[ICLR2021] Learning Accurate Entropy Model with Global Reference for Image Compression [pdf]
The official repository for Learning Accurate Entropy Model with Global Reference for Image Compression.
Evaluation on Kodak Dataset
Clone the repo and create a conda environment as follows:
conda create --name ref python=3.6
conda activate ref
conda install pytorch=1.1 torchvision cudatoolkit=10.0
(We use PyTorch 1.1, CUDA 10.1.)
Kodak Dataset
kodak
├── image1.jpg
├── image2.jpg
└── ...
Evaluation:
# Kodak
sh test.sh [/path/to/kodak] [model_path]
Compress:
sh compress.sh original.png [model_path]
Decompress:
sh decompress.sh original.bin [model_path]
Download the pre-trained models optimized by MSE.
Note: We reorganize code and the performances are slightly different from the paper's.
Codebase from L3C-image-compression , torchac
If you find this code useful for your research, please cite our paper
@InProceedings{Yichen_2021_ICLR,
author = {Qian, Yichen and Tan, Zhiyu and Sun, Xiuyu and Lin, Ming and Li, Dongyang and Sun, Zhenhong and Li, Hao and Jin, Rong},
title = {Learning Accurate Entropy Model with Global Reference for Image Compression},
booktitle = {International Conference on Learning Representations},
month = {May},
year = {2021},
}