notesum.ai
Published at November 22BIP3D: Bridging 2D Images and 3D Perception for Embodied Intelligence
cs.CV
cs.AI
Released Date: November 22, 2024
Authors: Xuewu Lin1, Tianwei Lin1, Lichao Huang1, Hongyu Xie1, Zhizhong Su1
Aff.: 1Horizon Robotics, Beijing, China

| Methods | Overall | Head | Common | Tail | Small | Medium | Large | ScanNet | 3RScan | MP3D |
|---|---|---|---|---|---|---|---|---|---|---|
| VoteNet [30] | 5.18 | 10.87 | 2.41 | 2.07 | 0.16 | 5.30 | 5.99 | 9.90 | 7.69 | 3.82 |
| ImVoxelNet [35] | 8.08 | 3.11 | 7.05 | 3.73 | 0.06 | 7.95 | 9.02 | 11.91 | 2.17 | 5.24 |
| FCAF3D [34] | 13.86 | 22.89 | 9.61 | 8.75 | 2.90 | 13.90 | 10.91 | 21.35 | 17.02 | 9.78 |
| EmbodiedScan-D [38] | 15.22 | 24.95 | 10.81 | 9.48 | 3.28 | 15.24 | 10.95 | 22.66 | 18.25 | 10.91 |
| BIP3D-RGB | 17.40 | 24.20 | 14.76 | 12.94 | 3.45 | 18.19 | 14.06 | 20.38 | 27.23 | 8.77 |
| BIP3D | 20.91 | 27.57 | 18.77 | 16.03 | 5.72 | 21.48 | 15.20 | 23.47 | 32.48 | 10.09 |