quantization matrix of elements of the N x M array and derives a second array of orthogonal transform coefficients from the second array of quantized coefficients using at least a second quantization matrix of elements of the N x M array. An inverse orthogonal transform unit inverse orthogonally transforms the first array of orthogonal transform coefficients to generate a first prediction residual of pixels of a P x Q array and inverse orthogonally transforms the second array of orthogonal transform coefficients to generate a second prediction residual of pixels of the N x M array.

The present application provides an image decoding apparatus and method, an image encoding apparatus and method, and a storage medium. A decoding unit decodes data corresponding to a first array of quantized coefficients including an N x M array of quantized coefficients corresponding to a first block and data corresponding to a second array of quantized coefficients including an N x M array of quantized coefficients corresponding to a second block. An inverse quantization unit derives a first array of orthogonal transform coefficients from the first array of quantized coefficients using at least a first quantization matrix of elements of the N x M array and derives a second array of orthogonal transform coefficients from the second array of quantized coefficients using at least a second quantization matrix of elements of the N x M array. An inverse orthogonal transform unit inverse orthogonally transforms the first array of orthogonal transform coefficients to generate a first prediction residual of pixels of a P x Q array and inverse orthogonally transforms the second array of orthogonal transform coefficients to generate a second prediction residual of pixels of the N x M array.

An embodiment includes generating, using a first set of parameters, a first quantization matrix. The first set of parameters includes a set of coefficients of a formula that generates quantization values in the first quantization matrix. An embodiment includes encoding, using the first quantization matrix, a frame in an uncompressed video stream, the encoding generating a compressed video stream corresponding to the uncompressed video stream.

This invention relates to the field of data processing and intelligent collaborative control technology, specifically disclosing a full-link data collaboration system based on deep computing. The system collects multi-source heterogeneous data and fuses it to generate an initial fused data set; it performs causal relationship mining, constructs a causal structure topology graph, and calculates the causal influence weights of variables, generating a data set with causal weighting features; when an abnormal event occurs, it intervenes and assigns values ​​to target variables based on the causal structure topology graph, simulating and generating virtual data sequences under counterfactual scenarios, forming a counterfactual comparison data set; by comparing the differences between actual and virtual data layer by layer, it calculates the marginal causal contribution of the abnormal event to each node, generating a cascading influence quantification matrix; based on this matrix and constrained by minimizing the overall loss of the entire link, it dynamically optimizes the data interaction rules and collaboration parameters of each link, outputting collaborative control commands; this invention achieves accurate quantification and dynamic collaborative optimization of abnormal events.

The application provides a weight data processing method, device, medium and product of a neural network model, and relates to the technical field of machine learning. The method comprises the following steps: performing quantization processing on an initial weight matrix obtained by training a neural network model after splitting the initial weight matrix, to obtain a significant quantization matrix and a non-significant quantization matrix; obtaining a significant difference matrix based on the initial weight matrix and the significant quantization matrix and storing the significant difference matrix, obtaining a non-significant difference matrix based on the initial weight matrix and the non-significant quantization matrix and storing the non-significant difference matrix; then, determining a weight matrix used for model inference based on a target number of significant difference matrices, a target number of non-significant difference matrices, the significant quantization matrix and the non-significant quantization matrix; and during model inference, obtaining weight matrices with different bit numbers according to memory occupation information, so that better model inference effects are obtained without increasing the calculation overhead, the utilization rate of the memory is improved, and the maximum utilization of the memory is realized.

This application provides a video decoding method, a video encoding method, an apparatus, a device, and a storage medium, relating to the technical field of video encoding and decoding processing. [Solution] The decoding method includes the steps of: obtaining a first parameter set corresponding to the video frame to be decoded; determining a valid QM based on the syntax elements included in the first parameter set, wherein the valid QM refers to the QM actually used when performing inverse quantization on the quantized transformation coefficients in the decoding process of the video frame to be decoded; and decoding with the valid QM. By employing the technical means of the present invention, the decoder only needs to decode with the valid QM, thereby reducing the complexity of the decoder's calculations.

The application provides a model compression method, system, terminal and storage medium, the method comprises the following steps: model training is carried out on a to-be-compressed model, a regularization term and singular value decomposition are added in the to-be-compressed model, a singular value matrix is obtained, the step of executing model training on the to-be-compressed model and subsequent steps are returned according to the singular value matrix, until the to-be-compressed model meets a performance decline condition, and the to-be-compressed model is output; parameter clustering is carried out according to a weight tensor of the to-be-compressed model, a weight parameter matrix is obtained, weight quantization is carried out on the weight parameter matrix, and a clustered quantization matrix is obtained; parameter setting is carried out on the to-be-compressed model according to the clustered quantization matrix, and a compressed model is obtained. The application realizes maximum model compression on the premise of ensuring accuracy by jointly compressing the model from the global perspective based on the joint model compression mode of sparse regularization, iterative pruning and clustered quantization.

The present application provides an image decoding apparatus and method, an image encoding apparatus and method, and a storage medium. A decoding unit decodes data corresponding to a first array of quantized coefficients including an N x M array of quantized coefficients corresponding to a first block and data corresponding to a second array of quantized coefficients including an N x M array of quantized coefficients corresponding to a second block. An inverse quantization unit derives a first array of orthogonal transform coefficients from the first array of quantized coefficients using at least a first quantization matrix of elements of the N x M array and derives a second array of orthogonal transform coefficients from the second array of quantized coefficients using at least a second quantization matrix of elements of the N x M array. An inverse orthogonal transform unit inverse orthogonally transforms the first array of orthogonal transform coefficients to generate a first prediction residual of pixels of a P x Q array and inverse orthogonally transforms the second array of orthogonal transform coefficients to generate a second prediction residual of pixels of the N x M array.

A pixel group that has been determined to be similar to a pixel group adjacent to a block to be encoded in a frame is searched for from encoded pixel groups in the frame, and a predicted image composed of encoded pixels of the frame is generated on the basis of the searched pixel group. Then, a prediction error is calculated using the predicted image, frequency conversion is performed on the prediction error to generate a conversion coefficient, the conversion coefficient is quantized using a quantization matrix to generate a quantization coefficient, and the quantization coefficient is encoded.

The invention provides a model quantification method and device, a model reasoning method and device, equipment, a medium and a product. Constructing an input base of a sample input space based on the sample set, and defining an energy distribution condition of a base direction of the sample input space; the method comprises the following steps of: mapping an original weight matrix to a sample input space through the original weight matrix based on an input base and an expert module to obtain a mapping weight matrix, so that a quantization process of a weight parameter is explicitly aligned with a model input space; then, carrying out generality extraction on the mapping weight matrix of each expert module to obtain a specific weight matrix and a generality weight matrix of each expert module; and the specific weight matrix is quantized to obtain a quantization matrix, so that only the specific part of each expert module is taken as a to-be-quantized object, the limited quantization matrix can focus on the difference between experts and contains more information amount, the information utilization rate of the quantization matrix is remarkably improved, and the accuracy of model quantization is improved.