• iwave
• • our contibutions
  • file structure
  • environment
  • command
  • test
  • • • use test_ms.py to test mindspore model
      • use test_torch.py to test pytorch model
  • comparision
  • • • quality measurements
  • source
  • citation
  • contributors

iwave

《End-to-End Optimized Versatile Image Compression With Wavelet-Like Transform》

key words:image compression, wavelet transform, end-to-end optimization

iwave proposes a new end-to-end optimized image compression scheme, in which iwave, a trained wavelet-like transform, converts images into coefficients without any information loss. Then the coefficients are optionally quantized and encoded into bits. It achieves state-of-the-art compression efficiency compared with deep network-based methods.

our contibutions

1.translate from pytorch to mindspore

2.successfully run forward, backward and parameter update

3.test and compare between the mindspore version and the pytorch version with pretrained parameter

file structure

the translated model file is iwave/source/train/Model/Model.py

the pretrain paremeter is iwave/source/train/Model/iwave_mse.pth for pytorch and iwave/source/train/Model/iwave_mse.ckpt for mindspore

environment

• mindspore-dev==2.0.0dev20230116
• ubuntu 16.04
• cuda 11.1
• python 3.7

command

to use the translated model, you can choose more models in test.sh

python test_ms.py --mse_path 0.1_mse --init_scale 14 --mode w  --header 
python test_ms.py --mse_path 0.024_mse --init_scale 34 --mode a 
python test_ms.py --mse_path 0.0095_mse --init_scale 28 --mode a 
python test_ms.py --mse_path 0.0037_mse --init_scale 58 --mode a 

test

we choose div2k as the test dataset in pytorch and mindspore

use test_ms.py to test mindspore model

1.cd iwave/source/train

2.bash test.sh

use test_torch.py to test pytorch model

1.cd iwave_torch/source/train

2.bash test.sh

comparision

quality measurements

The results of different bitrate models on div2k test dataset:

mse	bpp	PSNR	MSSSIM	time_cost	GPU M
0.1_mse	1.725	38.223	0.990	63.697	5028
0.024_mse	0.660	33.568	0.974	65.273	5028
0.0095_mse	0.952	34.196	0.976	65.261	5028
0.0037_mse	0.411	30.767	0.950	65.783	5028

source

translated from pytorch

link:iWave / iWave · GitLab

citation

@article{ma2020end,
  title={End-to-end optimized versatile image compression with wavelet-like transform},
  author={Ma, Haichuan and Liu, Dong and Yan, Ning and Li, Houqiang and Wu, Feng},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  volume={44},
  number={3},
  pages={1247--1263},
  year={2020},
  publisher={IEEE}
}

contributors

name: Kaiyu Zheng, Yongchi Zhang