SK124/Blood-Cell-Enumeration ? reverse-engineered prompt

Build me a notebook based project that can detect and count blood cells in microscope images, specifically red blood cells, white blood cells, and platelets.

I want it to use the BCCD dataset, turn the XML annotations into the text format needed for training, and include the full workflow from preprocessing to training to testing. Please set it up so I can train a YOLO style object detection model, ideally with custom anchor boxes calculated for this dataset, then run inference on new images and show the detected boxes with counts for each cell type.

It should also include a simple way to review performance on the test set, like precision, recall, F1, IoU, and mAP, plus sample visual results. If it helps, use the kind of training settings that work well for small, crowded objects. Keep it practical and notebook friendly, and look up current docs online if you need to.