Fine-tuning classification networks and using GradCAM to obtain object detection results on histopathological slides of breast cancer.
pip install breakhis_gradcam
Later
- Examples of how to use this package
- Using some form of unsupervised bounding box detection (e.g. attention cropping) to pull bounding box predictions
- This is probably more reasonable to do with a dataset that has these labels, as more training can be done to get better bounding box predictions
- Using a variant of the SimCLR framework to fine-tune the trained models based on similarity between detected regions of slides within the same subclass