notesum.ai
Published at December 5Targeted Hard Sample Synthesis Based on Estimated Pose and Occlusion Error for Improved Object Pose Estimation
cs.CV
Released Date: December 5, 2024
Authors: Alan Li1, Angela P. Schoellig1
Aff.: 1University of Toronto

| Object | Correct Detection Rate (10% ADD-S Error) | |||||
|---|---|---|---|---|---|---|
| DC-Net | MP-AAE | Sim2Real | FoundationPose | Ours (Baseline) | Ours (Pose+Occ.) | |
| Zigzag | 30.9 | 92.4 | 86.7 | 97.1 | 89.3 | 95.5 |
| Eye Bolt | 53.5 | 67.3 | 89.2 | 93.9 | 82.5 | 96.7 |
| DIN | 18.7 | 21.8 | 24.2 | 64.3 | 31.8 | 56.4 |
| Gear | 77.6 | 82.1 | 76.8 | 87.4 | 72.7 | 81.9 |
| Tube Fit. | 74.8 | 91.6 | 97.5 | 94.2 | 71.1 | 89.3 |
| Screw | 67.7 | 90.3 | 78.1 | 76.4 | 89.6 | 93.8 |
| D-Sub | 10.6 | 7.35 | 14.6 | 47.4 | 25.6 | 48.2 |