parmanandsharma/Lightweight_AI — reverse-engineered prompt

Build me a Python package for lightweight medical image segmentation using the LWBNA Unet idea from the README. I want it to let researchers train a small Unet style model on biological or ophthalmic images without needing a complicated setup.

The main thing should be an easy training function where I can point it at folders of images and matching masks, choose image size, batch size, epochs, resize or preprocessing options, validation handling, early stopping, and a save location. It should also expose a function that returns the LWBNA Unet segmentation model so advanced users can compile it themselves with their own loss and optimizer.

Please include clear setup instructions, expected dataset folder structure, and a simple example script. Use Python with TensorFlow, plus common image and data libraries. Keep it suitable for limited hardware as much as possible, and make the README explain that commercial use may require checking the patent license.