microsoft/Swin-Transformer ? reverse-engineered prompt

Build me a clean, easy to use image recognition project based on Swin Transformer.

I want to be able to train and test it on my own image data, and also try a pretrained model right away on a sample folder of images. Please make the setup straightforward, with simple commands for running classification and for evaluating a checkpoint. If it makes sense, include a small demo or example script that shows the model making predictions on an image.

Keep the project organized so it is easy to switch between the common Swin model sizes and the newer variants mentioned in the repo, and make sure the training and logging feel solid. If there are current best practices or docs I should follow, look them up online and use them. I do not need the other research tasks right now, just a practical version of the Swin Transformer code that I can run without fighting the setup.