123 results about "Inverse quantization" patented technology

The invention relates to the technical field of integrated circuit design and artificial intelligence hardware acceleration, and particularly discloses a quantization precision adaptive switching hardware acceleration architecture. Hardware is divided into a static area and a dynamic area, wherein the static area comprises a pre-processing module, a post-processing module, a control module and a reconfigurable logic loading control block which run all the time; the dynamic region bears a quantization and inverse quantization module composed of a plurality of reconfigurable logic units. The method is characterized in that a model precision instruction is analyzed through a control module, and a reconfigurable logic file with corresponding precision is dynamically loaded to a dynamic area from external starting equipment by a loading control block according to an internal address mapping table, so that seamless switching of different quantization precision is realized on the hardware level, and the whole chip does not need to be burnt again. According to the design, through a modularized and reconfigurable hardware structure, the limitation of fixed-precision hardware in coping with diversified AI models is effectively solved, and the deployment flexibility and energy efficiency are remarkably improved.

The invention relates to a task-oriented semantic communication-oriented adaptive source channel coding system, which belongs to the field of wireless communication, and comprises a sending end for extracting a semantic feature map for evaluation from input data through a feature extractor initialized based on a pre-training large model, executing nested quantization and arithmetic coding based on a calculated importance score, and sending the semantic feature map to a task-oriented semantic communication-oriented adaptive source channel coding system; generating a scalable semantic bit stream with importance levels; the physical layer is used for carrying out reordering and grouping according to importance levels of bit streams, a channel encoder is combined with an interleaver to realize unequal error protection of semantic matching, and encoded bits are mapped to sign bits with different fault-tolerant capabilities in a constellation diagram through division modulation; and a receiving end: recovering the semantic feature map through arithmetic decoding and nested inverse quantization. According to the invention, on the premise of keeping compatibility with a digital communication protocol stack, through distortion-driven joint rate self-adaption, the transmission effectiveness for machine tasks and human vision is obviously improved under a limited bandwidth.

The embodiment of the invention discloses a processor, a chip product, computer equipment and a data processing method. The processor comprises a block matrix calculation unit, a storage unit and a target register, wherein the storage unit is used for storing an input matrix to be processed and inverse quantization data corresponding to the input matrix; the block matrix calculation unit is used for reading the block matrix from the storage unit and reading the last target operation result from the target register; performing accumulation operation on the last target operation result and the block matrix to obtain a candidate operation result; and according to the inverse quantization parameter corresponding to the block matrix read from the storage unit, performing data precision adjustment on each element in the candidate operation result to obtain a current target operation result, and updating and storing the current target operation result in a target register.

FIELD: video signal encoding and decoding.SUBSTANCE: residual coefficient of the residual block is extracted. The quantization parameter for the residual block is calculated. Inverse quantization on the residual coefficient using the calculated quantization parameter is performed. A residual sample of the residual block is formed by performing an inverse transform on the inversely quantized residual coefficient. The encoded residual coefficient for the partial region in the residual block is not transmitted from the encoding device. The residual sample of the partial region in the residual block is set to the default value preset in the decoder, which is 0. The default value is set based on at least one of the size or shape of the residual block. A partial region is a region excluding at least one of the N columns on the left side of the residual block or M rows on the top side of the residual block.EFFECT: increased efficiency of video signal coding.9 cl, 16 dwg

A displacement decoding unit (205) of a mesh decoding device (200) according to the present invention includes: a bypass arithmetic decoding unit (205A) configured to generate a coefficient level value by performing bypass arithmetic decoding on a displacement bit stream; an inverse quantization unit (205B) configured to generate a first transformed coefficient by performing inverse quantization on the coefficient level value; an adder (205D) configured to generate a second transformed coefficient by adding a prediction transformed coefficient and a prediction residual; an inter prediction unit (205F) configured to generate the prediction transformed coefficient by performing inter prediction by using the second transformed coefficient of a reference frame read from the frame buffer; and a second inverse transform unit (205G) configured to generate a decoded displacement by performing second inverse transform on the second transformed coefficient.

A decoder decodes quantized wavelet coefficients representing the displacement of vertices in a set of vertices of a three-dimensional (3D) mesh from a bitstream, and decodes quantization information from the bitstream indicating a quantization offset associated with a subset of vertices in the set of vertices. The decoder further performs inverse quantization on the quantized wavelet coefficients to determine wavelet coefficients, and adjusts the wavelet coefficients according to the quantization offset based on the vertices being in the subset to determine the displacement of the vertices.

The embodiment of the application discloses a lossy compression method, a decompression method and equipment for time sequence floating point data, which are used for improving the data compression rate and the decompression speed of the time sequence floating point data. The method comprises the following steps: acquiring a floating point array, the floating point array comprising a plurality of floating point data arranged in time sequence; sequentially determining other floating point data in the floating point array as target floating point data, and determining a previous floating point data of the target floating point data from the floating point array; determining a difference between the target floating point data and the previous floating point data as a difference data of the target floating point data; performing quantization processing on the difference data based on a preset error boundary to obtain quantized difference data corresponding to the target floating point data, wherein a difference between lossy data corresponding to the target floating point data and the target floating point data is not greater than the preset error boundary, and the lossy data corresponding to the target floating point data is obtained by performing inverse quantization processing on the quantized difference data corresponding to the target floating point data; and performing compression processing on each quantized difference data.

This provides a video encoding method. [Solution] The method includes determining a quantification parameter for residual data of a coding unit (CU), deriving a scale value by scaling the quantification parameter by a scale factor, determining a plurality of coefficients associated with the CU, determining a plurality of parameters associated with the CU, obtaining a plurality of bit shifts by bit shifting the plurality of parameters, and obtaining a quantification level based on the scale value, the plurality of coefficients, and the plurality of bit shifts.

An apparatus receives a compressed bitstream including an encoded displacements bitstream and packing method information indicating whether the displacement component samples are packed in ascending order or in descending order. The apparatus video-decodes the encoded displacements bitstream to generate a displacement video frame, wherein padding is added at a bottom of the displacement video frame irrespective of whether the displacement component samples are packed in ascending order or in descending order. The apparatus image-unpacks the displacement video frame to generate an array of quantized displacement wavelet coefficients, inverse-quantizes the array of quantized displacement wavelet coefficients to generate displacement wavelet coefficients, and inverse-wavelet-transforms the displacement wavelet coefficients to generate displacement component samples.

A deep learning-based signal processing method according to the present invention may: obtain a quantized feature map from a bitstream; reconstruct the feature map by performing inverse quantization on the quantized feature map; and synthesize the reconstructed feature map on the basis of a neural network.

The invention relates to a non-standard JPEG (Joint Photographic Experts Group) coding processing method for dynamic code rate control. The method comprises the following steps: presetting a multi-frame joint image processing scene, obtaining an original reference image, and carrying out combined processing of boundary expansion and random bias addition to obtain a preprocessed reference image; dividing macro blocks for the preprocessed reference image, and distributing an independent coding parameter configuration space after verification and compliance; frequency domain conversion is executed according to a division result, an initial quantization parameter is calculated through an index model, and a dynamic quantization table is dynamically adjusted; executing non-standard entropy coding to obtain an indexed macro block coding code stream; combining the indexed code stream and a macro block division result to construct an index table containing a macro block serial number, an initial address and a code stream length; analyzing the code stream to obtain quantized data, and performing inverse quantization by using a dynamic quantization table to obtain spatial domain data; enabling the format of the reconstructed image to be consistent with that of the preprocessed reference image through reverse preprocessing, and performing real-time processing on the linkage input image to obtain a reconstructed reference image; and non-standard JPEG coding processing of dynamic code rate control is realized.

A video decoding method is disclosed, comprising the steps of: determining a quadrilateral scanning region comprising all significant transform coefficients within a current block; scanning information related to the transform coefficients within the quadrilateral scanning region according to a predetermined scanning order; obtaining the transform coefficients of the current block based on the scanned information related to the transform coefficients; performing inverse quantization and inverse transform on the transform coefficients of the current block to generate a residual block of the current block, and recovering the current block based on the generated residual block.

The present application relates to a kind of power grid energy consumption prediction method and system based on depth compression federal learning, comprising: server first distributes global model parameter to each client;After each client completes model training based on local data, in turn, client pruning, quantization and coding operation are executed, to obtain compressed local model.Subsequently, client uploads its compressed model to server end, server decodes the received model, restores as quantization model, and the quantization model of all client is aggregated;To improve model security and prevent potential malicious gradient attack, server further executes server pruning strategy after aggregation, and the aggregation result is optimized.Finally, server distributes the quantization model after pruning to each client again.After receiving model, client executes alignment and inverse quantization operation based on local information, to update local model, to complete a round of federal optimization process.

The application discloses a nano unmanned aerial vehicle image navigation method and system based on an ultra-low-power edge chip, relates to the cross technical field of unmanned aerial vehicle autonomous navigation and edge computing, and comprises the following steps: constructing a lightweight feature extraction backbone network based on a deep separable convolution, generating a basic navigation model suitable for an edge chip based on early feature fusion and end-to-end decision calibration; obtaining a quantization model suitable for a chip instruction set through full-link fixed-pointing, operator smelting, and integer and inverse quantization collaborative optimization; combining a multi-level storage structure of the chip, demarcating a target memory space, establishing a memory alignment and hardware protection mechanism, and building a stable bottom-layer running environment; and achieving full-airborne high-frame-rate real-time navigation operation of the nano unmanned aerial vehicle through tensor block constraint planning and asynchronous inference pipeline design. Through optimization of a network architecture, power scheduling and a memory mechanism, the application realizes airborne independent low-power navigation, and solves the problems of high navigation power consumption of the unmanned aerial vehicle and difficulty in adaptation.

The present invention provides an image data encoding / decoding method and apparatus that improves compression performance. [Solution] The 360-degree image decoding method receives a bitstream encoded with a 360-degree image, generates a predicted image by referring to syntactic information obtained from the received bitstream, combines the generated predicted image with a residual image obtained by inverse quantization and inverse transformation of the bitstream to obtain a decoded image, and reconstructs the decoded image into a 360-degree image in a projection format. The generation of the predicted image involves obtaining a group of motion vector candidates, including the motion vectors of blocks adjacent to the current block to be decoded, from the motion information contained in the syntactic information, deriving a predicted motion vector from the group of motion vector candidates based on selection information extracted from the motion information, and using the final motion vector derived by adding the predicted motion vector with the difference motion vector extracted from the motion information, the predicted block of the current block to be decoded is determined.

The present invention provides an image data encoding / decoding method and apparatus that improves compression performance. [Solution] The 360-degree image decoding method receives a bitstream encoded with a 360-degree image, generates a predicted image by referring to syntactic information obtained from the received bitstream, combines the generated predicted image with a residual image obtained by inverse quantization and inverse transformation of the bitstream to obtain a decoded image, and reconstructs the decoded image into a 360-degree image in a projection format. The generation of the predicted image involves obtaining a group of motion vector candidates, including the motion vectors of blocks adjacent to the current block to be decoded, from the motion information contained in the syntactic information, deriving a predicted motion vector from the group of motion vector candidates based on selection information extracted from the motion information, and using the final motion vector derived by adding the predicted motion vector with the difference motion vector extracted from the motion information, the predicted block of the current block to be decoded is determined.

The present invention provides an image data encoding / decoding method and apparatus that improves compression performance. [Solution] The 360-degree image decoding method receives a bitstream encoded with a 360-degree image, generates a predicted image by referring to syntactic information obtained from the received bitstream, combines the generated predicted image with a residual image obtained by inverse quantization and inverse transformation of the bitstream to obtain a decoded image, and reconstructs the decoded image into a 360-degree image in a projection format. The generation of the predicted image involves obtaining a group of motion vector candidates, including the motion vectors of blocks adjacent to the current block to be decoded, from the motion information contained in the syntactic information, deriving a predicted motion vector from the group of motion vector candidates based on selection information extracted from the motion information, and using the final motion vector derived by adding the predicted motion vector with the difference motion vector extracted from the motion information, the predicted block of the current block to be decoded is determined.

The invention provides a matrix calculation device and method, electronic equipment and a storage medium. The matrix calculation device is used for the artificial intelligence processor and comprises at least one inverse quantization calculation module and an inverse quantization accumulation module. And each inverse quantization calculation module comprises at least one processing unit and an accumulation unit. According to the matrix calculation device, through two-stage inverse quantization operation, the overall precision loss can be reduced, seamless connection of calculation assembly lines can be achieved, and the overall calculation throughput rate is increased.

The present invention provides an image data encoding / decoding method and apparatus that improves compression performance. [Solution] The 360-degree image decoding method receives a bitstream encoded with a 360-degree image, generates a predicted image by referring to syntactic information obtained from the received bitstream, combines the generated predicted image with a residual image obtained by inverse quantization and inverse transformation of the bitstream to obtain a decoded image, and reconstructs the decoded image into a 360-degree image in a projection format. The generation of the predicted image involves obtaining a group of motion vector candidates, including the motion vectors of blocks adjacent to the current block to be decoded, from the motion information contained in the syntactic information, deriving a predicted motion vector from the group of motion vector candidates based on selection information extracted from the motion information, and using the final motion vector derived by adding the predicted motion vector with the difference motion vector extracted from the motion information, the predicted block of the current block to be decoded is determined.

The present disclosure relates to the technical field of image processing and discloses an image decoding and encoding method and apparatus, a device and a storage medium. In the present disclosure, feature channel distinguishing information and / or spatial point distinguishing information are obtained, and image code streams are decoded to obtain quantized residual values; an inverse quantization precision parameter value is set for each quantized residual value according to the feature channel distinguishing information and / or the spatial point distinguishing information; inverse quantization is performed on the quantization residual values according to the inverse quantization precision parameter values to obtain reconstructed residual values; synthetic transformation is performed on the reconstructed residual values to obtain a reconstructed image block.

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 invention relates to an online computing method and device, computer equipment and a storage medium. The method comprises the following steps: receiving first quantized data and a first scaling coefficient sent by computing equipment; performing inverse quantization processing on the first quantized data based on the first scaling coefficient to obtain first inverse quantization data, performing operation processing on the first inverse quantization data to obtain second inverse quantization data, and performing quantization processing on the second inverse quantization data to obtain second quantized data and a second scaling coefficient; and sending the second quantized data and the second scaling coefficient to the next network node. By adopting the method, the data transmission efficiency and the calculation efficiency in the artificial intelligence network can be improved.

An apparatus for decoding encoded dynamic mesh data may be configured to: determine a set of quantized integer coefficient values ​​for the displacement vectors of the encoded dynamic mesh data; inverse quantize the set of quantized integer coefficient values ​​to determine a set of fixed-point inverse quantization coefficient values; determine a set of fixed-point transform coefficient values ​​based on the set of fixed-point inverse quantization coefficient values; convert the set of fixed-point transform coefficient values ​​into a set of floating-point transform coefficient values; inverse transform the set of floating-point transform coefficient values ​​to determine a set of reconstructed displacement vectors; and determine a reconstructed deformable mesh based on the set of reconstructed displacement vectors.

The application provides an image adaptive quantization communication method and system, relates to the technical field of artificial intelligence and semantic communication, and comprises the following steps: performing feature extraction on an input image to form an initial code word representation; adopting an adaptive dynamic quantization mechanism to perform quantization processing on the initial code word representation to generate a code word index as a quantization feature; after the receiving end obtains the code word index of the sending end, performing inverse quantization by using a codebook knowledge base, and reconstructing a recovered image by using a decoder. The method and system provided by the application avoid representation collapse and realize adaptive matching and dynamic updating between codebook elements and semantic features through multi-codebook cooperation, a learnable linear transformation and a dynamic selection mechanism based on attention; and through quantization representation and transmission of the image and then decoding and recovery, the efficiency of image transmission is greatly improved, and an efficient and intelligent basic framework is provided for a next-generation semantic-oriented image communication system.

The present invention provides an image data encoding / decoding method and apparatus that improves compression performance. [Solution] The 360-degree image decoding method receives a bitstream encoded with a 360-degree image, generates a predicted image by referring to syntactic information obtained from the received bitstream, combines the generated predicted image with a residual image obtained by inverse quantization and inverse transformation of the bitstream to obtain a decoded image, and reconstructs the decoded image into a 360-degree image in a projection format. The generation of the predicted image involves obtaining a group of motion vector candidates, including the motion vectors of blocks adjacent to the current block to be decoded, from the motion information contained in the syntactic information, deriving a predicted motion vector from the group of motion vector candidates based on selection information extracted from the motion information, and using the final motion vector derived by adding the predicted motion vector with the difference motion vector extracted from the motion information, the predicted block of the current block to be decoded is determined.

A chip system, a method of processing data, and a computer readable storage medium are provided. The chip system comprises: a memory for storing first data to be input to a current layer of a neural network and weight data of the current layer; the quantization circuit is used for quantizing the first data by adopting a plurality of quantization modes, and the plurality of quantization modes are used for quantizing the first data into second data with a plurality of data types; the operation circuit is used for determining third data according to the second data and the weight data of the current layer; and the inverse quantization circuit is used for performing inverse quantization on the third data and determining fourth data output by the current layer. According to the scheme, the data of the current layer is subjected to hybrid quantization by adopting multiple quantization modes, so that the quantization error can be reduced, and the operation performance of a chip system is improved.

An image decoding method and apparatus according to an embodiment may extract, from a bitstream, a quantization coefficient generated through core transformation, secondary transformation, and quantization; generate an inverse-quantization coefficient by performing inverse quantization on the quantization coefficient; generate a secondary inverse-transformation coefficient by performing secondary inverse-transformation on a low frequency component of the inverse-quantization coefficient, the secondary inverse-transformation corresponding to the secondary transformation; and perform core inverse-transformation on the secondary inverse-transformation coefficient, the core inverse-transformation corresponding to the core transformation.

Disclosed in the embodiments of the present application are an encoding method, a decoding method, an encoder, a decoder and a storage medium. The decoding method comprises: when a current block meets a preset condition, parsing a first syntax element in a bitstream; determining at least two candidate intra-frame prediction modes of the current block; on the basis of the value of the first syntax element and the at least two candidate intra-frame prediction modes, determining a transform set of the current block; on the basis of the transform set of the current block, determining a transform kernel of the current block; and on the basis of the transform kernel, performing inverse transform on an inverse quantization coefficient of the current block, so as to determine a residual sample of the current block. In this way, the compression efficiency of a transform technique can be improved, thereby improving the encoding and decoding performance.

The invention discloses a large model three-valued zero inverse quantization reasoning method based on position mask coding, and belongs to the field of artificial intelligence. According to the method, positive and negative mask byte pairs are generated for each 8-element subgroup in a quantization stage aiming at three-valued representation {+ 1, 0,-1} of a large model weight, and memory access is optimized by adopting a mask cross storage layout. In the reasoning stage, conditional accumulation of activation values is guided by directly utilizing a position mask through a hardware bit mask instruction, and an inverse quantization link and element-by-element multiplication calculation in a traditional three-valued quantization model are thoroughly eliminated. According to the method, symbol information is converted into a bit-level control signal which can be directly analyzed by hardware, so that the three-valued weight can be more directly mapped to a vector operation unit of a general CPU, and the reasoning efficiency of a large model on resource-constrained equipment is greatly improved.

In an embodiment, an image decoding method for quantization shift is provided. The image decoding method may include obtaining information indicating a quantization parameter of a current block from a bitstream. The image decoding method may include obtaining a transform coefficient of a current block by performing inverse quantization based on information indicating a quantization parameter. The image decoding method may include determining one of a plurality of quantization offsets based on at least one of a prediction mode of a current block, a frequency band of the current block, or a quantization parameter of the current block. The image decoding method may include changing a transform coefficient by using the determined quantization offset.