Deformable 3D Convolution for Video Super-Resolution

Pytorch implementation of deformable 3D convolution network (D3Dnet). [PDF]

Our code is based on cuda and can perform deformation in any dimension of 3D convolution.

Overview

Architecture of D3Dnet

Architecture of D3D

Requirements

• Python 3
• pytorch (1.0.0), torchvision (0.2.2) / pytorch (1.2.0), torchvision (0.4.0)
• numpy, PIL
• Visual Studio 2015

Build

Compile deformable 3D convolution:

1. Cd to code/dcn.
2. For Windows users, run cmd make.bat. For Linux users, run bash make.sh. The scripts will build D3D automatically and create some folders.
3. We offer customized settings for any dimension (e.g., Temporal, Height, Width) you want to deform. See code/dcn/test.py for more details.

Datasets

Training dataset

1. Download the Vimeo dataset and put the images in code/data/Vimeo.
2. Cd to code/data/Vimeo and run generate_LR_Vimeo90K.m to generate training data as below:

  Vimeo
    └── sequences
           ├── 00001
           ├── 00002
           ├── ...
    └── LR_x4
           ├── 00001
           ├── 00002
           ├── ...		
    ├── readme.txt 
    ├── sep_trainlist.txt
    ├── sep_testlist.txt
    └── generate_LR_Vimeo90K.m      

Test dataset

1. Download the dataset Vid4 and SPMC-11 dataset in https://pan.baidu.com/s/1PKZeTo8HVklHU5Pe26qUtw (Code: 4l5r) and put the folder in code/data.
2. (optional) You can also download Vid4 and SPMC-11 or other video datasets and prepare test data in code/data as below:

 data
  └── dataset_1
         └── scene_1
               └── hr    
                  ├── hr_01.png  
                  ├── hr_02.png  
                  ├── ...
                  └── hr_M.png    
               └── lr_x4
                  ├── lr_01.png  
                  ├── lr_02.png  
                  ├── ...
                  └── lr_M.png   
         ├── ...		  
         └── scene_M
  ├── ...    
  └── dataset_N      

Results

Quantitative Results

We have organized the Matlab code framework of Video Quality Assessment metric T-MOVIE and MOVIE. [Code]
Welcome to have a look and use our code.

Qualitative Results

A demo video is available at https://wyqdatabase.s3-us-west-1.amazonaws.com/D3Dnet.mp4

Citiation

@article{D3Dnet,
  author = {Ying, Xinyi and Wang, Longguang and Wang, Yingqian and Sheng, Weidong and An, Wei and Guo, Yulan},
  title = {Deformable 3D Convolution for Video Super-Resolution},
  journal = {IEEE Signal Processing Letters},
  volume = {27},
  pages = {1500-1504}，
  year = {2020},
}

Acknowledgement

This code is built on [DCNv2] and [SOF-VSR]. We thank the authors for sharing their codes.

Contact

Please contact us at [email protected] for any question.