This is the official implementation and appendix of the paper:
Structure-Preserving Motion Estimation for Learned Video Compression. Han Gao, Jinzhong Cui, Mao Ye, Shuai Li, Yu Zhao, Xiatian Zhu. ACM Multimedia 2022. [pdf]
Change the configs in class named HEVC_dataset
of the file dataset.py to the path of the data to be tested, e.g. :
root="/xxx/HEVC_dataset/Class_B", filelist="./Tools/filelists/B.txt"
Run test.py for testing, in which the config named --model_path
is the pretrained model path, and --lambda_weight
is the lambda value of the prerained model, e.g. :
python -u test.py --model_path="./Checkpoints/2048.pth" --lambda_weight=2048
During implementation, we drawed on the experience of CompressAI, PyTorchVideoCompression and DCVC. The model weights of intra coding are from CompressAI.
If you find this paper useful, kindly cite:
@inproceedings{gao2022structure,
title={Structure-Preserving Motion Estimation for Learned Video Compression},
author={Gao, Han and Cui, Jinzhong and Ye, Mao and Li, Shuai and Zhao, Yu and Zhu, Xiatian},
booktitle={Proceedings of the 30th ACM International Conference on Multimedia (MM’22)},
year={2022}
}
If any questions, kindly contact with Han Gao via e-mail: han.gao@std.uestc.edu.cn / gaohan_vc@163.com.