Pairwise Perceived Cycling Safety Modeling on PCS Berlin (test)
74.88Rank AccuracyGII injection
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GII injectionBackbone=DINOv3, Attention Mode=–2026.05 | 74.88 | 74.55 | |
| BaselineBackbone=DINOv3, Attention Mode=–2026.05 | 74.73 | 74.06 | |
| EG-PCS-NetBackbone=DINOv3, Attention Mode=Raw2026.05 | 74.66 | 74.64 | |
| EGViTBackbone=DINOv3, Attention Mode=–2026.05 | 74.41 | 74.21 | |
| EG-PCS-NetBackbone=DINOv3, Attention Mode=Rollout2026.05 | 74.32 | 74.6 | |
| EG-PCS-NetBackbone=DeiT III, Attention Mode=Rollout2026.05 | 74.2 | 73.93 | |
| EGViTBackbone=DeiT III, Attention Mode=–2026.05 | 73.94 | 73.35 | |
| BaselineBackbone=DeiT III, Attention Mode=–2026.05 | 73.84 | 73.2 | |
| EG-PCS-NetBackbone=DeiT III, Attention Mode=Raw2026.05 | 73.78 | 73.52 | |
| GII injectionBackbone=CLIP, Attention Mode=–2026.05 | 73.36 | 73.42 | |
| BaselineBackbone=CLIP, Attention Mode=–2026.05 | 73.34 | 72.11 | |
| EG-PCS-NetBackbone=CLIP, Attention Mode=Raw2026.05 | 73.32 | 73.06 | |
| GII injectionBackbone=DeiT III, Attention Mode=–2026.05 | 73.29 | 72.82 | |
| EGViTBackbone=CLIP, Attention Mode=–2026.05 | 73.03 | 72.15 | |
| EG-PCS-NetBackbone=CLIP, Attention Mode=Rollout2026.05 | 72.71 | 71.68 |