allenai/satlas-super-resolution — reverse-engineered prompt

Build me a Python project for satellite image super resolution like Satlas. I want to take low resolution Sentinel 2 image tiles and generate sharper, higher resolution looking imagery similar to NAIP, with scripts for both running inference with downloaded model weights and training or validating on the provided datasets.

Please make it easy to set up in a fresh environment, download the sample validation or test data and pretrained weights, then run one command that produces before and after image outputs I can inspect. Include sensible defaults for the main ESRGAN model, and keep support for the other provided model weights if it is already in the repo. Add clear README instructions for installing, downloading data, running inference, and doing a quick validation run.

I’m not trying to build a web app right now. I mostly want a clean research codebase that works end to end and can reproduce the main Satlas super resolution workflow. Look up current docs online if you need to.