Our paper was accepted to CVPR2023. You can also find our full-version paper on arXiv

🚀Quickstart

1. Setup a conda environment and install the pytorch

conda create -n TINC python=3.9
conda activate TINC
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia

2. Installing python libraries

pip install -r requirements.txt

3. Compression

(1) Single compression task

relevant compression parameters can be modified in opt/SingleTask/default.yaml.

python main.py -p opt/SingleTask/default.yaml -g 0 

final compressed file path: outputs/default_{time}/compressed/

final decompressed file path: outputs/default_{time}/decompressed.tif

training result:

tensorboard --logdir=outputs/default_{time}

(2) Run multiple tasks at once

relevant compression parameters can be modified in opt/MultiTask/default.yaml.

python MultiTask.py -p opt/MultiTask/default.yaml -g 0,1,2,3 -stp main.py -debug

😘Citations

@inproceedings{yang2023tinc,
  title={TINC: Tree-structured Implicit Neural Compression},
  author={Yang, Runzhao},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={18517--18526},
  year={2023}
}

💡Contact

If you need any help or are looking for cooperation feel free to contact us. yangrz20@mails.tsinghua.edu.cn