notesum.ai
Published at November 25Diffusion Features for Zero-Shot 6DoF Object Pose Estimation
cs.CV
Released Date: November 25, 2024
Authors: Bernd Von Gimborn1, Philipp Ausserlechner2, Markus Vincze2, Stefan Thalhammer1
Aff.: 1Department of Industrial Engineering, University of Applied Sciences Technikum Wien; 2Automation and Control Institute, TU Wien

| Method | Training | LMO | YCBV | TLESS | |
|---|---|---|---|---|---|
| a | CosyPose [27] | instance-level | 0.536 | 0.333 | 0.520 |
| GDRnet [30] | instance-level | 0.672 | 0.755 | 0.512 | |
| HiPose [5] | instance-level | 0.799 | 0.907 | 0.833 | |
| b | MegaPose [7] | supervised zero-shot | 0.187 | 0.139 | 0.197 |
| OSOP [9] | supervised zero-shot | 0.274 | 0.296 | 0.403 | |
| GigaPose [40] | supervised zero-shot | 0.299 | 0.290 | 0.273 | |
| c | FoundPose | unsupervised zero-shot | 0.396 | 0.452 | 0.338 |
| \cdashline2-6 | ZS6D [6] | unsupervised zero-shot | 0.298 | 0.324 | 0.210 |
| DZOP ([6]+ours) | unsupervised zero-shot | 0.328 | 0.365 | 0.267 |