[Paper Review] PVN3D: A Deep Point-wise 3D Keypoints Voting Network for 6DoF Pose Estimation

Yisheng He, Wei Sun+4 more|arXiv (Cornell University)|Nov 11, 2019

Robot Manipulation and Learning61 references37 citations

TL;DR

PVN3D presents a 3D keypoint voting network with instance segmentation to estimate 6DoF pose from a single RGBD image, outperforming prior methods on YCB-Video and LineMOD.

ABSTRACT

In this work, we present a novel data-driven method for robust 6DoF object pose estimation from a single RGBD image. Unlike previous methods that directly regressing pose parameters, we tackle this challenging task with a keypoint-based approach. Specifically, we propose a deep Hough voting network to detect 3D keypoints of objects and then estimate the 6D pose parameters within a least-squares fitting manner. Our method is a natural extension of 2D-keypoint approaches that successfully work on RGB based 6DoF estimation. It allows us to fully utilize the geometric constraint of rigid objects with the extra depth information and is easy for a network to learn and optimize. Extensive experiments were conducted to demonstrate the effectiveness of 3D-keypoint detection in the 6D pose estimation task. Experimental results also show our method outperforms the state-of-the-art methods by large margins on several benchmarks. Code and video are available at https://github.com/ethnhe/PVN3D.git.

Motivation & Objective

• Motivate robust 6DoF pose estimation from RGB-D by leveraging 3D keypoints rather than direct pose regression.

Proposed method

• Develop a deep 3D keypoints Hough voting network that learns per-point translation offsets to selected 3D keypoints and uses clustering to obtain keypoints.

Experimental results

Research questions

• RQ1 Can 3D keypoint-based voting outperform direct pose regression and 2D-keypoint approaches for 6DoF pose estimation?

Key findings

• Outperforms state-of-the-art methods on YCB-Video and LineMOD datasets in ADD-S and ADD-S AUC metrics.
• 3D keypoint formulation outperforms direct pose regression and 2D keypoint approaches in experiments.
• Jointly training 3D keypoint voting with instance semantic segmentation improves pose accuracy and segmentation quality.