57 results about "Video reconstruction" patented technology

A system and method for latent geodesic traversal across multi-axis hyperspaces for real-time video reconstruction and augmentation. Spatiotemporal video data are compressed into navigable latent representations using hierarchical and Lorentzian autoencoders that preserve geometric and temporal structure. A geodesic traversal engine computes paths across spatial, temporal, spectral, and semantic axes, guided by symbolic anchors and spatiotemporal routing protocols. A correlation network restores fine detail, while an augmentation generator synthesizes additional or counterfactual content to enable infinite zoom, continuous multi-scale exploration, and temporally coherent augmentation. A strategy caching system preserves successful traversal patterns for reuse, supporting persistent learning and adaptive real-time performance.

The invention relates to a video processing method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises: acquiring an original video; tracking and detecting a target object in the original video to obtain an original motion track of the target object in the original video; obtaining a target picture size and a target shooting style; determining a picture extension parameter according to the original motion track, the target picture size and the original picture size of the original video; performing picture expansion on the original video according to the picture expansion parameter to obtain an expanded video; determining a cutting frame sequence according to the target movement track and the mapped movement track; and according to the cutting frame sequence, performing picture cutting on the expanded video according to the target picture size to obtain a target video of the original video in the target shooting style. By adopting the method, the efficiency and the effect of video reconstruction images can be improved.

The invention relates to a high-fidelity generative video stream transmission system based on a visual base model, which belongs to the field of image communication, and is characterized in that a visual enhancement-oriented generative codec is designed, and high-fidelity video reconstruction under a high compression ratio is realized through an asymmetric space-time compression strategy and time sequence consistency enhancement; a resolution scaling module is provided, the calculation complexity is remarkably reduced through a video super-resolution recovery module of adaptive resolution control and joint optimization, and real-time high-definition video processing is achieved; and constructing a network adaptive video stream transmission controller, an intelligent token discarding mechanism based on semantic importance and a mixed packet loss processing strategy to realize code rate scalable control and robust transmission under network fluctuation.

The invention discloses a bidirectional adaptive video super-resolution method based on frame difficulty index, and belongs to the field of computer vision. The invention provides a frame-level dynamic reconstruction method for solving the problems that simple frame calculation is redundant and difficult frame reconstruction is insufficient due to the fact that an existing model adopts a fixed calculation strategy for video frames with different difficulties. According to the method, a motion detail decoupling propagation network is constructed, motion information is efficiently transmitted by utilizing a shallow forward propagation branch, and a deep backward propagation branch focuses on recovering texture details so as to decouple a time sequence propagation task; meanwhile, a frame reconstruction difficulty evaluation network is introduced to generate a global difficulty index, so that the receptive field weight of the adaptive time sequence fusion network and the refining depth of the dynamic refining network are regulated and controlled. According to the method, through explicit modeling frame-level reconstruction difficulty, adaptive matching of the model capacity and the video frame feature complexity is realized, and the video reconstruction performance is remarkably improved under limited computing power.

The invention provides a video coding method and device and a video decoding method and device, and relates to the technical field of video coding and decoding. Comprises: obtaining a student model; the student model is a model obtained by guiding a first network model to train through a teacher model, the teacher model is a model obtained by training a second network model, and when the same video is coded based on the first network model and the second network model respectively, the first network model and the second network model are selected; the code rate of the coding result of the first network model is smaller than that of the coding result of the second network model; and obtaining a coding result of the video to be coded according to the student model. According to some embodiments of the invention, in a video coding scheme for carrying out video coding by using a deep learning network model, the video reconstruction quality is ensured, and the code rate of a video coding result is reduced at the same time.

The invention discloses a loop filtering method and device for reconstructing a video. The method receives input data of a current block, wherein the input data includes reconstructed samples of the current block. At least two loop filters are applied to a current block, where the at least two loop filters belong to a loop filter bank comprising a bilateral filter (BIF) and an adaptive loop filter (ALF), and the classification processes of the at least two loop filters share one input source, one or more classification rules, one or more processing units or a combination thereof. A filtered output generated by applying the at least two loop filters to the current block is provided.

This invention relates to the field of infrared imaging and low-level hardware processing technology, and discloses a parallel processing method and circuit for infrared image data based on FPGA. The method includes: establishing a pixel spatiotemporal mapping coordinate system and extracting pixel mapping point coordinate pairs; performing hardware topology sensing and spatial decoupling to output a spatial topology matrix; performing real-time pixel-level non-uniformity correction; implementing phase-locked noise reduction through a virtual offset field to obtain an equalized corrected bitstream; generating a quantization mapping curve based on saliency sensing; and performing real-time video reconstruction combined with adaptive power consumption control to output a video signal. This invention solves the problems of latency and instantaneous heat accumulation in large-area infrared data processing, achieving high-fidelity restoration of spatial topology and suppression of non-uniform noise; closed-loop energy management eliminates image blurring caused by thermal drift, comprehensively improving the system's imaging clarity and hardware operating efficiency.

The invention discloses a construction method of a video reconstruction system of a joint information source channel, the video reconstruction system, equipment and a medium. At a transmitting end, the method comprises the following steps: performing multi-frame joint semantic coding on a video sequence, extracting a potential representation simultaneously containing spatial information, time information and high-level semantic information, and directly mapping the potential representation to a wireless channel for transmission; and at a receiving end, a diffusion generation model is constructed based on the denoising network and the diffusion model, and semantic features are introduced to carry out condition guidance on the diffusion denoising process, so that generative reconstruction of the potential representation damaged by noise is realized. According to the method, the diffusion generation model is integrated into a deep joint source channel coding framework, so that the video reconstruction quality and semantic consistency are remarkably improved in a low signal-to-noise ratio and complex channel environment, the cliff effect is effectively weakened, and higher robustness and adaptive ability are achieved.

A system and method for generating synthetic video from diverse sensor inputs within a unified computational framework. The system receives heterogeneous data such as acoustic, thermal, and textual streams, encodes each into modality-specific latent representations, and projects them into a shared geometric manifold. Within this manifold, convergence points known as multimodal landmarks are established and used to compute geodesic trajectories that describe relationships among the inputs. The trajectories are verified for reversibility to ensure that forward and reverse mappings remain consistent. A Lorentzian autoencoder then decodes the validated trajectories into temporally coherent video sequences derived from the multimodal evidence rather than reconstructed imagery. The system records geometric states for auditability and persistently stores the resulting landmarks and trajectories for reuse, enabling reversible, verifiable generation of synthetic video that accurately reflects the integrated sensor data.

The application relates to the field of film and television rendering, in particular to a configuration decision optimization system and method for film and television rendering, which comprises a shooting calibration module, an intra-frame rendering module, a scene reconstruction module, a video rendering module and a configuration compression module; the shooting calibration module is used for video de-jittering and adjusting a picture center; the intra-frame rendering module is used for area light rendering; the scene reconstruction module is used for generating a point cloud model; the video rendering module is used for merging a rendered video; and the configuration compression module is used for compressing a video stream; the application can ensure the definition and consistency of an output picture, reduce high-frequency artifacts and fuzzy sawteeth caused by shooting angle problems, reduce video rendering result jitter, improve image rendering quality, solve a sampling rate deficiency problem, improve the scale of multi-path video rendering, improve video reconstruction speed and rendering efficiency, and realize efficient video compression and transmission.

The application discloses a video compression transmission method and related equipment, the method comprises the following steps: obtaining a video sequence to be compressed, taking the first frame and the last frame as conditional frames; inputting the video sequence into the encoder of a pre-trained variational autoencoder to obtain a target latent space representation; processing the target latent space representation through the downsampling module of the compressor to generate an extreme compression representation; transmitting the conditional frames and the extreme compression representation to the receiving end to enable the receiving end to reconstruct the video through the upsampling module of the compressor, the generation model and the decoder of the variational autoencoder to obtain a reconstructed video sequence; the application provides time sequence boundary information through the conditional frames, and in combination with the reconstruction capability of the generation model, can effectively reduce the block effect, blur and high-frequency detail loss commonly seen in traditional methods; the conditional frames and the extreme compression representation significantly reduce the transmission code rate and bandwidth demand, can significantly improve the rate distortion performance, and can be widely applied to the technical field of video compression.

This invention proposes a video reconstruction method and system based on prior features and global frequency domain filtering, relating to the field of video coding technology. The method involves inputting video frames into a pre-trained encoder to extract multi-scale features, and progressively fusing these deep features to obtain the first prior feature. Hybrid residual features are then extracted from the video frames based on a hybrid residual grid. The first prior feature and the hybrid residual feature are input into convolutional blocks respectively to obtain the second prior feature and residual grid features. These two features are then fused based on adaptive weights to obtain a fused feature containing both prior and grid information. The fused feature is input into a coupled mapping RNN module to obtain state features, which are then input into a global frequency domain filtering upsampling module for video frame reconstruction, resulting in reconstructed video frames containing information at different frequencies. This invention improves the high-quality reconstruction performance of the implicit neural video representation model by enhancing the representational power of intermediate features and the decoding power of the upsampling decoding module.

The invention provides a video self-encoding method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence, and is suitable for the financial field and the medical field. The method comprises the following steps: carrying out pixel preprocessing on each image frame of an original video, and carrying out down-sampling on obtained initial image features to obtain down-sampled image features; performing feature coding on the down-sampling image feature and the first memory feature of the current image frame to obtain a coded image feature of the current image frame; performing feature decoding on the coded image feature and the second memory feature of the current image frame to obtain a decoded image feature of the current image frame; performing up-sampling on the decoded image features to obtain up-sampled image features; and carrying out pixel post-processing on the up-sampling image features of all the image frames to obtain a target video. According to the method, the calculation complexity can be reduced, the video reconstruction accuracy can be improved, and particularly, the situation that jitter artifacts occur in the reconstructed video can be remarkably reduced.

The invention discloses an event camera video reconstruction method and system based on active aperture modulation, and the method comprises the steps: introducing an aperture modulation strategy for the first time, and reconstructing an initial frame with good quality by periodically adjusting the opening and closing of an aperture and actively triggering a dense global event signal; the problem of low background reconstruction quality caused by sparse events in a static region in the prior art is solved, and good prior information is provided for subsequent dynamic scene reconstruction. By constructing a forward-reverse bidirectional network, rich intensity information is provided for a static scene, and the problems of background disappearance and error accumulation in long-time operation in a traditional method are effectively solved. And meanwhile, high-time-resolution capture of a dynamic region is kept, and high-fidelity and high-dynamic-range video reconstruction is realized.

The invention discloses a real-time video super-resolution method based on multi-scale space-time motion estimation, and the method comprises the steps: carrying out the feature extraction of a low-resolution video frame through convolution operation, and obtaining a basic visual feature; performing efficient fusion on the spatial features by using an encoder module to generate multi-scale visual feature representation; through a space-time motion refinement module, space consistency alignment of motion is realized on visual features of different space scales, and meanwhile, motion estimation of a current frame is iteratively updated based on a motion estimation result of a historical frame, so that time sequence coherence of the motion is realized; a motion alignment decoder module is adopted, the reconstruction features of the historical frame layer and the visual features of the current frame layer are fused, the reconstruction features of the previous layer are refined, and the high-quality reconstruction features of the current frame are obtained; and adopting reconstruction operation to generate a final super-resolution image frame according to the final reconstruction feature. According to the method, low delay and calculation efficiency can be considered, and the visual quality and detail representation of video reconstruction are improved.

The present invention can be applied to the technical field of computer vision and graphics. Provided are a method for disentangled reconstruction of a dynamic digital human, and an electronic device and a storage medium. The method comprises: using a technique for reconstructing three-dimensional shapes of a human body and garments from a monocular human body video, and using a representation method in which explicit geometry is combined with an implicit signed distance field (hmSDF), such that high-quality reconstruction of a dynamic disentangled digital human from a monocular video can be achieved. In the method, garments and a human body are separated and separately modeled, optimized hmSDF is used to achieve accurate segmentation of visible regions, and an SMPL model is also used to complete occluded human body regions, thereby ensuring the consistency and fidelity of the overall geometry. A linear blend skinning (LBS) deformation field and a non-rigid deformation field are used to capture human body motion and detailed variations, such that high-fidelity and continuous spatio-temporally disentangled human body and garment geometry is ultimately generated.

The application discloses a three-dimensional dynamic human body light field content generation method, and relates to the field of dynamic human body reconstruction. The method first reconstructs an implicit model according to input multi-view color video to obtain a drivable parameterized template; then, a Gaussian graph is predicted by a generation network based on a posture condition, and three-dimensional Gaussian points in a canonical space are transformed to a target posture space by linear mixed skinning; then, template parameters, neural network weights and Gaussian attributes are jointly optimized according to input multi-view images to obtain a drivable human body Gaussian model; finally, an efficient de-redundancy light field encoding algorithm is used to realize generation and three-dimensional display of three-dimensional dynamic human body light field content. The method can maintain high-quality human body reconstruction while significantly improving the generation efficiency and rendering performance of dynamic human body light field content, and is suitable for digital human generation, human-computer interaction, virtual reality and three-dimensional light field display.

The invention discloses an end-to-end snapshot compression computer vision method based on a pseudo-random mask array. The method comprises the following steps of: performing single exposure on a video sequence subjected to space-time light intensity modulation by using a two-dimensional image sensor and a pseudo-random binary mask to obtain a corresponding two-dimensional (2D) compression measurement value; a video reconstruction is taken as a proxy task, a trained compression denoising auto-encoder model is utilized to extract potential spatial-temporal feature representations from the 2D compression measurement values, and the compression denoising auto-encoder model comprises a shared encoder and at least one task-specific decoder. The shared encoder is configured to extract the potential spatio-temporal feature representation from the 2D compressed measurements; and inputting the potential spatio-temporal feature representation into the decoder to obtain a dynamic result of the downstream computer vision task. According to the method, the image reconstruction complexity and the calculation cost are remarkably reduced, the excellent performance can still be kept especially in an extremely low illumination scene, and the privacy protection characteristic is achieved.

The application relates to the technical field of computer vision, and discloses a video compression method, a video compression system, a terminal and a storage medium for cross-domain online learning, the method comprising: constructing corresponding latent motion features for each frame of image data of a target video to construct a multi-scale time context of a current frame; encoding the target video by using the multi-scale time context to obtain a latent representation of the current frame; fine-tuning a weighting factor according to a distortion difference of the current frame, and adding the weighting factor to a target loss function to fine-tune the latent representation to obtain an updated latent representation; and decoding the updated latent representation to output image features of the target video. When facing a test video of a non-natural scene or an unknown distribution, the application greatly enhances the cross-domain generalization capability of a neural video compression model, can significantly improve rate-distortion performance, and realizes lower bit rate and higher video reconstruction quality.

The invention provides an ultrasonic image reconstruction method. The method comprises the following steps: taking original radio frequency data received by an ultrasonic probe as an image reconstruction object; firstly, a sound field characteristic matrix of an ultrasonic probe in a two-dimensional plane is obtained through measurement of a hydrophone; according to the sound field characteristic matrix, a pulse-echo matrix corresponding to a two-dimensional plane is obtained through a classic sound field propagation rule, and probe excitation signals, probe response characteristics and sound field distribution prior are introduced into a reconstruction process; then, through a classic ultrasonic scattering echo model, a mapping relation among ultrasonic echo data, a pulse-echo matrix and an ultrasonic two-dimensional image is obtained; then, by introducing a multi-frame shared sparse constraint and optimization method, high-speed video reconstruction is realized instead of traditional single-frame reconstruction; the method has the beneficial effects that the signal-to-noise ratio and the signal-to-background ratio are remarkably improved, and meanwhile, side lobe and grating lobe artifacts are reduced.

This invention discloses a model training method and apparatus based on images and text. The method includes: performing frame extraction on a training video to obtain multiple training video frames; determining a target representation frame image from the multiple training video frames; copying the target representation frame image to obtain multiple copied representation frame images; inputting the multiple copied representation frame images and descriptive text into a video reconstruction prediction model for training; calculating the video loss function value and text loss function value output by the video reconstruction prediction model during training; and optimizing the model parameters of the video reconstruction prediction model based on the video loss function value and text loss function value until convergence, thereby obtaining a trained video reconstruction prediction model. It is evident that this invention enables the model to learn the image relationship between specific video frames and the entire video, as well as the text relationship with the descriptive text, during training, thereby improving the video reconstruction capability of the finally trained model.

This electronic device may comprise: a memory storing instructions; and at least one processor comprising a processing circuit. The instructions, when executed individually or collectively by the at least one processor, may cause the electronic device to: receive an input for generating a second video reconstructed from a first video; on the basis of the input, acquire, from the first video, first motion data of a first visual object within the first video and second motion data of a second visual object within the first video; and on the basis of the first motion data that is greater than the second motion data, generate the second video representing the first visual object moving in front of the second visual object.

This invention provides a video reconstruction method, apparatus, system, terminal, and storage medium. The method includes: receiving current frame encoding features obtained by encoding a target video at an encoding end; decoding the current frame encoding features to obtain current frame reconstruction features; acquiring target reconstruction features corresponding to at least one encoded frame in the target video; extracting semantic features from the target reconstruction features; inputting the current frame reconstruction features into a preset diffusion model, and using the semantic features as conditional information of the diffusion model to generate enhanced reconstruction features; and obtaining the current reconstructed video frame based on the enhanced reconstruction features. This application requires no additional text or auxiliary encoder. The semantic features dynamically change with the video content, providing more precise guidance. The semantic features contain high-level semantics and are spatially aligned with the reconstruction features, resulting in more realistic and consistent details in the generated content, effectively avoiding semantic drift, and thus improving the accuracy of video reconstruction.

The present application relates to the technical field of video encoding / decoding, and provides a video representation method, a video construction method, and an apparatus. The video representation method comprises: creating an information grid corresponding to each video frame of a target video sequence; and cyclically using each video frame as a current video frame to execute the following steps until a training stop condition is met: acquiring a first hidden state on the basis of the information grid corresponding to the current video frame, a first recurrent neural network, and a hidden state of a video frame previous to the current video frame; acquiring a reconstructed video frame on the basis of the first hidden state and a reconstruction network; on the basis of the current video frame, the reconstructed video frame, and a preset loss function, adjusting at least one of the parameter of the information grid, the parameter of the first recurrent neural network, and the parameter of the reconstruction network; and when the training stop condition is met, generating representation data of the target video sequence on the basis of the information grid corresponding to each video frame, the first recurrent neural network, and the reconstruction network. Some embodiments of the present application are used for improving the representation efficiency of videos.

The invention discloses an sRGB-to-RAW video reconstruction method and system based on compressed key frame guidance, and relates to the technical field of video reconstruction. The method comprises the following steps: designing a key frame selection strategy based on a real-world RAW-sRGB pairwise data set, and dynamically selecting a proper key frame; designing an RAW compression reconstruction module, a time domain alignment module, a space-frequency combined mask module and an sRGB guided refined recovery module, and constructing an sRGB-to-RAW video reconstruction model based on compressed key frame guidance; designing a loss function, training an sRGB-to-RAW video reconstruction model by using a deep learning Pytorch framework, and repeatedly traversing a data set until the model is converged; and inputting the RAW key frame and the sRGB video sequence into the trained sRGB-to-RAW video reconstruction model to obtain a reconstructed RAW video sequence. According to the method disclosed by the invention, the high-quality RAW video is reconstructed under the condition that the storage is limited so as to be used for subsequent video editing or visual tasks.

The application relates to a real-world video super-resolution method, system, device and medium, which comprises the following steps: inputting an original video sequence; video embedding: extracting the spatial features of each frame from the original video sequence to obtain first features, and inputting the first features into a double-axis space-time attention mechanism module to obtain second features; wherein the double-axis space-time attention mechanism module comprises a vertical-time attention block and a horizontal-time attention block, the feature blocks generated after the first features are processed through rotation position coding are converted into token sequences, and the token sequences are rearranged and then respectively sent into the vertical-time attention block and the horizontal-time attention block to simulate spatial texture and motion characteristics; space-time reconstruction: time attention is adopted to integrate time information, and a video output with higher space-time quality is generated. Compared with the prior art, the application has the advantages of good video reconstruction quality, high robustness and low cost.

The invention discloses a video generation method and device, computer equipment and a storage medium, belongs to the technical field of artificial intelligence, and is applied to the financial field or the health medical field. In the tensor algebraic level, high-order tensor decomposition is carried out on an input image by using a transformation tensor product, and the input image is projected to an orthogonal potential subspace. Then, in a feature processing process, weighting the space factor matrix by adopting a local attention window, and combining with a preset mask matrix constraint; a recursive converter with orthogonal constraints is introduced to process a time sequence factor matrix, and it is ensured that the hidden state sequence is stably expressed in long-time dependence modeling. And finally, in a video reconstruction stage, performing up-sampling and reconstruction on the fusion tensor by adopting a tensor completion algorithm based on nuclear norm constraint and a low-rank synthesis strategy. According to the method, the unification of spatial decoupling, time sequence stability and global low-rank performance is realized, so that the generated video reaches a better level in the aspects of spatial resolution, time sequence dynamics and overall visual quality.

The invention provides a video discrete coding length adaptive adjustment method and system, and the method comprises the steps: obtaining an original video frame sequence to obtain a space-time fusion video image block sequence and text data, and obtaining a first visual pooling feature; constructing an initial policy network based on preset policy network parameters; screening the video image block sequence based on the network to obtain a first video image block sequence and reconstruct a first video frame sequence; and establishing a first loss function and a first reward function, updating the strategy network, obtaining a strategy network parameter, and obtaining a second video image block sequence if a preset stop condition is reached, thereby realizing self-adaption of the video discrete coding length. According to the video discrete coding length adaptive adjustment method and system provided by the invention, redundancy or information loss caused by a fixed coding length is avoided, the balance between the video reconstruction quality and the compression ratio is realized, and adaptive adjustment of the video discrete coding length is achieved.

A system and method for latent geodesic traversal across multi-axis hyperspaces for real-time video reconstruction and augmentation. Spatiotemporal video data are compressed into navigable latent representations using hierarchical and Lorentzian autoencoders that preserve geometric and temporal structure. A geodesic traversal engine computes paths across spatial, temporal, spectral, and semantic axes, guided by symbolic anchors and spatiotemporal routing protocols. A correlation network restores fine detail, while an augmentation generator synthesizes additional or counterfactual content to enable infinite zoom, continuous multi-scale exploration, and temporally coherent augmentation. A strategy caching system preserves successful traversal patterns for reuse, supporting persistent learning and adaptive real-time performance.

The application discloses a kind of double-branch HDR video reconstruction methods based on Raw domain, belong to video signal processing technical field;A kind of double-branch HDR video reconstruction methods based on Raw domain, specifically includes the following steps: S1, establishes the Raw domain video HDR dataset of synthesis;S2, designs double-branch Raw video HDR reconstruction algorithm based on S1;S3, training model;S4, the low dynamic range (LDR) of Raw video sequence in test set is input into model, and the corresponding high dynamic range output result is obtained;The first video HDR dataset of simulating real noise distribution is synthesized in Raw domain in the application, and the training and evaluation of HDR reconstruction method under night or extreme scene are provided with benchmark dataset;Meanwhile, the dynamic range expansion of noisy LDR video under difficult scene is realized using the proposed content enhancement module.