• Checkerboard-Context-Model
• • Our Contibutions
  • File Structure
  • Environment
  • Command
  • Train
  • test
  • Comparison
  • • Reconstructed image
    • Quality measurements on Kodak
    • • MindSpore version, models trained on div2k
  • Citation
  • Contributors

Checkerboard-Context-Model

Checkerboard Context Model for Efficient Learned Image Compression

key words: Lossless Image Compression, Checkerboard Context Model

This paper proposes a parallel context model based on the checkerboard-shaped convolution and develops a two-pass parallel decoding scheme. Compared with the serial context model, it allows decoding to be implemented in a highly parallel manner. The paper is published in 2021, and readers can read the original paper via the link.

Our Contibutions

1. Translate from pytorch to mindspore
2. Successfully run forward, backward and parameter update
3. Test the mindspore version

File Structure

source
├── __pycache__
├── akg_kernel_meta      # akg kernel
├── decoded files        # decoded images
├── model          
│   └── best.ckpt        # model file with best performance
│   └── train_log.txt    # train log
├── layer.py             # CheckerboardContext layer
├── model_baseline.py    # model baseline
├── test.py              # test in reality
└── train.py             # train

Environment

• mindspore-dev==2.0.0dev20230116
  • It is recommended to install through the link.
• cuda 11.1
• python 3.7

Command

Please intall compressai via

pip install -e .

under dictionary CompressAI_MindSpore.

Train

Go to directory source.
Train the model via:

python train.py -d "your/own/dataset/address" --seed 0  --batch-size 16 --test-batch-size 1 --save --lambda 0.01 -e 10

test

python test.py -d "your/own/dataset/address" --seed 0   --test-batch-size 1 --lambda 0.01 --pretrained_file "model/best.ckpt"

Comparison

Reconstructed image

mindspore version

Comparison of the original images with reconsructed ones (original above, reconstructed below)

Quality measurements on Kodak

MindSpore version, models trained on div2k

bpp	PSNR	MSSSIM	run_time	GPU Memory(MiB)	lambda
0.138	22.91	0.852	27.009	4002	lambda0.01
0.263	23.213	0.863	27.085	4002	lambda0.03
0.468	25.838	0.925	32.072	4002	lambda0.06
0.686	27.468	0.953	33.721	4002	lambda0.22
0.953	27.621	0.957	23.823	4002	lambda0.4

Citation

@misc{he2021checkerboard,
      title={Checkerboard Context Model for Efficient Learned Image Compression}, 
      author={Dailan He and Yaoyan Zheng and Baocheng Sun and Yan Wang and Hongwei Qin},
      year={2021},
      eprint={2103.15306},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}

Contributors

Name:

Chenhao Zhang