Pytorch implementation of our ACCV 2018 paper "Learning for Video Super-Resolution through HR Optical Flow Estimation" and TIP 2020 paper "Deep Video Super-Resolution using HR Optical Flow Estimation".
Figure 1. Overview of our SOF-VSR network.
Figure 2. Comparison with the state-of-the-arts.
- Python 3
- pytorch (0.4), torchvision (0.2)
- numpy, PIL
- Matlab (For PSNR/SSIM evaluation)
We collect 145 1080P video clips from the CDVL Database for training.
We use the Vid4 dataset and a subset of the DAVIS dataset (namely, DAVIS-10) for benchmark test.
- Vid4(BaiduPan, GoogleDrive)
- DAVIS-10
We use 10 scenes in the DAVIS-2017 test set including boxing, demolition, dive-in, dog-control, dolphins, kart-turn, ocean-birds, pole-vault, speed-skating and wings-trun.
Figure 3. Comparative results achieved on the Vid4 dataset. Zoom-in regions from left to right: IDNnet, VSRnet, TDVSR, our SOF-VSR, DRVSR and our SOF-VSR-BD.
Figure 4. Comparative results achieved on the DAVIS-10 dataset. Zoom-in regions from left to right: IDNnet, VSRnet, our SOF-VSR, DRVSR and our SOF-VSR-BD.
Figure 5. Visual comparison of 4x SR results. From left to right: VSRnet, TDVSR, our SOF-VSR and the groundtruth.
@InProceedings{Wang2018accv,
author = {Longguang Wang and Yulan Guo and Zaiping Lin and Xinpu Deng and Wei An},
title = {Learning for Video Super-Resolution through {HR} Optical Flow Estimation},
booktitle = {ACCV},
year = {2018},
}
@Article{Wang2020tip,
author = {Longguang Wang and Yulan Guo and Li Liu and Zaiping Lin and Xinpu Deng and Wei An},
title = {Deep Video Super-Resolution using {HR} Optical Flow Estimation},
journal = {{IEEE} Transactions on Image Processing},
year = {2020},
}
For questions, please send an email to wanglongguang15@nudt.edu.cn