Submission: Megapose-CNOS_fastSAM+MultiHyp/YCB-V

Download submission
Submission name
Submission time (UTC) Sept. 26, 2023, 3 p.m.
User agimus-happypose
Task Model-based 6D localization of unseen objects
Dataset YCB-V
Training model type Default
Training image type Synthetic (provided)
Description Submission prepared by Médéric Fourmy, Elliot Maître, Yann Labbé
Evaluation scores

Method: Megapose-CNOS_fastSAM+MultiHyp

User agimus-happypose
Publication Labbé et al.: MegaPose: 6D Pose Estimation of Novel Objects via Render & Compare, CoRL 2022
Training image modalities RGB
Test image modalities RGB

This submission was prepared by Mederic Fourmy, Elliot Maître, Lucas Manuelli, Yann Labbé

In this submission, CNOS_fastSAM [A] detections (default detections for Task 4) are used as input to the MegaPose pose estimation method [B].

For each detection, we run the MegaPose coarse network. Then, each of the top-5 hypotheses are refined using 5 iterations of the refinement network. The refined hypotheses are then scored using the coarse network and the best one is considered the pose estimate for the initial detection.

The following improvement has been made over the original MegaPose paper [B]: - The orientation of the coarse hypotheses are obtained my discretizing SO(3) in 576 values using [C].
- The refinement network is ran on multiple pose hypotheses coming from the coarse network.

Instructions to reproduce these results will be made available on

[A] Nguyen et al.: CNOS: A Strong Baseline for CAD-based Novel Object Segmentation, arXiv 2023
[B] Labbé et al.: MegaPose: 6D Pose Estimation of Novel Objects via Render & Compare, CoRL 2022
[C] Yershova et al.: Generating Uniform Incremental Grids on SO(3) Using the Hopf Fibration, IJRR 2009, reference implementation:

This work was granted access to the HPC resources of IDRIS under the allocation 011014301 made by GENCI

Computer specifications NVIDIA Tesla V100 32Go