106 results about "Tight frame" patented technology

The invention provides a cross-frame code migration method and system, electronic equipment and a storage medium, and the code migration method comprises the steps: carrying out the frame-level conversion of a first frame code, and generating a target engineering skeleton adaptive to a second frame protocol; analyzing the construction configuration and the entry file, determining an entry point and a compiling rule of code migration, analyzing the first frame code through an abstract syntax tree, and generating a target code range list needing to be migrated; and according to the target code range list and the dependency item list, applying a preset conversion rule to the first framework code, and generating a second framework code adaptive to a second framework protocol. Through automatic framework-level conversion and dynamic rule adaptation, rapid migration from the first framework code to the second framework code is realized, a multi-terminal code library is unified, the manpower development cost is reduced, synchronous operation of business iteration and migration is supported, and the development efficiency is improved.

The invention relates to the technical field of deep learning, in particular to a PyTorch framework-based compatibility expansion system, which comprises a cross-framework model converter, a heterogeneous hardware abstraction layer, a hybrid computational graph execution engine, an intelligent distributed trainer and a self-adaptive optimizer, the cross-frame model converter converts input model formats of different frames into PyTorch executable formats, the heterogeneous hardware abstraction layer supports rear ends of various hardware and automatically selects an optimal calculation path, and the hybrid calculation graph execution engine fuses dynamic graph flexibility and static subgraph optimization and supports dynamic control function execution flow. An intelligent distributed trainer automatically selects a parallel scheme and is compatible with multi-protocol communication, a self-adaptive optimizer performs dynamic optimization based on hardware characteristics and a PyTorch tensor, a cross-frame model converter supports model conversion of multiple deep learning frames, the cost of migration among different frames by a user is reduced, and the universality of the PyTorch frame is improved.

The invention provides a long video understanding method capable of relieving time sequence illusion in a video language large model. The long video understanding method is based on a static bias adaptive frame selection mechanism and a cross-modal feature fusion strategy. According to the static bias mechanism, inter-frame similarity is evaluated through a discriminator, redundant frames are identified, key frames are selected or a complete sequence is reserved, so that calculation overhead is reduced, and spatio-temporal information integrity is kept; a video frame and a text are mapped to a shared semantic space, the single-frame semantic understanding ability is enhanced, then an embedded sequence serves as a soft prompt to be input into a large language model, and a final answer is generated in an autoregression mode. According to the method, the efficiency and accuracy of long video understanding and video question and answer tasks can be remarkably improved; the problem of low training and reasoning efficiency caused by time sequence dependence redundancy and excessive computing resource consumption is effectively relieved; and through a dynamic multi-modal task processing framework and a space-time memory bank compression mechanism, the modeling capability and generalization performance of the model on a long video sequence are further improved.

Techniques are described herein for processing video data using a neural network system. For instance, a process can include generating, by a first convolutional layer of an encoder sub-network of the neural network system, output values associated with a luminance channel of a frame. The process can include generating, by a second convolutional layer of the encoder sub-network, output values associated with at least one chrominance channel of the frame. The process can include generating, by a third convolutional layer based on the output values associated with the luminance channel of the frame and the output values associated with the at least one chrominance channel of the frame, a combined representation of the frame. The process can further include generating encoded video data based on the combined representation of the frame.

The invention relates to a low-dose computed tomography (CT) denoising method of a tight frame wavelet, and the method comprises the steps: S1, collecting a low-dose CT picture and a conventional-dose CT picture to construct a CT data set, and converting the CT data set into a Numpy matrix; s2, GTF (Geometric Tigh Framelet) transformation is carried out on the image data to obtain one low-frequency component and eight high-frequency components, the high-frequency components are grouped, and the number of the high-frequency components in each group is 2, 3 and 3; s3, creating a tight frame diffusion model (Tight Frame Diffusion Model), and carrying out the denoising of the low-frequency component; s4, a group high-frequency enhanced network (GHFEN) is created, and the high-frequency component is recovered by the group high-frequency enhanced network (GHFEN), and the high-frequency component of the high-frequency component is recovered by the group high-frequency enhanced network (GHFEN), and the high-frequency component of the high-frequency component is recovered by the group high-frequency enhanced network (GHFEN); and S5, finally, carrying out inverse GTF transformation on the enhanced and denoised components to obtain a denoised image. By using the CT image denoising method and the CT image denoising device, a CT image denoising service which is more in line with a human eye observation effect can be provided for a low-dose CT obtained by a user through a CT scanner. The multi-level feature perception low-dose CT denoising method can be widely applied to the field of CT denoising.

The invention discloses a space-time consistent video depth completion method under a zero sample unified diffusion framework. The method comprises the following steps: constructing a depth completion model comprising a variational auto-encoder, a semantic coding network and a space-time diffusion generation network; preparing training data, and generating a frame-level semantic feature vector and a conditional latent variable fusing an original depth and a relative depth for a video frame; training the space-time diffusion generation network in stages by taking the conditional latent variable sequence as input and the semantic features as conditions; in the inference stage, a video sequence to be complemented is processed through a sliding window fusion mechanism, a de-noising depth latent variable is obtained through a trained network, and finally a complemented depth sequence is output through decoding and scale recovery of a variational auto-encoder. The method has the advantages that a depth sequence with measurement consistency, structural integrity and time stability can be generated when depth completion is performed on a sparse, noisy or structurally damaged long sequence video.

In order to overcome the defects of an existing low-rank model in non-stationary trend, pseudo-periodic mode and transient anomaly processing, the invention provides a time series prediction method based on self-adaptive low-rank representation, original time series data is decomposed into a low-rank component (L) and a sparse component (S) through robust principal component tracking (PCP) and principal component analysis (PCA) algorithms, and the low-rank component (L) and the sparse component (S) are subjected to low-rank prediction. And adaptive separation of a trend-periodic component and a residual (abnormal / transient) component is realized. According to the method, an orthogonal transformation matrix (A) is constructed, a data-driven orthogonal basis (B) and a Fourier orthogonal basis (UF, VF) are fused in the matrix, and time sequence data are mapped to a low-rank potential space with higher characterization capacity. The method adopts a convolution kernel norm minimization (CNNM) frame for prediction, a learnable dynamic search window is established under the frame, the size (wx, n) of the window is adaptively adjusted according to a sequence local feature (| un |), and through fusion of fast Fourier transform (FFT) decomposition and introduced motion vector information, the motion vector information of the motion vector is obtained. And accurate acquisition of multi-scale time sequence characteristics (such as periodic modes of different frequencies) is realized. Under the low-rank constraint, the method provided by the invention can effectively process the complex dynamic characteristics of high-dimensional sensing data, and meanwhile, the prediction precision of a non-stationary sequence is remarkably improved through a parameter adaptive mechanism.

The invention discloses a frame building collapse analysis method based on parametric modeling. The method comprises the steps that 1, the macroscopic size, the component size, the component section reinforcement parameters and the material parameters of a frame building are preset; step 2, presetting grid division parameters; 3, the column dismantling position and dismantling time are preset; 4, generating node, unit, section and component data of the beam, the column and the plate; 5, defining an equivalent material model of the beam; 6, defining an equivalent material model of the board; 7, defining an equivalent material model of the column which is not dismantled; 8, defining an equivalent material model of the dismantled column; step 9, applying gravity and surface load; step 10, defining calculation duration, output step length and other solving control parameters, and generating a K file; and step 11, carrying out collapse calculation. According to the method, the pretreatment efficiency and the collapse calculation efficiency of frame building collapse analysis are greatly improved, the single-working-condition K file generation time is a second level, and the single-working-condition collapse calculation time is a minute level.

An example method of video coding includes receiving a video bitstream that includes a plurality of blocks. The method also includes determining a scanning order for a motion vector list for a first block of the plurality of blocks based on one or more of: a number of neighboring blocks of the current block with corresponding temporal motion vectors, a number of neighboring blocks of the current block encoded in an inter prediction mode, a mode of the current block, and a reference frame index for the current block. The method further includes generating the motion vector list according to the scanning order and identifying, from the motion vector list, a motion vector predictor for the current block. The method also includes decoding the current block using the identified motion vector predictor.

An apparatus directed to improvements to update processing for inverse linear lifting transforms is provided. The apparatus receives a compressed bitstream, decodes a base mesh sub-bitstream to reconstruct a base mesh, decodes a displacements sub-bitstream to reconstruct displacement wavelet coefficients, performs an update process with the displacement wavelet coefficients to generate updated displacement wavelet coefficients, performs a prediction process with the updated displacement wavelet coefficients to generate displacements, and reconstructs a mesh frame based on the displacements. During the update process, the apparatus determines a first and second valence indicating a number of connected edges at a first and second vertices which form an edges to which a current vertex belongs, based on whether the first and second vertices belong to the base mesh, and updates a displacement wavelet coefficient of the first and second vertices based on the first and second valence.

A computer-implemented method for predicting a next frame for a video is provided. Aspects include obtaining the video having a sequence of frames, obtaining a plurality of machine learning models (MLMs) that are trained to perform next frame prediction, and inputting a subset of the sequence of frames into each of the plurality of MLMs. Aspects also include calculating, for the subset of the sequence of frames, a score for each of the plurality of MLMs and generating, using one of the plurality of the MLMs having a highest score, the next frame, where the next frame is a frame immediately subsequent to the subset of the sequence of frames.

The present disclosure provides a computer-implemented method for decoding video. The method includes receiving a bitstream including a coding unit of a video frame, the coding unit being coded in a skip mode or a direct mode; determining whether the coding units has a width greater than a threshold width or a height greater than a threshold height; in response to the coding unit having the width greater than the threshold width or the height greater than the threshold height, enabling a motion vector angular prediction process to store one or more motion vector angular prediction candidates in a candidate list storing a plurality of motion candidates associated with the coding unit; and performing an inter prediction to the coding unit according to the candidate list and an index parsed from the bitstream.

The invention discloses a lightweight hash function construction method and device based on a chaos sponge structure, equipment, a storage medium and a program product, and relates to the technical field of encryption algorithms, and the method comprises an initialization stage, an absorption stage and an extrusion stage. Wherein in the initialization stage, input messages are subjected to standardized filling, and an initialized input sequence is obtained; in the absorption stage, the initialization input sequence and the external part of the internal state are subjected to XOR, and the internal state after XOR is obtained; diffusing the internal state after XOR through the dynamic replacement layer to obtain an absorbed internal state; and an extrusion stage: iteratively extracting Hash output based on the absorbed internal state. Due to the fact that the lightweight hash function which is optimized for low resources and provided with the dynamic replacement layer is designed, the hash function is instantiated through a sponge structure frame, the output length flexibility and the collision attack resistance are kept, and strict cryptographic analysis is supported.

Aspects of the disclosure provide methods and apparatuses for mesh coding. In some examples, an apparatus for mesh coding includes processing circuitry. The processing circuitry receives a bitstream carrying a 3D mesh frame. The 3D mesh frame represents a surface of an object with polygons, the polygons are defined by vertices and edges connecting the vertices, the 3D mesh frame is parameterized into a plurality of 2D charts in a 2D UV atlas. The processing circuitry determines respective chart based coding modes for the plurality of 2D charts according to one or more indices decoded from the bitstream, and decodes, from the bitstream, at least respective geometry information associated with the plurality of 2D charts according to the respective chart based coding modes. The processing circuitry reconstructs the 3D mesh frame according to the respective geometry information associated with the plurality of 2D charts.

The application provides a multiscale geological model cross-scale nesting and fusion modeling method, belonging to the geological field, including four steps of multiscale data preprocessing, data fusion, cross-scale nesting and fusion and dynamic coupling. This method can integrate multi-source heterogeneous data, realize data fusion through joint probability space, variational assimilation framework, fuzzy logic conflict factor and alternating direction multiplier method, utilize deep learning auxiliary fusion technology such as multi-scale convolutional adversarial network to improve the stability and precision of the model. At the same time, the model parameters are determined through volume average upsampling algorithm and random field conditional simulation, and the consistency of the parameters between the models is realized through dynamic adjustment of ensemble Kalman filter and localized ensemble transform Kalman filter, bidirectional feedback and iterative optimization. Finally, through the double grid strategy and the restrictive interpolation, the multiscale convergence is realized, and the convergence and dynamic balance of the model on the scale of kilometers to microns are ensured.

This invention discloses a method and system for generating static code checkers based on a large language model. The method involves: acquiring rule descriptions and a test suite containing positive and negative examples; generating detection logic corresponding to the rule descriptions and their corresponding checker specifications using the large language model; retrieving the corresponding API context based on the detection logic; generating candidate checker code based on the checker specifications, rule descriptions, test suites, and the retrieved API context; validating the candidate checker code based on the test suites and relevant rule descriptions; and obtaining the final checker code that passes the verification. The method proposed in this invention significantly reduces the technical threshold and development cost of customizing checkers for static analysis frameworks, efficiently generating checker code with high accuracy, low false positive rate, and low false negative rate, and possesses good scalability.

The invention provides a complex tubular structure modeling method based on 2D segmented algebraic splines, and relates to the technical field of geometric modeling and computer graphics. The method comprises the following steps: firstly, constructing a section control polygon and generating a high-order algebraic spline primary function; calculating a Frenet frame corresponding to the guide curve, and mapping the section control polygon along the curve to generate a multi-section structure; and constructing an implicit field based on the parameters, and extracting contour surfaces by using a Marching Cubes algorithm to generate a triangular mesh model. The method is small in number of control points, high in structural continuity and suitable for efficient modeling and batch generation of complex tubular structures.

This disclosure generally relates to video coding / decoding, and in particular to enhancing block-adaptive weighted prediction. The method involves receiving a coded video bitstream, which includes the current block of the current frame and a first syntax element indicating the prediction mode of the current block, wherein a plurality of scaling factor lookup tables are stored and include different ranges of scaling factors. In multiple scaling factor lookup tables, the step size or precision of the scaling factor in each lookup table is the same, and A step of determining a prediction mode based on the value of a first syntax element, wherein the prediction mode is used to predict the current block based on the reference block of the reference frame. The steps include determining a scaling factor from one of several scaling factor lookup tables, The process includes the step of reconstructing the current block based on a reference block and a determined scaling factor.

The video temporal sentence localization method based on four constraints and partial supervision comprises the following steps: high-level semantic features are extracted from query text and video by using a deep neural network; a time stamp corresponding to the query event is obtained from the video features by fusing a frame-by-frame algorithm and a candidate frame algorithm, and the video features fused with the query text features are input into an event detector; a four-constraint model is constructed to facilitate learning of a semi-supervised temporal sentence localization task; a partial-full merging framework is used to predict the temporal localization result according to the features, so as to flexibly handle two types of supervision, i.e., partial supervision and full supervision; partial supervision branch optimization and full supervision branch optimization are performed, the partial supervision branch optimization comprises optimization of the four-constraint model and optimization of a localization pseudo-label localization loss constructed according to partial supervision information; and the full supervision branch optimization is performed by using a localization learning framework and relying on supervised learning optimization of the localization pseudo-label.

The invention relates to the field of 3D data compression, in particular to a 3D video sequence data compression, decompression and transmission method and electronic equipment, and the compression method comprises the steps: obtaining RGB-D sequence frame data, and setting a label for each frame of RGB-D sequence frame data; the RGB-D sequence frame data comprises RGB texture sequence frame data and 32-bit depth sequence frame data; selecting RGB-D sequence frame data of a corresponding frame based on the label, and compressing the RGB-D sequence frame data to obtain RGB-D sequence frame compressed data with the label; the compression comprises the following steps: compressing the RGB texture sequence frame data by adopting a standard video compression protocol to obtain RGB texture sequence frame compression data; and compressing the 32-bit depth sequence frame data by adopting a 32-bit half-truncation compression frame to obtain 32-bit depth sequence frame compressed data. Through a 32-bit half-truncation compression technology, a high compression ratio is realized on the premise of ensuring depth data precision, the storage and transmission pressure of massive 3D video data is effectively relieved, and real-time application is supported.

This invention discloses a method and system for order reduction and decoupling of high-dimensional stochastic systems based on stochastic approximation conjugacy, belonging to the field of complex system modeling and analysis technology. This method acquires multi-condition state data of the high-dimensional system, constructs a reduced-order analysis framework including embedding mapping, a low-dimensional decoupled dynamic model, and lifting mapping. Using stochastic approximation conjugacy as a constraint, it minimizes the total loss function to complete joint model training. After solving the system's stochastic response characteristics within a low-dimensional feature manifold, it maps back to the high-dimensional space to complete dynamic behavior analysis. The corresponding system includes modules for data acquisition, model construction, model training, low-dimensional solution, and mapping analysis. This invention solves the curse of dimensionality problem of traditional methods, ensures the reliability and interpretability of the reduced-order model, has strong versatility, and can be widely applied to intelligent operation and maintenance and control optimization of major engineering systems.

Here we describe a method to improve mesh reconstruction by repositioning vertices on the boundary of patches to ensure there are no gaps between adjacent patches (also known as zippering). We disclose six different ways to implement the post-processing operation, as well as syntax elements and semantics for the transmission of filter parameters. A hierarchical method indicates geometric distortions that may cause gaps between patches. A value is transmitted per frame, or per patch, or per boundary object. The number of bits to encode the value also depends on the previous geometric distortion. Another method is to transmit an index match instead of a geometric distortion. A matching index is transmitted for each boundary vertex, but a method of transmitting only one index of a pair is also implemented.

The invention discloses a method and a device for realizing reliability evaluation of an aero-engine, a double-framework model comprising a structural BN and a dependent BN is constructed based on a model framework such as BN, a recursive algorithm uses the condition independence of the BN to reduce the complexity to a pseudo polynomial level, and the calculation efficiency is remarkably improved. In addition, large deviation caused by a traditional independent hypothesis is avoided, and the evaluation precision is improved. According to the embodiment of the invention, any layered structure and cross-module dependence are supported, key risk path analysis can be output, and technical support is provided for efficient implementation of aero-engine design optimization, predictive maintenance and airworthiness certification.

Post training compression of 3DGS data is agnostic to training in a traditional signal compression perspective. Gaussian parameters are treated as signals. Pre-processing and transform coding techniques are used to compress the signals effectively. Firstly, lossless / lossy compression is performed on 3DGS geometry (positions, scales, rotations) using a point cloud coding-based (e.g., G-PCC, GeS) framework. Positions are compressed using occupancy tree coding. Scales and rotations are encoded as attributes using transform coding. The widely used block-based graph Fourier transform (GFT) is used to compress the attributes (base colors, spherical harmonic coefficients and opacities). In addition, a graph construction strategy is used for 3DGS data that computes the edge weights based on similarity (or dissimilarity) between the 3D Gaussian distributions using KL-divergence. Alternatively, positions can be encoded using occupancy tree (e.g., G-PCC, GeS) or AI-based PCC methods, and any subset of Gaussian parameters or the transformed coefficients of Gaussian parameters can be mapped into 2D frames and encoded by video coders.

A system and method for joint coding of motion vector differentials using template matching-based scaling factor derivation includes receiving a current frame in a video bitstream, determining that a current block in the current frame is coded in joint motion vector differential (JMVD) mode, selecting a first adjacent reconstructed sample and a second adjacent reconstructed sample of the current block as template regions to be used to predict the current block in the JVMD mode, determining a predictive block from a reference frame based on a scaling factor derived from the selected template region and applied to a motion vector differential (MVD) associated with the current block, and reconstructing the current block in the JVMD mode based on at least the predictive block.

A frame processor may generate a mask based on one or more static regions of a first set of frames of a plurality of previous frame and adjust the mask to at least one of determine alpha data or conceal distorted content associated with the one or more static regions of the first set of frames. The distorted content may be caused by extrapolation of a frame from a second set of frames of the plurality of previous frames. The frame processor may generate a composite frame based on application of at least one of the mask or the alpha data to a previous frame of the plurality of frames, and application of the previous frame based on the at least one of the mask or the alpha data to the frame extrapolated from the second set of frames of the plurality of previous frames.

The invention provides a dynamic grid animation coding method, a decoding method, a coding device and a decoding device based on a depth entropy model.The dynamic grid animation coding method comprises the steps that A1, vertex position offset of a grid animation frame sequence is coded, and a prediction model is constructed through vertex position information of a preorder frame; a2, based on the vertex position information of the preorder frame, dividing the vertex into a base layer and an inference layer by adopting hierarchical detail layering and uniform sampling layering; a3, for the vertexes in each layer, using inverse distance weighted interpolation based on nearest neighbors to predict the offset of the vertexes, and calculating a prediction residual error; a4, carrying out the coding of the prediction residual error of the vertex of the base layer through run-length coding; and coding the inference layer vertex by using a depth attention model based on an adaptive position descriptor. According to the method disclosed by the invention, the vertex position offset of two adjacent frames of animation sequences is coded in combination with a coding framework of deep learning, so that the compression performance of the grid animation is improved.

The invention provides a parameterized swept curved surface design method, and relates to the field of aircraft design. The method comprises the following steps of: processing a two-dimensional graph by adopting a sequence-preserving and feature-preserving discrete sampling method, so that a sampling sequence has consistent point sequence index rules and adjacency relations while keeping geometric details; under a normalized coordinate system, a measurement base is constructed by using the reference graph and the coordinate sequence of the interpolation center, parameterization of other graphs is completed, and an analysis framework with consistent meaning is formed; graphic shape change is achieved through coordinate transformation based on the frame, and smooth deformation between complex graphs is guaranteed with an S-shaped curve with the acceleration at the two ends being 0 as a control rule; and finally, describing shape change, size change and scanning path design through a parameterized curve or a discrete point list, and realizing parameterized generation of a scanning curved surface by combining parameter coordinates of each section graph and different section assembly strategies. By adopting the method, different types of sweeping requirements can be flexibly met.

The application discloses a kind of multi-agent preset time formation control methods based on backstepping method under distance rigidity matrix, comprising: 1) based on the double integrator dynamics model of multi-agent system is established based on rigidity graph theory, define the minimum and infinitesimal rigidity frame corresponding to the expected formation;2) introduce backstepping design framework, define virtual speed error variable;3) introduce time-varying scaling function to construct time scale transformation;4) design distributed preset time controller;5) through numerical simulation and physical platform are verified, verify the effectiveness of distance error in user preset time under different initial conditions converges to the vicinity of zero.The application is fused by backstepping and preset time scaling function, for the first time in only local relative measurement distance rigidity formation Convergence time user self-defined preset is realized, and convergence time upper bound is independent of initial condition and controller parameter, provides reliable, engineering implementation solution for time-critical multi-agent cooperative task.

The invention relates to a coding representation, retrieval and matching method of a lead frame design model, and belongs to the field of computer aided design (CAD). According to the method, on the basis of text readability of a DXF format, a deep learning model DualDXF is used for carrying out overall coding representation on an LF design model, so that the retrieval problem of 2D LF design is solved; in addition, the invention further provides an effective LF design matching method which is used for comparing two similar LF design models and visually marking the differences in the LF design of the DXF format. According to the method, the problem of input dimension disasters when a deep learning model retrieves high-complexity LF design is solved, and retrieval of similar LF design can be completed without manual annotation.