This repository is the our solution for VCIP2022. The model is modified based on convNeXt.
PyTorch = 1.8
python = 3.8
The docker environment is recommended in Docker Hub.
pip install requirements
# train
python train.py --config=./config.yaml --name=nic_cvt # --resume '/workspace/lm/TIC/results/nic_cvt/8/snapshots/best.pt'
# eval
python eval.py --snapshot='./results/tic_/8/snapshots/best.pt' --quality=8
# compress
python codec_VR.py encode -q 6 -o I01.bin -lb 1000 -ph 1472 -pw 1280 /data/Dongmuchen/VCIP_mytestdata/mytest/1.png
/workspace/sharedata/VCIP2022/val/1.png
# decompress
python codec_VR.py decode -o I01dec.png I01.bin
The framework is based on CompressAI, we thank the authors for their great work.