PPCN (Team: Rainbow)

[Paper_download][Paper_CVF][Paper_Springer]

Super-Resolution Task

The schematics of the proposed network for image super-resolution

Training

First, download the DIV2K dataset and unzip it in train_SR/ folder.

Run the following command to train the SR model

python train_SR.py

Testing

First, download the SR_Test_Datasets and put them in test/SR_test_data folder.

Run the following command to super-resolve low-resolution images

python evaluate_super_resolution.py

Enhancement Task

DPED image enhanced by our method

The structure of the proposed generator and discriminator for image enhancement

Training

• Step1: download the pre-trained VGG19 model and put it into train/vgg_pretrained/ folder
• Step2: download DPED dataset and extract it into train/dped/ folder.
• Step3: train the teacher model by executing the following command

python train_teacher.py

• Step4: train the student model by running

python train_student.py

Testing

Run the following command to enhance low-quality images

python evaluate_enhancement.py

Citation

If you find PPCN useful in your research, please consider citing:

@inproceedings{Hui-PPCN-2018,
  title={Perception-Preserving Convolutional Networks for Image Enhancement on Smartphones},
  author={Hui, Zheng and Wang, Xiumei and Deng, Lirui and Gao, Xinbo},
  booktitle={ECCV Workshop},
  pages = {197--213},
  year={2018}
}

Code References

[1]https://github.com/aiff22/ai-challenge

[2]https://github.com/aiff22/DPED

[3]https://github.com/roimehrez/contextualLoss