sml2h3/ddddocr ? reverse-engineered prompt

Build me a simple Python tool for recognizing captcha images locally.

I want it to take an image file or raw bytes and return the text it sees, with a small demo script so I can try it right away. It should also handle slider captchas by finding the missing piece location, and support a basic target detection mode too. Please make the setup easy, keep dependencies light, and include a few example images or sample inputs so I can test it without extra work.

If it helps, add a tiny command line way to run it on one image or a folder of images, and let me turn on GPU if it’s available. It would be great if the code is clean enough that I can also plug in my own model later. Look up current docs online if you need to, and make sure the examples are easy to follow.