Introduction

IPCodec is an open source entropy-based compression toolbox with tensorflow 1.12.0, which contains Interpolation Variable Rate (IVR) I-frame and P-frame compression models.

---

License

This project is developed by Alibaba and licensed under the Apache 2.0 license.

---

Installation

Prerequisites

• Linux
• Python 3.6+
• tensorflow 1.12.0
• CUDA 10.0+

Prepare environment

1. Create a conda virtual environment and activate it.

   conda create -n tf12 python=3.6 -y
   conda activate tf12

2. Install other packages with the following command.

   pip install -r requirements.txt

---

Easy to use

Use the checkpoint/download.sh to download all pretrained models.

cd checkpoint
sh download.sh

Train or test for image compression

• If not use Unet post network, set is_post to False.
• Download Kodak test image to dataset/kodak

python I_train.py/I_test.py --loss_metric PSNR --model_name CM \
--is_post True  --with_context_model True  --is_multi True

python I_train.py/I_test.py --loss_metric SSIM --model_name CM \
--is_post True  --with_context_model True  --is_multi True

python I_train.py/I_test.py --loss_metric PSNR --model_name NoCM \
--is_post False  --with_context_model False  --is_multi True

Train or test for video compression

• If not use Unet post network, set is_post to False.
• Convert video to PNG images to test_set_dir in IP_config.py.
• IP_test_multi.py is the Multi videos processing script, need to carefully use it since the model_path is changed.

python IP_train.py/IP_test.py --loss_metric PSNR --model_name STPM \
--is_post True  --with_context_model True  --is_multi True --ckpt_dir_pre ./checkpoint/I_model/CM_PSNR  --idx_test 0

python IP_train.py/IP_test.py --loss_metric SSIM --model_name STPM \
--is_post True  --with_context_model True  --is_multi True --ckpt_dir_pre ./checkpoint/I_model/CM_SSIM  --idx_test 0

python IP_train.py/IP_test.py --loss_metric PSNR --model_name NoSPM \
--is_post True  --with_context_model False  --is_multi True --ckpt_dir_pre ./checkpoint/I_model/CM_PSNR  --idx_test 0

python IP_train.py/IP_test.py --loss_metric SSIM --model_name NoSPM \
--is_post True  --with_context_model False  --is_multi True --ckpt_dir_pre ./checkpoint/I_model/CM_SSIM  --idx_test 0

---

Results

Results for I-frame compression (Image compression) on Kodak Dataset.

Results for P-frame compression (Video compression) on MCL-JCV, UVG, HEVC-ClassB Dataset.

Note: If you find this useful, please support us by citing the following paper.

@inproceedings{ivr,
  title={Interpolation variable rate image compression},
  author={Sun, Zhenhong and Tan, Zhiyu and Sun, Xiuyu and Zhang, Fangyi and Qian, Yichen and Li, Dongyang and Li, Hao},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
  pages={5574--5582},
  year={2021}
}

@article{sun2021spatiotemporal,
  title={Spatiotemporal Entropy Model is All You Need for Learned Video Compression},
  author={Sun, Zhenhong and Tan, Zhiyu and Sun, Xiuyu and Zhang, Fangyi and Li, Dongyang and Qian, Yichen and Li, Hao},
  journal={arXiv preprint arXiv:2104.06083},
  year={2021}
}

---

Main Contributors

Zhenhong Sun,Dongyang Li, Xiuyu Sun.