FUT-HEROS — RF-DETR-Seg-Small (football: ball / player / goal)
RF-DETR-Seg-Small instance-segmentation model that detects ball, player, goal in amateur football clips. Distilled from SAM3 + NVIDIA LocateAnything-3B auto-labels with zero manual annotation — see FUT-HEROS for the full pipeline.
Results (held-out 12-clip test set, via the FUT-HEROS pipeline)
| detector | goal | leg | pose-capture |
|---|---|---|---|
| SAM3+LA (teacher) | 83% | 82% | 92% |
| this model (student) | 75% | 75% | 100% |
Real-time, runs ~50x faster than the SAM3+LA teacher; beats it on pose (cleaner player masks) and recovers the ball on clips SAM3 missed entirely.
Use
from rfdetr import RFDETRSegSmall
m = RFDETRSegSmall(pretrain_weights="checkpoint_best_ema.pth")
det = m.predict(image, threshold=0.3) # classes: ball, player, goal
Classes: 0 ball, 1 player, 2 goal. Trained at 384px on ~1.8k auto-labelled frames.