pnbao / rsgunet-testing

Team: Mt.Phoenix (1st place)

• http://ai-benchmark.com/challenge.html#challenge

How to use the code:

1. Use a checkpoint file to save your model and a dataset file to place your dataset.
2. Use a pre-trained vgg model which can be found on the https://drive.google.com/file/d/0BwOLOmqkYj-jMGRwaUR2UjhSNDQ/view?usp=sharing
3. To train the model you just need run the squid/train.py.

Requierments

• tensorflow 1.15

data structure

BASE_FOLER/ ┐ ├ training_data/ ┐ │ ├ │ ├ canon/ ┐ │ │ ├ 0.jpg │ │ : │ ├ iphone/ ┐ │ │ ├ 0.jpg │ │ : │ └ dataset.txt └ test_data/patches/┐ ├ ├ canon/ ┐ │ ├ 0.jpg │ : ├ iphone/ ┐ │ ├ 0.jpg │ : ├ dataset.txt └ imagenet-vgg-verydeep-19.mat

Please cite our paper:

@InProceedings{RSGUNet2018,
author = {J. Huang and P. Zhu and M. Geng and J. Ran and X. Zhou and C. Xing and P. Wan and X. Ji},
title={Range Scaling Global U-Net for Perceptual Image Enhancement on Mobile Devices},
booktitle={European Conference on Computer Vision Workshops},
year={2018},
}

• http://openaccess.thecvf.com/content_ECCVW_2018/papers/11133/Huang_Range_Scaling_Global_U-Net_for_Perceptual_Image_Enhancement_on_Mobile_ECCVW_2018_paper.pdf