Submission: GigaPose+MegaPose's refinement/HB/GigaPose + MegaPose's refinement

Download submission
Submission name GigaPose + MegaPose's refinement
Submission time (UTC) Nov. 27, 2023, 3:49 p.m.
User nvnguyen
Task 6D localization of unseen objects
Dataset HB
Description We use GigaPose to extract 5 hypothesis and run MegaPose's refinement.
Evaluation scores
AR:0.722
AR_MSPD:0.765
AR_MSSD:0.713
AR_VSD:0.688
average_time_per_image:8.426

Method: GigaPose+MegaPose's refinement

User nvnguyen
Publication https://arxiv.org/abs/2311.14155
Implementation https://github.com/nv-nguyen/gigapose
Training image modalities RGB
Test image modalities RGB
Description

GigaPose: Fast and Robust Novel Object Pose Estimation via One Correspondence.

GigaPose is a "hybrid" template-patch correspondence approach to estimate 6D pose of novel objects in RGB images: GigaPose first uses templates, rendered images of the CAD models, to recover the out-of-plane rotation (2DoF) and then uses patch correspondences to estimate the remaining 4DoF.

We use GigaPose to extract 5 hypothesis and run MegaPose's refinement.

Computer specifications V100