Video Rescaling Networks with Joint Optimization Strategies for Downscaling and Upscaling (CVPR 2021)

Pytorch Implementation of the paper "Video Rescaling Networks with Joint Optimization Strategies for Downscaling and Upscaling (CVPR 2021)".

Project Page: Link

Paper (arXiv): Link

Prerequisite

• Python 3 via Anaconda (recommended)
• PyTorch >= 1.4.0
• NVIDIA GPU + CUDA
• Python Package: pip install numpy opencv-python lmdb pyyaml

Dataset Preparation

Training and testing dataset can be found here. We adopt the LMDB format and also provide the script in codes/data_scripts. For more detail, please refer to BasicSR.

Usage

Pretrained weight can be downloaded from Google Drive.

All the implementation is in /codes. To run the code, select the corresponding configuration file in /codes/options/ and run as following command (MIMO-VRN for example):

Training

python train.py -opt options/train/train_MIMO-VRN.yml

Testing

python test.py -opt options/test/test_MIMO-VRN.yml

Quantitative Results

HR Reconstruction on Vid4

Qualitative Results

HR Reconstruction on Vid4

LR Reconstruction on Vid4

Citation

@InProceedings{Huang_2021_CVPR,
    author    = {Huang, Yan-Cheng and Chen, Yi-Hsin and Lu, Cheng-You and Wang, Hui-Po and Peng, Wen-Hsiao and Huang, Ching-Chun},
    title     = {Video Rescaling Networks With Joint Optimization Strategies for Downscaling and Upscaling},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {3527-3536}
}

Acknowledgement

Our project is heavily based on Invertible-Image-Rescaling and they adopt BasicSR as basic framework.