Spatial-Temporal Reasoning and Trajectory Analysis on STB24 (test)
100Direction Detection AccuracySTAR
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| STARBackbone=Qwen3-8B, Parameters=8B, Deployment=local2026.05 | 100 | 81.7 | 67.4 | 92.4 | 36.3 | 112.6 | 100 | 90.4 | 89.1 | 62.5 | 47.6 | |
| GPT-4o2026.05 | 54.3 | 11 | 60.2 | 91.9 | 43.2 | — | — | 95.9 | 96.6 | 75.5 | 60.3 | |
| Open-7B‡Model Class=7B-class, Type=open-source2026.05 | 19.7 | 27.9 | 51 | 90.4 | 31.9 | 139.4 | 55.4 | 44.6 | 30.1 | 38.9 | 27 | |
| ChatGPT2026.05 | 17 | 27.9 | 53.8 | 92.4 | 37.3 | — | 75.2 | 79.3 | 83.6 | 43.8 | 39.8 |