rapidonot/wafer-defect-classifier

This repository builds a multi-label wafer-bin-map defect classifier and provides a Streamlit image-upload demo. The user workflow is upload an image, convert it to the model’s categorical input, run ResNet-18 inference, then view confidence scores and thresholded defect labels. Separate workflows prepare datasets, train ResNet-18 or Vanilla CNN models with BCE or focal loss, and benchmark predictions. The tree and excerpts support these major capabilities, but do not establish that training produces the dashboard’s bundled weights or that the evaluation model path exists.