Submission: Coupled Iterative Refinement/YCB-V

Download submission
Submission name
Submission time (UTC) April 24, 2022, 9:58 p.m.
User lahav
Task Model-based 6D localization of seen objects
Dataset YCB-V
Training model type Default
Training image type Synthetic + real
Description Single view. Dataset splits: train_real, train_synt
Evaluation scores
AR:0.893
AR_MSPD:0.885
AR_MSSD:0.924
AR_VSD:0.871
average_time_per_image:-1.000

Method: Coupled Iterative Refinement

User lahav
Publication Coupled Iterative Refinement for 6D Multi-Object Pose Estimation, CVPR 2022
Implementation https://github.com/princeton-vl/coupled-iterative-refinement
Training image modalities RGB-D
Test image modalities RGB-D
Description

We propose a new approach to 6D object pose estimation which consists of an end-to-end differentiable architecture that makes use of geometric knowledge. Our approach it- eratively refines both pose and correspondence in a tightly coupled manner, allowing us to dynamically remove outliers to improve accuracy. We use a novel differentiable layer to perform pose refinement by solving an optimization problem we refer to as Bidirectional Depth-Augmented Perspective- N-Point (BD-PnP).

Computer specifications 2 RTX-3090s