3D Video Object Detection on EmbodiedScan v1.0 (test)
54.2P@0.25GAP-MLLM-3B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GAP-MLLM-3B3D Input=false, Frame Setting=4-frame2026.03 | 54.2 | 48.1 | 50.6 | |
| GAP-MLLM-3B3D Input=false, Frame Setting=6-frame2026.03 | 52.9 | 45.4 | 48.5 | |
| VG-LLM-4B (w/ GAP)3D Input=false, Frame Setting=6-frame2026.03 | 48.8 | 40.2 | 43.6 | |
| VG-LLM-4B (w/ GAP)3D Input=false, Frame Setting=4-frame2026.03 | 47 | 41.2 | 43.5 | |
| VG-LLM-7B3D Input=false, Frame Setting=6-frame2026.03 | 43.5 | 38.7 | 40.8 | |
| VG-LLM-7B3D Input=false, Frame Setting=4-frame2026.03 | 43.4 | 39.6 | 41.2 | |
| SpatialLM3D Input=true, Frame Setting=6-frame2026.03 | 42.1 | 26.1 | 31.1 | |
| VG-LLM-4B3D Input=false, Frame Setting=4-frame2026.03 | 41.7 | 35.7 | 38.2 | |
| SpatialLM3D Input=true, Frame Setting=4-frame2026.03 | 40.8 | 23.7 | 29.1 | |
| VG-LLM-4B3D Input=false, Frame Setting=6-frame2026.03 | 39.7 | 34 | 36.4 |