Open-vocabulary 3D segmentation images (2D_seg)

Naming convention used by the Open-vocabulary 3D segmentation section:

• RGB (one per scene): {scene}_rgb.png — e.g. tum1_rgb.png, tum3_rgb.png
• Feature-3DGS / M3 / LatentAM (per scene + query): {scene}_{slot}_{query}.png

  • slot: feat

    m3

    pro (Feature-3DGS, M3, LatentAM)

  • query: see below

TUM1 queries

• book, monitor, keyboard

Example: tum1_feat_book.png, tum1_m3_monitor.png, tum1_pro_keyboard.png

TUM3 queries

• keyboard, monitor, teddy

Example: tum3_feat_teddy.png, tum3_m3_monitor.png, tum3_pro_keyboard.png

Columns (left to right)

1. RGB — {scene}_rgb.png
2. Feature-3DGS (Online) — {scene}_feat_{query}.png
3. M3 (Online) — {scene}_m3_{query}.png
4. LatentAM (ours) — {scene}_pro_{query}.png