• Coarse2Fine
• • Our Contibutions
  • File Structure
  • Environment
  • test
  • Comparison
  • • Reconstructed image
    • Quality measurements on Kodak of MindSpore version
  • Citation
  • Contributors

Coarse2Fine

Coarse-to-Fine Hyper-Prior Modeling for Learned Image Compression

key words: Image Compression, Hyper-Prior Model

This paper proposes a coarse-to-fine framework with hierarchical layers of hyper-priors to conduct comprehensive analysis of the image and more effectively reduce spatial redundancy, which improves the rate-distortion performance of image compression significantly. The paper is published in 2020, and readers can read the original paper via the link.

Our Contibutions

1. Translate from pytorch to mindspore
2. Test the mindspore version

File Structure

Coarse2Fine/mindspore
├── __pycache__
├── akg_kernel_meta      # akg kernel
├── example.bin          # coded binary file
├── example.png          # original image
├── module_arithmeticcoding.cpp #arithmetic coder
├── model_baseline.py    # model baseline
├── my_model.ckpt        # model params
├── reconstruct.png      # decoded image
└── test.py              # test

Environment

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

test

cd Coarse2Fine/mindspore
python test.py

Comparison

Reconstructed image

mindspore version

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

Quality measurements on Kodak of MindSpore version

bpp	PSNR	MSSSIM	enc_time	dec_time	GPU Memory(MiB)	lambda
0.208	30.063	0.946	31.16	36.773	5036	qp1
0.309	31.611	0.963	28.08	39.709	5036	qp2
1.06	38.025	0.991	51.601	61.22	6060	qp5
1.439	40.049	0.994	52.225	64.669	6060	qp7

MindSpore models are converted from officially pretrained PyTorch models.

Citation

@inproceedings{hu2020coarse,
    title={Coarse-to-Fine Hyper-Prior Modeling for Learned Image Compression},
    author={Hu, Yueyu and Yang, Wenhan and Liu, Jiaying},
    booktitle={AAAI Conference on Artificial Intelligenc},
    year={2020}
}

Contributors

Name:

Chenhao Zhang