Wide Activation for Efficient and Accurate Image Super-Resolution

Tech Report | Approach | Results | TensorFlow | Other Implementations | Bibtex

Update (Oct, 2018): We have re-implemented WDSR on TensorFlow for end-to-end training and testing. Pre-trained models are released. The runtime speed of weight normalization on tensorflow is also optimized.

Run

0. Requirements:
   • Install PyTorch (tested on release 0.4.0 and 0.4.1).
   • Clone EDSR-Pytorch as backbone training framework.
1. Training and Validation:
   • Copy wdsr_a.py, wdsr_b.py into EDSR-PyTorch/src/model/.
   • Modify EDSR-PyTorch/src/option.py and EDSR-PyTorch/src/demo.sh to support --n_feats, --block_feats, --[r,g,b]_mean option (please find reference in issue #7, #8).
   • Launch training with EDSR-Pytorch as backbone training framework.
2. Still have questions?
   • If you still have questions, please first search over closed issues. If the problem is not solved, please open a new issue.

Overall Performance

Network	Parameters	DIV2K (val) PSNR
EDSR Baseline	1,372,318	34.61
WDSR Baseline	1,190,100	34.77

We measured PSNR using DIV2K 0801 ~ 0900 (trained on 0000 ~ 0800) on RGB channels without self-ensemble. Both baseline models have 16 residual blocks.

More results:

Number of Residual Blocks	1			3
SR Network	EDSR	WDSR-A	WDSR-B	EDSR	WDSR-A	WDSR-B
Parameters	0.26M	0.08M	0.08M	0.41M	0.23M	0.23M
DIV2K (val) PSNR	33.210	33.323	33.434	34.043	34.163	34.205

Number of Residual Blocks	5			8
SR Network	EDSR	WDSR-A	WDSR-B	EDSR	WDSR-A	WDSR-B
Parameters	0.56M	0.37M	0.37M	0.78M	0.60M	0.60M
DIV2K (val) PSNR	34.284	34.388	34.409	34.457	34.541	34.536

Comparisons of EDSR and our proposed WDSR-A, WDSR-B using identical settings to EDSR baseline model for image bicubic x2 super-resolution on DIV2K dataset.

WDSR Network Architecture

Left: vanilla residual block in EDSR. Middle: wide activation. Right: wider activation with linear low-rank convolution. The proposed wide activation WDSR-A, WDSR-B have similar merits with MobileNet V2 but different architectures and much better PSNR.

Weight Normalization vs. Batch Normalization and No Normalization

Training loss and validation PSNR with weight normalization, batch normalization or no normalization. Training with weight normalization has faster convergence and better accuracy.

Other Implementations

• TensorFlow-WDSR (official)
• Keras-WDSR By Martin Krasser

Citing

Please consider cite WDSR for image super-resolution and compression if you find it helpful.

@article{yu2018wide,
  title={Wide Activation for Efficient and Accurate Image Super-Resolution},
  author={Yu, Jiahui and Fan, Yuchen and Yang, Jianchao and Xu, Ning and Wang, Xinchao and Huang, Thomas S},
  journal={arXiv preprint arXiv:1808.08718},
  year={2018}
}

@inproceedings{fan2018wide,
  title={Wide-activated Deep Residual Networks based Restoration for BPG-compressed Images},
  author={Fan, Yuchen and Yu, Jiahui and Huang, Thomas S},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
  pages={2621--2624},
  year={2018}
}