Home

Awesome

COMPASS: High-Efficiency Deep Image Compression with Arbitrary-scale Spatial Scalability

ICCV 2023

Paper | Project Page | Video

Installation

COMPASS supports python 3.7+ and Pytorch 1.13+.

git clone https://github.com/ImJongminPark/COMPASS.git
cd COMPASS
conda create -n compass python=3.7 -y
pip install torch torchvision

Requirements

Datasets

You can download the training and test datasets via this link.

mkdir datsets_img
mv <YOUR_DOWNLOAD_PATH>/train_512.zip datasets_img
mv <YOUR_DOWNLOAD_PATH>/test.zip datasets_img

cd datasets_img
unzip train_512.zip -d train_512
unzip test.zip -d test

Training

Before the training process, download the pre-trained residual compression module and LIFF module via this link.

mkdir pretrained
mv <YOUR_DOWNLOAD_PATH>/pretrained.zip pretrained
cd pretrained
unzip pretrained.zip

For the training process, choose a lambda value from the set [0.0018, 0.0035, 0.0067, 0.013]. Then, assign this selected value to the 'lmbda' parameter within the 'cfg_train.yaml' configuration file. Ensure this lambda value is consistent with the pre-trained residual compression module you intend to use.

python -m torch.distributed.launch --nproc_per_node=<NUM_OF_GPUS> train.py

Evaluation

Before the evaluation process, download the whole pre-trained COMPASS model via this link.

mkdir checkpoints
mv <YOUR_DOWNLOAD_PATH>/checkpoints.zip checkpoints
cd checkpoints
unzip checkpoints.zip

For the evaluation process, choose a lambda value from the set [0.0018, 0.0035, 0.0067, 0.013]. Then, assign this selected value to the 'lmbda' parameter within the 'cfg_eval.yaml' configuration file.

python update.py
python eval.py

Acknowledgements

This work was supported by internal fund/grant of Electronics and Telecommunications Research Institute (ETRI). [23YC1100, Technology Development for Strengthening Competitiveness in Standard IPR for communication and media]

Authors

Citation

If you use this project, please cite the relevant original publications for the models and datasets, and cite this project as:

@inproceedings{park2023compass,
  title={COMPASS: High-Efficiency Deep Image Compression with Arbitrary-scale Spatial Scalability},
  author={Park, Jongmin and Lee, Jooyoung and Kim, Munchurl},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={12826--12835},
  year={2023}
}