Low performance compared to pix2pix

[XenMil]

Many thanks for the great repository! I've trained both pix2pixHD and pix2pix for an image enhancement task on a rather large dataset (over 30k images, 512x512). Surprisingly, pix2pixHD significantly underperforms compared to pix2pix on all common metrics (MSE, PSNR, SSIM), even after trying to optimize the model and training parameters.

Can you provide any hints as to why this is the case? Is the model unsuitable for this kind of tasks, or should I try any specific parameter values? Any help appreciated!