-
This pj owes greatly to MaybeShewill-CV.So please refer to Readme.md to learn about the way to operate on this pj.
-
My tiny work on attentive GAN(Tensorflow version) Your can find my report here.
-
Simply set
weights_path,gt_path,data_pathandsave_pathin predict.py, then run the command:
python tools/predict.py
You will get PSNR and SSIM metrics. When running niqe.m, you'd get NIQE metric.
| Model | Dataset | PSNR | SSIM | NIQE |
|---|---|---|---|---|
| Torch ver | test_a | 31.5160 | 0.9213 | 3.6448 |
| test_b | 24.9249 | 0.8091 | 4.1802 | |
| TF ver(trained by MaybeShewill-CV) | test_a | 25.7948 | 0.9043 | 3.7938 |
| test_b | 24.3363 | 0.8409 | 4.5349 | |
| TF ver(smoothed mask) | test_a | 26.1360 | 0.9109 | 3.6528 |
| test_b | 24.5049 | 0.8435 | 4.2612 | |
| TF ver(ssim-loss) | test_a | 26.3702 | 0.9151 | 3.5834 |
| test_b | 24.5303 | 0.8451 | 4.0724 |