Safety Map Segmentation on Custom Urban Delivery Dataset
79.7IoUSee&Say
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| See&SayMultiple-run evaluation=5x per frame2026.04 | 79.7 | 88 | 88 | 91.3 | 86.8 | 89.1 | 88.6 | 88 | |
| RTDETR+SAM2+GradientModel components=RT-DETR, SAM2, Gradient2026.04 | 61.9 | 74.4 | 74.4 | 96.5 | 62.8 | 98.5 | 81.7 | 80.6 | |
| YOLOv8+SAM2+GradientModel components=YOLOv8, SAM2, Gradient2026.04 | 61.9 | 74.4 | 74.4 | 96.6 | 62.7 | 98.6 | 81.7 | 80.6 | |
| DINOx+GradientMultiple-run evaluation=5x per frame, Model components=DINO-X, Gradient2026.04 | 61.9 | 74.4 | 74.4 | 96.6 | 62.7 | 98.6 | 81.7 | 80.6 | |
| GradientMethod components=Gradient2026.04 | 61 | 73.6 | 73.6 | 96.5 | 61.8 | 98.6 | 81.3 | 80.2 | |
| DINOx (Flat ground category)Multiple-run evaluation=5x per frame, Prompt category=Flat ground2026.04 | 56 | 69 | 69 | 63.6 | 85.7 | 47.7 | 63.3 | 66.7 | |
| Vanilla VLM (o3)Multiple-run evaluation=5x per frame, Model version=o32026.04 | 53.5 | 68.7 | 68.7 | 70.6 | 68.4 | 78.1 | 74.6 | 73.2 | |
| RTDETR+SAM2Model components=RT-DETR, SAM22026.04 | 8.2 | 14.4 | 14.4 | 99.3 | 8.2 | 100 | 59.2 | 54.1 | |
| YOLOv8+SAM2Model components=YOLOv8, SAM22026.04 | 8.2 | 14.3 | 14.3 | 95.6 | 8.2 | 100 | 59.3 | 54.1 | |
| DINOx (VisDrone category)Multiple-run evaluation=5x per frame, Prompt category=VisDrone2026.04 | 8.1 | 14.2 | 14.2 | 99.7 | 8.1 | 100 | 59.2 | 54.1 |