GAN-based image super-resolution using perceptual content losses (tensorflow implementation)
This repository include a implementation of GAN-based single image super-resolution using perceptual content losses (PCL), which considers both distortion- and perception-based quality of super-resolved images. In the PIRM Challenge on Perceptual Super Resolution at ECCV 2018, Our team (Yonsei-MCML) won the 2nd place for Region 1. (Our team also won the 2nd place for Region 2 based on 4PP-EUSR model.)
Please cite following papers when you use the code, pre-trained models, or results:
- M. Cheon, J.-H. Kim, J.-H. Choi, J.-S. Lee: Generative adversarial network-based image super-resolution using perceptual content losses. arXiv:1809.04783 (2018) (To appear in ECCV 2018 workshop)
- J.-H. Kim, J.-S. Lee: Deep residual network with enhanced upscaling module for super-resolution. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pp. 913-921 (2018)
The perceptual index is calculated by two no-reference quality measurements, Ma and NIQE. Lower score means better perceptual quality. The detail of this index is explained in the PIRM Challenge.
Final ranking of our method (Yonsei-MCML) (please check the details in PIRM website)
The instructions for the usage of testing code is below. Generating super-resolved images from the pre-trained models can be done by <test/test.py>. Now, we only support x4 super-resolution for the challenge.
- Download and copy the trained model available in Downloads section to the
<test/>folder. - Place the low-resolution images (PNG files) to the
<test/LR/>folder. - Run
<python test.py> - The super-resolved images will be available on the
<test/LR/>folder.
The training code will be upaded soon.
Pre-trained models (for the PIRM Challenge)
- Download PIRM Challenge version : eusr-pcl_pirm.pb