miseplastic/firstmise

This is a microplastic-detection dataset research repository, currently focused on auditing train/validation data for image integrity, possible duplicates, and annotation-layout similarities. The documented workflow uses a notebook to create an image inventory, then an audit script to analyze the datasets and produce artifacts summarized in a report. Group-based splitting and YOLOv5s evaluation are planned, not implemented. The graph preserves those research-planning records but does not depict them as runtime stages; unsampled notebook behavior is included only as documented in the README.