Introduction

This is the implementation of paper "Optical-Flow-Reuse-Based Bidirectional Recurrence Network for Space-Time Video Super-Resolution".

Pre-trained models

BaiduCloud

password: opd3

dropbox

Environment

We are good in the environment:

python 3.7

CUDA 9.1

Pytorch 1.5.0

Generate image

To test on vimeo,

cd src

python test_vimeo.py --datapath VIMEOPATH --outputpath  OUTPUTPATH --weight PATHTOWEIGHT

To test on REDS,

cd src

python test_reds.py --datapath REDSPATH --outputpath  OUTPUTPATH --weight PATHTOWEIGHT

To test on VID4,

cd src

python test_vid4.py --datapath VID4PATH --outputpath  OUTPUTPATH --weight PATHTOWEIGHT

Calculate criteria

you should specify the GT path and output path first, and run:

cd src

python eval.py

or you may directly get all evaluation results in src/evaluation_results

Run a demo

cd src

python demo.py

Other STVSR works

We have conducted a series of video spatiotemporal super-resolution-related works, which include not only OFR-BRN but also:

• Yuantong Zhang, Huairui Wang, Zhenzhong Chen: Controllable Space-Time Video Super-Resolution via Enhanced Bidirectional Flow Warping. VCIP 2022
• Yuantong Zhang, Daiqin Yang, Zhenzhong Chen, Wenpeng Ding: Continuous Space-Time Video Super-Resolution with Multi-stage Motion Information Reorganization. ACM Transactions on Multimedia Computing Communications and Applications.
• Yuantong Zhang, Hanyou Zheng, Daiqin Yang, Zhenzhong Chen, Haichuan Ma, Wenpeng Ding: Space-Time Video Super-resolution with Neural Operator. CoRR abs/2404.06036 (2024)

Acknowledgment

Our code is built on

Zooming-Slow-Mo-CVPR-2020

open-mmlab

bicubic_pytorch

FLAVR

RIFE

We thank the authors for sharing their codes!