phillipi/pix2pix ? reverse-engineered prompt

Build me a working pix2pix image to image translation project from this repo.

I want to be able to take paired images, like sketches and photos, maps and satellite views, or labels and facades, train a model on them, then run it on new images and see the generated results saved to a simple web page or image folder. Please include the basic setup so I can point it at my own dataset, with train, val, and test folders, and make it easy to switch the translation direction either way.

If there are sample datasets or pretrained models already supported here, wire those up too so I can try it quickly. I’d also like clear training and testing commands, and if there’s a way to watch results while training, set that up as well. Keep it runnable in the current Torch and Lua setup from the repo, and look up current docs online if you need to.