Image Classification with Transfer Learning

You have only a few hundred labelled images per class, far too few to train an image model from scratch.

Transfer learning reuses a model already trained on millions of images, so you only fit a small head on your classes and reach high accuracy fast.

Use this for any custom image classifier (defects, categories, document types) with a modest labelled set.

Python, PyTorch or TensorFlow, and a folder of images arranged one subfolder per class.