Coarse-to-Fine Spatio-Temporal Information Fusion for Compressed Video Quality Enhancement (SPL 2022)
The PyTorch implementation for the CF-STIF: Coarse-to-Fine Spatio-Temporal Information Fusion for Compressed Video Quality Enhancement which is accepted by [IEEE SPL 2022].
conda create -n CF-STIF python=3.10.11
conda activate CF-STIF
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.3 -c pytorch
git clone --depth=1 https://github.com/xiaomingxige/CF-STIF
cd CF-STIF
pip install -r requirements.txt
cd ops/dcn/
# You may need to modify the paths of cuda before compiling.
bash build.sh
python simple_check.py
The DCNv2 source files here is different from the open-sourced version due to incompatibility. [issue]
Please check here.
You should modify option_mfqev2_1G.yml
according to your environment
1G
: 1 GPUs will be used for the below training.
You can also edit
option_mfqev2_2G.yml
andoption_mfqev2_4G.yml
if needed.
We now generate LMDB to speed up IO during training.
python create_lmdb_mfqev2.py --opt_path option_mfqev2_1G.yml
Finally, the MFQEv2 dataset root will be sym-linked to the folder ./data/
automatically.
So that we and programmes can access MFQEv2 dataset at
./data/
directly.
See script.sh
.
We utilize 1 NVIDIA GeForce RTX 3090 GPU for testing:
python test_video.py
If you find this project is useful for your research, please cite:
@article{luo2022coarse,
title={Coarse-to-fine spatio-temporal information fusion for compressed video quality enhancement},
author={Luo, Dengyan and Ye, Mao and Li, Shuai and Li, Xue},
journal={IEEE Signal Processing Letters},
volume={29},
pages={543--547},
year={2022},
publisher={IEEE}
}
This work is based on STDF-Pytoch. Thank RyanXingQL for sharing the codes.