ZoomNAS: Searching for Whole-body Human Pose Estimation in the Wild

TL;DR

This paper introduces ZoomNAS, a neural architecture search framework for accurate and efficient 2D whole-body human pose estimation, supported by a new large-scale dataset COCO-WholeBody V1.0.

Contribution

The paper presents a novel neural architecture search method, ZoomNAS, tailored for whole-body pose estimation, and introduces the first large-scale dataset for this task.

Findings

ZoomNAS improves pose estimation accuracy.

The dataset enables comprehensive evaluation of whole-body pose models.

ZoomNAS achieves a good balance between accuracy and computational efficiency.

Abstract

This paper investigates the task of 2D whole-body human pose estimation, which aims to localize dense landmarks on the entire human body including body, feet, face, and hands. We propose a single-network approach, termed ZoomNet, to take into account the hierarchical structure of the full human body and solve the scale variation of different body parts. We further propose a neural architecture search framework, termed ZoomNAS, to promote both the accuracy and efficiency of whole-body pose estimation. ZoomNAS jointly searches the model architecture and the connections between different sub-modules, and automatically allocates computational complexity for searched sub-modules. To train and evaluate ZoomNAS, we introduce the first large-scale 2D human whole-body dataset, namely COCO-WholeBody V1.0, which annotates 133 keypoints for in-the-wild images. Extensive experiments demonstrate the effectiveness of ZoomNAS and the significance of COCO-WholeBody V1.0.