Questions about the experimental part

[raywzy]

For the quantitative comparsions, how do you obtain the numerical scores such as PSNR, SSIM of other methods? Do you re-train all the methods using your own training data or just use their pretrained model?

[Lotayou]

We retrained (or fine-tuned) most of the methods on our dataset. However, some baselines did not release the training code (for instance GFRNet and RIDNet), in which case we evaluate on the pretrained model instead.

[raywzy]

We retrained (or fine-tuned) most of the methods on our dataset. However, some baselines did not release the training code (for instance GFRNet and RIDNet), in which case we evaluate on the pretrained model instead.

Thanks for your reply. When do you plan to release the pretrained model? I reall want to try the performance.

[Lotayou]

We retrained (or fine-tuned) most of the methods on our dataset. However, some baselines did not release the training code (for instance GFRNet and RIDNet), in which case we evaluate on the pretrained model instead.

Thanks for your reply. When do you plan to release the pretrained model? I reall want to try the performance.

This work is done at Alibaba and the code release is subject to company regulations. If you are keen to get started, try first switch the input of SPADE with LQ face images and see if it works. Good luck:)

[Lotayou]

@raywzy Checkpoint released