Pytorch Implementation of "ShiftLIC: High-Efficiency Learned Image Compression with Spatial-Channel Shift Operations"

This repository contains the Pytorch implementation of the paper "ShiftLIC: High-Efficiency Learned Image Compression with Spatial-Channel Shift Operations" in IEEE Transactions on Circuits and Systems for Video Technology.

TODO

• Provide pretrained models
• Offer supplementary materials

Architectures

Figure 1: Overview of the ShiftLIC Architecture

Results

Figure 2: Comprehensive Performance Comparison

Figure 3: Rate-Distortion-Complexity-Latency Trade-off

Dependencies

To run this code, you will need the following dependencies:

• Pytorch 1.9
• CompressAI

You can install the necessary dependencies using the following command:

pip install torch==2.0.1 compressai=1.2.4

Pretrained Models

Pretrained models will be made available soon. Stay tuned for updates.

Training

To train the models, follow these steps:

1. Clone the repository:

   git clone https://github.com/baoyu2020/ShiftLIC.git
   cd ShiftLIC/shiftlic/large/

2. Prepare your dataset and adjust the configuration in config.py.

3. Run the training script:

   python train.py -q 1 --out_dir ./Log/out_dir/ --batch_size 16 --metric mse --lr 1e-4 --epochs 100 --dataset Train_dataset_dir 

Testing

To test the models, follow these steps:

1. Run the testing script:

   python test.py --model_path ./Log/out_dir/model_best.pth --dataset Test_dataset_dir

2. The results will be saved in the ./Log/out_dir/ directory.

Acknowledgements

This repository is based on CompressAI and TinylLIC.

If you find this work useful for your research, please cite:

@article{10947057,
  author={Bao, Youneng and Tan, Wen and Jia, Chuanmin and Li, Mu and Liang, Yongsheng and Tian, Yonghong},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={ShiftLIC: Lightweight Learned Image Compression with Spatial-Channel Shift Operations}, 
  year={2025},
  doi={10.1109/TCSVT.2025.3556708}}

@article{lu2022high,
  title={High-Efficiency Lossy Image Coding Through Adaptive Neighborhood Information Aggregation},
  author={Lu, Ming and Ma, Zhan},
  journal={arXiv preprint arXiv:2204.11448},
  year={2022}
}

Contact

For any inquiries or issues, please contact us at ynbao@stu.hit.cn.

---

We hope you find this repository useful. Contributions are welcome!