Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields

TL;DR

This paper introduces Part Affinity Fields, a real-time, bottom-up approach for multi-person 2D pose estimation that accurately associates body parts to individuals in images.

Contribution

It proposes a novel Part Affinity Fields representation and a joint learning architecture that enables efficient, accurate multi-person pose estimation in real-time.

Findings

Achieved first place in COCO 2016 keypoints challenge.

Significantly outperformed previous methods on MPII Multi-Person benchmark.

Operates efficiently regardless of the number of people in the image.

Abstract

We present an approach to efficiently detect the 2D pose of multiple people in an image. The approach uses a nonparametric representation, which we refer to as Part Affinity Fields (PAFs), to learn to associate body parts with individuals in the image. The architecture encodes global context, allowing a greedy bottom-up parsing step that maintains high accuracy while achieving realtime performance, irrespective of the number of people in the image. The architecture is designed to jointly learn part locations and their association via two branches of the same sequential prediction process. Our method placed first in the inaugural COCO 2016 keypoints challenge, and significantly exceeds the previous state-of-the-art result on the MPII Multi-Person benchmark, both in performance and efficiency.