Egocentric Deictic Visual Grounding on EgoPoint-Ground (Real-World)
99.4Precision@0.3Human Study
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Human Study2026.03 | 99.4 | 91.8 | 87.4 | |
| SV-CoTBase Model=Qwen3-vl2026.03 | 82 | 81.8 | 72.7 | |
| Qwen3-VL-8B2026.03 | 78.9 | 70.1 | 59.3 | |
| Qwen2.5-VL-7B2026.03 | 77.6 | 68.8 | 53.3 | |
| Qwen3-VL-4B2026.03 | 74.7 | 69.9 | 61.6 | |
| Qwen2.5-VL-3B2026.03 | 73.8 | 63.2 | 48.8 | |
| LLaVA-OneVision-4B2026.03 | 65.4 | 43.1 | 12.9 | |
| InternVL-3.5-8B2026.03 | 59.2 | 42.7 | 21.4 | |
| DeepSeek-VL22026.03 | 36.4 | 29.6 | 24.5 | |
| GroundingGPT2026.03 | 35.4 | 25.9 | 12.7 | |
| LLaVA-v1.5-13B2026.03 | 27.4 | 12.8 | 4.2 | |
| Ferret-7B2026.03 | 25.8 | 20.3 | 16 | |
| Florence2-Large2026.03 | 23.4 | 19.3 | 15.3 | |
| Florence2-Base2026.03 | 14.1 | 12.1 | 7.5 |