| Submission name |
|
| Submission time (UTC) |
Oct. 12, 2022, 10:19 a.m.
|
| User |
Transfer-BOP
|
| Task |
Model-based 6D localization of seen objects |
| Dataset |
ITODD |
| Training model type |
Default |
| Training image type |
None |
| Description |
A variant implementation of Pair Point Feature algorithm with a normal-based clustering preprocessing step is introduced. Further more, we adopt an efficient visibility-based scoring process to measure weights of points for voting, improved method of pose hypothesis step to overcome the influence quadrature error make and utilize a method to remove overlapping instances before refinement for better performance.
Point-to-plane ICP is used for refinement.
CPU only, no GPU is used.
|
| Evaluation scores |
| AR: | 0.481 |
| AR_MSPD: | 0.501 |
| AR_MSSD: | 0.506 |
| AR_VSD: | 0.437 |
| average_time_per_image: | -1.000 |
|