Open-vocabulary 2D object retrieval and localization on LERF
63.4mIoU (Ramen Scene)LESV
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| LESVmulti-level SAM mask generation=false, feature backbone=Radio-v32026.04 | 63.4 | 88.73 | 48.41 | 85.71 | 69.3 | 84.75 | 46.24 | 77.27 | 56.84 | 84.12 | |
| LangSplatV22026.04 | 51.8 | 74.7 | 56.4 | 82.1 | 72.2 | 93.2 | 59.1 | 86.4 | 59.88 | 84.1 | |
| LangSplat2026.04 | 51.2 | 73.2 | 44.7 | 80.4 | 65.1 | 88.1 | 44.5 | 95.5 | 51.38 | 84.3 | |
| GAGS2026.04 | 46.8 | 69 | 53.6 | 78.6 | 60.3 | 88.1 | 55.8 | 90.9 | 54.13 | 81.65 |