CN112533026A - Video frame interpolation method based on convolutional neural network - Google Patents
Video frame interpolation method based on convolutional neural network Download PDFInfo
- Publication number
- CN112533026A CN112533026A CN202011364443.0A CN202011364443A CN112533026A CN 112533026 A CN112533026 A CN 112533026A CN 202011364443 A CN202011364443 A CN 202011364443A CN 112533026 A CN112533026 A CN 112533026A
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- video frame
- video
- frame interpolation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/234—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs
- H04N21/2343—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving reformatting operations of video signals for distribution or compliance with end-user requests or end-user device requirements
- H04N21/234381—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving reformatting operations of video signals for distribution or compliance with end-user requests or end-user device requirements by altering the temporal resolution, e.g. decreasing the frame rate by frame skipping
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
- H04N21/4402—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving reformatting operations of video signals for household redistribution, storage or real-time display
- H04N21/440281—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving reformatting operations of video signals for household redistribution, storage or real-time display by altering the temporal resolution, e.g. by frame skipping
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Television Systems (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention provides a video frame interpolation method based on a convolutional neural network, which comprises the following steps: firstly, acquiring video continuous frames, inputting the video continuous frames into a convolutional neural network for down-sampling and up-sampling processing, and simultaneously extracting and outputting video frame interpolation characteristics; then circularly outputting the intermediate frame through the LSTM convolutional layer; and then calculating the mean square error of the optical flow and taking the mean square error as an optimization objective function of the video non-uniform frame interpolation calculation to realize the optimization of the video frame interpolation. The invention avoids the influence of the motion estimation process in the existing frame interpolation method on the frame interpolation quality, and can directly output the intermediate frame through the deep convolutional neural network.
Description
Technical Field
The invention belongs to the technical field of computer image processing, and particularly relates to a non-uniform video frame interpolation method.
Background
The video frame rate conversion technology is a technology for reconstructing intermediate frames by utilizing relevant information between two adjacent frames in a video and applying an interpolation method. The technology can remove redundant information in coding, reduce the frame rate in the video transmission process and reduce the data volume transmitted by a video network, so the technology can be applied to video compression or video continuity enhancement.
The conventional video frame interpolation method mainly comprises two steps, namely optical flow estimation and pixel synthesis. The effect of the video frame interpolation technology in the method is often dependent on the quality of the optical flow estimation, and the process of the optical flow estimation is easily affected by occlusion and blurring to generate obvious errors. With the development of deep learning, a video frame interpolation technology based on the deep learning has a new breakthrough, and the attempt of video frame interpolation by using a convolutional neural network is successful to a certain extent.
The video frame interpolation technology is to use the related information between adjacent front and back frames in a video to obtain an intermediate frame by an interpolation method. The purpose of video frame interpolation is to synthesize new intermediate frames in a video and improve the frame rate of the video.
Video interpolation frames can be divided into uniform interpolation frames and non-uniform interpolation frames according to the relation between the number of new interpolation frames and the number of input video frames. The uniform frame interpolation refers to that a new interpolated frame and an input video frame sequence are synthesized into a final new video sequence according to a ratio of 1:1, and the non-uniform frame interpolation refers to that a new interpolated frame and an input video sequence are synthesized into a new video sequence according to a certain ratio.
The traditional video frame interpolation technology mainly finds out an obvious corresponding relation between pixels of two frames of images before and after a video, and the most common method is to acquire optical flow information between the two frames before and after the video. And taking the information of the optical flow field as the corresponding relation between the front frame image and the rear frame image of the video, and synthesizing the intermediate frame image by using the information of the optical flow field. In the conventional method, the quality of the inserted frame depends on the quality of the optical flow field information to a great extent.
In the video frame interpolation technology, motion estimation plays an important role, and besides directly finding out the motion relationship between two adjacent frame images, methods capable of replacing the motion information of the adjacent frames to be directly estimated are continuously generated, the methods are all based on the adaptation of a phase method, and the main idea is to encode the motion part in the phase difference between the adjacent video frame images.
Disclosure of Invention
In order to avoid the influence of the motion estimation process between two adjacent frames on the quality of an inserted frame, the invention aims to provide a video frame inserting method based on a convolutional neural network.
The technical solution of the invention is as follows:
a video frame interpolation method based on a convolutional neural network comprises the following steps:
1) acquiring continuous frames of a video:
selecting related front and rear frames from the real video frames, normalizing the front and rear frames, and inputting the normalized front and rear frames into a convolutional neural network;
2) extracting video motion information and restoring video space:
the convolutional neural network front half part encoding module performs down-sampling processing on the front and rear frames after normalization processing, and extracts motion information between the front and rear frames; then, a second half decoding module of the convolutional neural network performs up-sampling processing on the motion information between the videos subjected to down-sampling processing, recovers the spatial dimension of the videos and compensates details; meanwhile, the middle part of the convolutional neural network transmits the information of the network bottom layer to a deep network in a skip-connection mode, and extracts and outputs the video frame interpolation characteristics;
3) outputting a plurality of intermediate frames:
inputting the information subjected to the upsampling processing in the step 2 and the video frame interpolation characteristics into the bidirectional LSTM convolutional layer, and outputting at least one intermediate video frame in a circulating manner;
4) calculating the mean square error of the optical flow:
respectively calculating the intermediate video frame optical flow and the real video frame optical flow in the step 3 through the pre-trained FlowNet; then calculating the mean square error between the intermediate video frame optical flow and the real video frame optical flow;
5) optimizing the frame insertion:
and 4, taking the mean square error in the step 4 as an optimization objective function of the non-uniform video frame interpolation calculation, and enabling the FlowNet to participate in the gradient backward propagation process in network optimization to realize the optimization of the video frame interpolation.
In the step 2, the convolutional neural network front half part encoding module preferably performs two times of downsampling processing on the front and rear frames after the normalization processing; and the second half decoding module of the convolutional neural network correspondingly performs two times of up-sampling processing on the motion information between the videos after the down-sampling processing.
The convolutional neural network is preferably an encoding and decoding U-Net module, and the encoding module is a 3-layer convolutional module.
Before upsampling and downsampling, a dense connection network can also be used for convolution feature extraction respectively.
The up-sampling adopts a sub-pixel algorithm; the down-sampling described above employs the Conv-LSTM algorithm.
In the step 2, the middle part of the convolutional neural network transmits the information of the network bottom layer to the deep network through skip-connection.
Step 2 above also includes expanding the network field of view of the sample from video space using a convolution algorithm with stride.
The invention has the beneficial effects that:
1. the video non-uniform frame interpolation algorithm based on video motion estimation utilizes the spatial geometric relationship and the time domain motion relationship to perform video frame interpolation, is used for motion estimation and compensation prediction through deep learning, can be used for improving the video frame rate, and increases the time resolution.
2. The method of the invention carries out video non-uniform frame interpolation based on video motion estimation, removes redundant information in coding, reduces the frame rate in the video transmission process, and reduces the data volume transmitted by a video network, thereby being applicable to video compression or enhancing the video continuity.
3. The method carries out video non-uniform frame interpolation based on video motion estimation, thereby improving the video time resolution, improving the video smoothness and compensating the missing motion information.
4. The method of the invention extracts the video motion information through down sampling, thereby reducing the spatial dimension, and expands the network view field and the network receptive field from the video space by sampling through the convolution algorithm with stride, thereby enhancing the motion estimation capability of the network.
5. The method of the invention uses a sub-pixel algorithm to perform convolution characteristic up-sampling in the up-sampling process to recover the video space dimension and compensate the details.
6. The method uses skip-connection algorithm to transmit the bottom layer information of the convolutional neural network to the deep layer of the network for feature extraction and output of video interpolation frames, improves the feature extraction capability of the convolutional neural network, and can avoid gradient explosion during network optimization.
7. The method learns the motion information on the video time sequence through a Conv-LSTM algorithm, and is used for realizing video multi-frame interpolation based on video motion estimation based on the characteristic that the Conv-LSTM has the capability of outputting a plurality of convolution results in a time dimension.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a diagram of a neural network used in the method of the present invention.
Fig. 3 and 4 are diagrams of the motion estimation module of the present invention, wherein fig. 3 is a down-sampling process of the first half of the convolutional neural network, i.e., motion estimation, and fig. 4 is an up-sampling process of the second half of the convolutional neural network, i.e., motion estimation.
Fig. 5 and 6 are two associated real video frames 1 and 2 of an embodiment of the invention.
Fig. 7 to 9 show three insertion frames, i.e., an intermediate insertion frame 1, an intermediate insertion frame 2, and an intermediate insertion frame 3, generated in the embodiment of the present invention.
Detailed Description
Referring to fig. 1, the video frame interpolation method based on the convolutional neural network of the present invention includes the following steps:
1) selecting related front and back frames from an input video, normalizing the front and back frames, and inputting the normalized front and back frames into a convolutional neural network;
2) referring to fig. 2, a dense connection network is used to perform convolution feature extraction on the frames before and after the normalization processing; then, a network visual field is expanded from a video space by sampling through a convolution algorithm with stride, then the former half part of a convolution neural network carries out down-sampling processing on the front frame and the rear frame after normalization processing (see figure 3), and motion information between the front frame and the rear frame is extracted; performing convolution feature extraction on the front and rear frames after the down-sampling processing by using a dense connection network again, then performing up-sampling processing on motion information between the front and rear frames after the down-sampling processing by using a rear half coding module of a convolution neural network (see fig. 4), recovering the video space dimensionality and compensating details; meanwhile, the middle part of the convolutional neural network transmits the information of the network bottom layer to a deep network through skip-connection; the overall structure of the network is an encoding and decoding U-Net module, the encoding module is a 3-layer convolution module, the down-sampling is preferably Conv-LSTM algorithm, and the up-sampling is preferably sub-pixel algorithm;
3) inputting the information subjected to the upsampling processing in the step 2 into the bidirectional LSTM convolutional layer, and circularly outputting a plurality of intermediate video frames;
4) respectively calculating the intermediate video frame optical flow and the real video frame optical flow in the step 3 through the pre-trained FlowNet; then calculating the mean square error between the intermediate video frame optical flow and the real video frame optical flow;
5) and 4, taking the mean square error in the step 4 as an optimization objective function of the non-uniform video frame interpolation calculation, and enabling the FlowNet to participate in the gradient backward propagation process in network optimization to realize the optimization of the video frame interpolation.
Taking three frames of the intermediate interpolated frame as an example, inputting two frames connected in the video, namely a real video frame 1 (figure 5) and a real video frame 2 (figure 6), into a convolutional neural network, respectively extracting the convolution characteristics of the front frame and the rear frame after normalization processing through a dense connection network, and extracting motion information between the front frame and the rear frame after processing by a down-sampling motion estimation module for two times; the second half part of the input convolutional neural network coding module is restored to the space size consistent with the input video after being processed by the up-sampling motion estimation module twice, and intermediate frame insertion results, namely an intermediate frame 1 (figure 7), an intermediate frame 2 (figure 8) and an intermediate frame 3 (figure 9), are output through a Conv-LSTM algorithm. It can be seen that the laser knife is gradually moved to the left with respect to the real video frame 1 and gradually approaches the position of the laser knife in the real video frame 2, while for relatively stable tumor portions in the video frame, the video frame interpolation results remain stable as with the real video frame 1 and the real video frame 2.
Claims (7)
1. The video frame interpolation method based on the convolutional neural network is characterized by comprising the following steps of:
1) acquiring continuous frames of a video:
selecting related front and rear frames from the real video frames, normalizing the front and rear frames, and inputting the normalized front and rear frames into a convolutional neural network;
2) extracting video motion information and restoring video space:
the convolutional neural network front half part encoding module performs down-sampling processing on the front and rear frames after normalization processing, and extracts motion information between the front and rear frames; then, a second half decoding module of the convolutional neural network performs up-sampling processing on the motion information between the videos subjected to down-sampling processing, recovers the spatial dimension of the videos and compensates details;
meanwhile, the middle part of the convolutional neural network transmits the information of the network bottom layer to a deep network in a skip-connection mode, and extracts and outputs the video frame interpolation characteristics;
3) outputting a plurality of intermediate frames:
inputting the information subjected to the upsampling processing in the step 2 and the video frame interpolation characteristics into the bidirectional LSTM convolutional layer, and outputting at least one intermediate video frame in a circulating manner;
4) calculating the mean square error of the optical flow:
respectively calculating the intermediate video frame optical flow and the real video frame optical flow in the step 3 through the pre-trained FlowNet; then calculating the mean square error between the intermediate video frame optical flow and the real video frame optical flow;
5) optimizing the frame insertion:
and 4, taking the mean square error in the step 4 as an optimization objective function of the non-uniform video frame interpolation calculation, and enabling the FlowNet to participate in the gradient backward propagation process in network optimization to realize the optimization of the video frame interpolation.
2. The convolutional neural network-based video frame interpolation method of claim 1, wherein: in the step 2, the convolutional neural network front half part encoding module performs two times of downsampling processing on the front and rear frames after the normalization processing; and the second half decoding module of the convolutional neural network performs up-sampling processing twice on the motion information between videos subjected to down-sampling processing.
3. The convolutional neural network-based video frame interpolation method of claim 1, wherein: the convolutional neural network is an encoding and decoding U-Net module, and the encoding module is a 3-layer convolutional module.
4. The convolutional neural network-based video frame interpolation method as claimed in claim 1, 2 or 3, wherein: before upsampling and downsampling, a dense connection network is used for convolution feature extraction respectively.
5. The convolutional neural network-based video frame interpolation method of claim 4, wherein: the up-sampling adopts a sub-pixel algorithm; the down-sampling employs the Conv-LSTM algorithm.
6. The convolutional neural network-based video frame interpolation method of claim 5, wherein: in step 2, the middle part of the convolutional neural network transmits the information of the network bottom layer to the deep network through skip-connection.
7. The convolutional neural network-based video frame interpolation method of claim 6, wherein: step 2 also includes expanding the network field of view of the sample from video space using a convolution algorithm with stride.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011364443.0A CN112533026A (en) | 2020-11-27 | 2020-11-27 | Video frame interpolation method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011364443.0A CN112533026A (en) | 2020-11-27 | 2020-11-27 | Video frame interpolation method based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112533026A true CN112533026A (en) | 2021-03-19 |
Family
ID=74994697
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011364443.0A Pending CN112533026A (en) | 2020-11-27 | 2020-11-27 | Video frame interpolation method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112533026A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113205148A (en) * | 2021-05-20 | 2021-08-03 | 山东财经大学 | Medical image frame interpolation method and terminal for iterative interlayer information fusion |
| CN114066761A (en) * | 2021-11-22 | 2022-02-18 | 青岛根尖智能科技有限公司 | Motion video frame rate enhancement method and system based on optical flow estimation and foreground detection |
| CN114418882A (en) * | 2022-01-17 | 2022-04-29 | 京东方科技集团股份有限公司 | Processing method, training method, device, electronic equipment and medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109068174A (en) * | 2018-09-12 | 2018-12-21 | 上海交通大学 | Video frame rate upconversion method and system based on cyclic convolution neural network |
| CN109379550A (en) * | 2018-09-12 | 2019-02-22 | 上海交通大学 | Video frame rate up-conversion method and system based on convolutional neural network |
| WO2019040134A1 (en) * | 2017-08-22 | 2019-02-28 | Google Llc | Optical flow estimation for motion compensated prediction in video coding |
| US20190289321A1 (en) * | 2016-11-14 | 2019-09-19 | Google Llc | Video Frame Synthesis with Deep Learning |
| CN110351511A (en) * | 2019-06-28 | 2019-10-18 | 上海交通大学 | Video frame rate upconversion system and method based on scene depth estimation |
| CN111105382A (en) * | 2019-12-31 | 2020-05-05 | 北京大学 | Video Repair Method |
| CN111405316A (en) * | 2020-03-12 | 2020-07-10 | 北京奇艺世纪科技有限公司 | Frame insertion method, electronic device and readable storage medium |
| WO2020150264A1 (en) * | 2019-01-15 | 2020-07-23 | Portland State University | Feature pyramid warping for video frame interpolation |
-
2020
- 2020-11-27 CN CN202011364443.0A patent/CN112533026A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190289321A1 (en) * | 2016-11-14 | 2019-09-19 | Google Llc | Video Frame Synthesis with Deep Learning |
| WO2019040134A1 (en) * | 2017-08-22 | 2019-02-28 | Google Llc | Optical flow estimation for motion compensated prediction in video coding |
| CN109068174A (en) * | 2018-09-12 | 2018-12-21 | 上海交通大学 | Video frame rate upconversion method and system based on cyclic convolution neural network |
| CN109379550A (en) * | 2018-09-12 | 2019-02-22 | 上海交通大学 | Video frame rate up-conversion method and system based on convolutional neural network |
| WO2020150264A1 (en) * | 2019-01-15 | 2020-07-23 | Portland State University | Feature pyramid warping for video frame interpolation |
| CN110351511A (en) * | 2019-06-28 | 2019-10-18 | 上海交通大学 | Video frame rate upconversion system and method based on scene depth estimation |
| CN111105382A (en) * | 2019-12-31 | 2020-05-05 | 北京大学 | Video Repair Method |
| CN111405316A (en) * | 2020-03-12 | 2020-07-10 | 北京奇艺世纪科技有限公司 | Frame insertion method, electronic device and readable storage medium |
Non-Patent Citations (2)
| Title |
|---|
| REN-YU TSENG等: "Adaptive Frame Interpolation using an End-to-End Deep Net with High Quality Flow Estimation", 《2019 INTERNATIONAL CONFERENCE ON TECHNOLOGIES AND APPLICATIONS OF ARTIFICIAL INTELLIGENCE 》 * |
| 张智峰: "基于深度学习的视频插帧技术研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113205148A (en) * | 2021-05-20 | 2021-08-03 | 山东财经大学 | Medical image frame interpolation method and terminal for iterative interlayer information fusion |
| CN114066761A (en) * | 2021-11-22 | 2022-02-18 | 青岛根尖智能科技有限公司 | Motion video frame rate enhancement method and system based on optical flow estimation and foreground detection |
| CN114418882A (en) * | 2022-01-17 | 2022-04-29 | 京东方科技集团股份有限公司 | Processing method, training method, device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Video super-resolution based on deep learning: a comprehensive survey | |
| CN112218072B (en) | Video coding method based on deconstruction compression and fusion | |
| CN117730338A (en) | Video super-resolution network and video super-resolution, encoding and decoding processing method and device | |
| CN112348766A (en) | Progressive feature stream depth fusion network for surveillance video enhancement | |
| CN111028150A (en) | A fast spatiotemporal residual attention video super-resolution reconstruction method | |
| CN112070665A (en) | Generative Adversarial Video Super-Resolution Reconstruction and Reconstruction Image Authenticity Discrimination Method | |
| CN112533026A (en) | Video frame interpolation method based on convolutional neural network | |
| CN115689917A (en) | Efficient space-time super-resolution video compression restoration method based on deep learning | |
| CN119313583A (en) | An image denoising method based on dual-branch feature fusion | |
| CN119205515A (en) | Image super-resolution generation model construction method and image super-resolution generation method and system | |
| CN120223924A (en) | Panoramic video frame insertion method based on latent diffusion model | |
| CN113850718B (en) | A video synchronization spatiotemporal super-resolution method based on inter-frame feature alignment | |
| CN120075449B (en) | A context-based binocular video compression method | |
| CN114663306B (en) | Video bit depth enhancement method and device based on pyramid multi-level information fusion | |
| CN117593185A (en) | A video super-resolution method and system based on deformable Transformer | |
| CN117640946A (en) | Systems and methods for transmitting and receiving image frames | |
| CN120088133B (en) | An Arbitrary Scale Video Super-Resolution Method Based on a Diffusion Model | |
| Gong et al. | Temporal transformer-based video super-resolution reconstruction with cross-modal attention | |
| CN119850422B (en) | A method and system for super-resolution reconstruction of synthetic aperture radar images based on Mamba's optical guidance | |
| CN120751217B (en) | A video processing method, apparatus, storage medium, and electronic device | |
| CN116033169B (en) | A video compression method and system considering long-distance temporal information | |
| CN113506219B (en) | Video super-resolution model training method and device | |
| CN120374457B (en) | Complex dramatic scene light field image enhancement and angle reconstruction method | |
| Wang et al. | Progressive local-to-global vision transformer for occluded face hallucination | |
| CN119850422A (en) | Mamba-based optical-guided synthetic aperture radar image super-resolution reconstruction method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210319 |
|
| RJ01 | Rejection of invention patent application after publication |