Infrastructure Improvement Proposal on Mapillary (val)
83.42RecallQwen2-VL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2-VL2026.03 | 83.42 | 21.43 | 32.98 | 21.07 | |
| Ours BaselineVision Encoder=ViT-B, Text Encoder=Long-CLIP2026.03 | 76.84 | 79.68 | 78.24 | 62.43 | |
| LLaVA-1.52026.03 | 75.52 | 14.71 | 22.03 | 14.05 | |
| Ours BaselineVision Encoder=Long-CLIP, Text Encoder=Long-CLIP2026.03 | 75.49 | 80.22 | 77.78 | 62.5 | |
| Ours BaselineVision Encoder=CLIP, Text Encoder=Long-CLIP2026.03 | 74.75 | 80.81 | 77.66 | 61.68 | |
| Ours BaselineVision Encoder=ResNet-50, Text Encoder=Long-CLIP2026.03 | 71.68 | 77.28 | 74.38 | 58.48 | |
| InternVL22026.03 | 68.27 | 21.07 | 30.73 | 20.45 | |
| OD-RASE (ResNet-50)Vision Encoder=ResNet-50, Text Encoder=RoBERTa-Base, Evaluation Setting=zero-shot, Training Dataset=BDD100K2026.03 | 64.38 | 69.08 | 66.65 | 34.97 | |
| OD-RASE (Long-CLIP)Vision Encoder=Long-CLIP, Text Encoder=RoBERTa-Base, Evaluation Setting=zero-shot, Training Dataset=BDD100K2026.03 | 62.68 | 74.91 | 68.25 | 39.29 | |
| OD-RASE (ViT-B)Vision Encoder=ViT-B, Text Encoder=RoBERTa-Base, Evaluation Setting=zero-shot, Training Dataset=BDD100K2026.03 | 62.49 | 74.34 | 67.9 | 38.02 | |
| OD-RASE (CLIP)Vision Encoder=CLIP, Text Encoder=RoBERTa-Base, Evaluation Setting=zero-shot, Training Dataset=BDD100K2026.03 | 60.64 | 73.76 | 66.56 | 38.02 | |
| Phi-32026.03 | 52.15 | 21.69 | 28.94 | 18.38 | |
| GPT-4o2026.03 | 47.19 | 17.86 | 25.27 | 16.86 | |
| GPT-4oEvaluation Setting=zero-shot2026.03 | 43.07 | 20.83 | 26.49 | 19.94 | |
| Phi-3Evaluation Setting=zero-shot2026.03 | 41.7 | 22.01 | 27.48 | 17.47 | |
| Qwen2-VLEvaluation Setting=zero-shot2026.03 | 41.68 | 17.94 | 24.09 | 15.46 | |
| LLaVA-1.5Evaluation Setting=zero-shot2026.03 | 39.78 | 22.67 | 22.51 | 15.88 | |
| InternVL2Evaluation Setting=zero-shot2026.03 | 38.23 | 14.87 | 20.17 | 12.73 |