Submission: MRC-Net/YCB-V/PBR(DefaultDetections2022)

Download submission
Submission name PBR(DefaultDetections2022)
Submission time (UTC) March 18, 2024, 8:49 p.m.
User ymao
Task Model-based 6D localization of seen objects
Dataset YCB-V
Training model type Default
Training image type Synthetic (only PBR images provided for BOP Challenge 2020 were used)
Description One model for all objects, 2D bounding box: BOP 2022 default detections, Official implementation of our CVPR 2024 paper "MRC-Net: 6-DoF Pose Estimation with MultiScale Residual Correlation", Contributors: Yuelong Li, Yafei Mao, Raja Bala, Sunil Hadap
Evaluation scores
AR:0.681
AR_MSPD:0.804
AR_MSSD:0.666
AR_VSD:0.572
average_time_per_image:1.134

Method: MRC-Net

User ymao
Publication Yuelong Li*, Yafei Mao*, Raja Bala, Sunil Hadap: MRC-Net: 6-DoF Pose Estimation with MultiScale Residual Correlation, CVPR 2024.
Implementation https://github.com/amzn/mrc-net-6d-pose
Training image modalities RGB
Test image modalities RGB
Description

MRC-Net: 6-DoF Pose Estimation with MultiScale Residual Correlation, CVPR 2024, https://arxiv.org/abs/2403.08019, MRC-Net is a simple correspondence-free RGB-only 6DoF pose estimation model with leading accuracy and near real-time speed! It reformulates the conventional multitask classification+regression approach into a new sequential design. The two-sequential stages are bridged by a novel multi-scale correlation structure.

Computer specifications Nvidia V100