Research Article
GAN-Based Image Super-Resolution with a Novel Quality Loss
Table 3
Test results (PI/SSIM/PSNR(dB)) when using four different training datasets.
| | Datasets | Set5 | Set14 | BSD100 |
| | ImageNet | 3.85/0.8510/30.12 | 2.97/0.6898/26.51 | 2.40/0.6470/25.31 | | DF2K | 3.70/0.8447/30.40 | 2.87/0.7055/25.96 | 2.34/0.6601/25.56 | | DF2K + OST | 3.43/0.8497/30.31 | 2.96/0.7015/26.60 | 2.37/0.6590/25.70 | | Augmentation | 3.25/0.8342/30.02 | 2.77/0.7021/26.37 | 2.29/0.6593/25.46 |
|
|
Bold indicates the best and italics indicate the second best performance in terms of PI. PSNR and SSIM are provided for reference.
|
