isl-org/MIDAS ? reverse-engineered prompt

Build me a simple depth estimation app that can take one or more photos and turn them into depth maps from a single image. I want it to work both on a folder of input images and from my webcam, and I want the results saved to an output folder as easy to view depth images. It should have a few model choices so I can pick between best quality and faster speed, and it should work well with the pretrained weights I drop into a weights folder.

Please make it easy to run locally, and also keep a Docker option so I can launch it without much setup. If there’s a straightforward way to support PyTorch Hub too, include that. Keep the interface simple, just one command to run inference, and let me optionally show the original image beside the depth map when using the camera. If you need to check current install or model details online, go ahead and do that.