sdjnzxc / dwsrx4 Goto Github PK

The testing code for Deep wavelet prediction for image super-resolution, CVPRW, 2017, NTIRE 2017 Super-Resolution Challenge - DWSRx4.

Python package requirement:

• tensorflow w/GPU @ https://github.com/tensorflow/tensorflow
• pywt @ https://github.com/PyWavelets/pywt
• cv2 @ https://github.com/opencv/opencv

1. In terminal, type in python DWSRx4.py
2. Then a promote asks for testing data set: Please enter the testing path [hit enter to run default set]:
3. Hit enter to run default testing set from DIV2K NTIRE which is stored at: ./Testx4Lum
4. The final results will be stored at: ./Resultx4Lum
5. Run FinalColorSRx4.m to generate final color SR and store the results in ./Resultx4Color

1. The testing data should be bicubic enlarged version of the original down-sampled version. For example, to generate x4 super-resolution results, the original x4 down-sampled low-resolution image should first be enlarged to x4 size, then fed the enlarged version to DWSR (as described in the fact sheet). Use generateTestX4.m to generate enlarged LR luminance image.
2. The DWSR weights are stored at: ./Weightx4
3. The DWSR model is defined in: netx4.py
4. The script is NOT for training.

The training code is not fully cleaned up; for academic purpose, please request training from here by providing basic usage information.

@inproceedings{guo2017deep,
  title={Deep wavelet prediction for image super-resolution},
  author={Guo, Tiantong and Mousavi, Hojjat Seyed and Vu, Tiep Huu and Monga, Vishal},
  booktitle={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  year={2017}
}

@inproceedings{timofte2017ntire,
  title={Ntire 2017 challenge on single image super-resolution: Methods and results},
  author={Timofte, Radu and Agustsson, Eirikur and Van Gool, Luc and Yang, Ming-Hsuan and Zhang, Lei and Lim, Bee and Son, Sanghyun and Kim, Heewon and Nah, Seungjun and Lee, Kyoung Mu and others},
  booktitle={Computer Vision and Pattern Recognition Workshops (CVPRW), 2017 IEEE Conference on},
  pages={1110--1121},
  year={2017},
  organization={IEEE}
}

Tiantong@iPAL2017, [email protected]