01
Catan Board Visualizer
Take one photo of a physical Settlers of Catan board and get back a clean digital version of it, plus every player's visible score.
- Built a CV referee using YOLO11 and homography to read and regenerate a Catan board from one photo and score players
- Labeled 3,100+ pieces and trained 5 YOLO11 models; cut false detections 43% by targeting lighting where the model failed
- Raised number-token accuracy from 80% to 100% by catching the model reading terrain, not digits, and masking it out
- Python
- PyTorch
- YOLO11
- OpenCV