shangqigao / gsq-image-SR

Multi-scale deep neural networks for real image super-resolution

MsDNN

This is an official implementation of "Multi-scale deep neural networks for real image super-resolution" via TensorFlow

Dependencies

• Tensorflow-gpu==1.9
• Python==2.7

Illustration

• MsDNN/model/msdnn_feed_v1_96a96_64blocks_1000000: a multi-scale model with 64 residual blocks
• Warning: The file model.checkpoint-999999.data-00000-of-00001 in the fold msdnn_feed_v1_96a96_64blocks_1000000 is large, one should download it alone and put into the corresponding folder.
• MsDNN/RealSR/Test_LR: The testing dataset provided by the NTIRE2019 SR challenge
• MsDNN/msdnn.py: The detailed structure of MsDNN
• MsDNN/msdnn_demo.py: The codes of obtaining high-resolution images

Quick test

Commands of getting high-resolution images:

python2 msdnn_demo.py

After executing the above command, there will exist a folder MsDNN/RealSR/Test_HR, which is the super-resolution of testing dataset.

Citation

If you find our work useful in your research or publication, please cite our work:

[1] Shangqi Gao, and Xiahai Zhuang, "Multi-scale deep neural networks for real image super-resolution", CVPR Workshops, 2019. [PDF] [arXiv]

@INPROCEEDINGS{msdnn/cvprw/2019, 
  author={S. {Gao} and X. {Zhuang}}, 
  booktitle={2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},   
  title={Multi-Scale Deep Neural Networks for Real Image Super-Resolution},   
  year={2019}, 
  pages={2006-2013}
}