Skip to content

Image downscaling & super-resolution project based on "Once for All: Train One Network and Specialize it for Efficient Deployment" (ICLR 2020)

License

Notifications You must be signed in to change notification settings

timfu248/ofa-for-super-resolution

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Image Downscaling & Super-Resolution based on Once-for-All

This repository contains image downscaling & super-resolution project code based on the paper "Once-for-All: Train One Network and Specialize it for Efficient Deployment" (ICLR 2020).

The objectives of this proejct are

  • Find the best image downscaling & super-resolution neural network architecture on mobile devices
  • Support both 2x, 4x super-resolution in a single architecture.

License and Citation

@inproceedings{
  cai2020once,
  title={Once for All: Train One Network and Specialize it for Efficient Deployment},
  author={Han Cai and Chuang Gan and Tianzhe Wang and Zhekai Zhang and Song Han},
  booktitle={International Conference on Learning Representations},
  year={2020},
  url={https://arxiv.org/pdf/1908.09791.pdf}
}
@inproceedings{
  kim2018tar,
  title={Task-Aware Image Downscaling},
  author={Heewon Kim and Myungsub Choi and Bee Lim and Kyoung Mu Lee},
  booktitle={European Conference on Computer Vision},
  year={2018},
  url={https://openaccess.thecvf.com/content_ECCV_2018/papers/Heewon_Kim_Task-Aware_Image_Downscaling_ECCV_2018_paper.pdf}
}

2x/4x Image Downscaling & Super-Resolution in a Single Mobile Architecture

Overview of Supernet Architecture

img

Progressive Shrinking of

  • Kernel Size
  • Network Depth
  • Expand Ratio
  • Number of Pixelshuffle

Comparison to CAR in terms of PSNR

"CAR: Learned Image Downscaling for Upscaling using Content Adaptive Resampler"

Dataset Ours CAR
Set14-2xUP 39.15 35.61
Set14-4xUP 31.01 30.30

img

img

About

Image downscaling & super-resolution project based on "Once for All: Train One Network and Specialize it for Efficient Deployment" (ICLR 2020)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 83.3%
  • Jupyter Notebook 16.7%