Regarding Fig. 8b, could you help clarify details about the validation dataset?
Using the released SAE_SPN, I used 500 randomly cropped 512x512 patches from Cityscapes validation images (leftImg8bit) but the FID is around 128.
I forgot hist_equal() initially, after correction, it's 115.96 16.77 0.518 0.768 for FID, PSNR, SSIM and FSIM respectively, still worse than reported metrics in the paper
Regarding Fig. 8b, could you help clarify details about the validation dataset? Using the released SAE_SPN, I used 500 randomly cropped 512x512 patches from Cityscapes validation images (leftImg8bit) but the FID is around 128.