Fine localization on LangLoc 3RScan 100-scene 1.0 (test)
0.926Positional Error (m) MeanLangLoc w/ dialog
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LangLoc w/ dialogdialog=true, dialog_source=human annotator2026.07 | 0.926 | 0.983 | 0.89 | 0.799 | 39.52 | 33.37 | 34.2 | |
| Midpointdefinition=floor centroid2026.07 | 1.416 | — | — | 1.347 | — | — | — | |
| VLMbackbone=Qwen2.5-VL-2B, rendering=top-down, zero-shot=true2026.07 | 1.582 | — | — | 1.42 | 85.48 | 85.54 | 6.2 | |
| LangLoc w/o dialogdialog=false2026.07 | 1.712 | 1.037 | 0.941 | 1.551 | 46.07 | 37.24 | 17.2 |