search

ck-kai / SGPA

[ICCV'21] SGPA: Structure-Guided Prior Adaptation for Category-Level 6D Object Pose Estimation
37 stars 4 forks source link

what's the performance of SGPA using pointnet rather than pointnet++？ #1

Closed Bingo-1996 closed 2 years ago

Bingo-1996 commented 2 years ago

Hi, author. Thank you for releasing such a wonderful work!

I have a question about "the performance of the model when replacing PointNet + + in SGPA with PointNet? Because SPD adopts PointNet instead of PointNet++." It seems that there are no ablation results for this change.

ck-kai commented 2 years ago

@Bingo-1996 Hi, thanks for your interest in our work.

If we directly apply our proposed prior adaptation module to SPD, the performance also gets significant improvement w.r.t. all metrics on two datasets. You can refer to the supplementary for detailed results. Since our proposed prior adaptation module is based on the geometry feature of point clouds, we therefore use PointNet++ as the point feature extractor, which provides more representative features for more effective prior adaptation and leads to a higher pose accuracy.

Bingo-1996 commented 2 years ago

Got it! Thank you! The idea of prior adaptation is really wonderful!