RMPE: Regional Multi-person Pose Estimation

ICCV 2017  ·  Hao-Shu Fang, Shuqin Xie, Yu-Wing Tai, Cewu Lu ·

Multi-person pose estimation in the wild is challenging. Although state-of-the-art human detectors have demonstrated good performance, small errors in localization and recognition are inevitable. These errors can cause failures for a single-person pose estimator (SPPE), especially for methods that solely depend on human detection results. In this paper, we propose a novel regional multi-person pose estimation (RMPE) framework to facilitate pose estimation in the presence of inaccurate human bounding boxes. Our framework consists of three components: Symmetric Spatial Transformer Network (SSTN), Parametric Pose Non-Maximum-Suppression (NMS), and Pose-Guided Proposals Generator (PGPG). Our method is able to handle inaccurate bounding boxes and redundant detections, allowing it to achieve a 17% increase in mAP over the state-of-the-art methods on the MPII (multi person) dataset.Our model and source codes are publicly available.

PDF Abstract ICCV 2017 PDF ICCV 2017 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Keypoint Detection COCO AlphaPose Test AP 73.3 # 11
FPS 23 # 1
Pose Estimation COCO test-dev RMPE AP 61.8 # 31
AP50 83.7 # 28
AP75 69.8 # 26
APL 67.6 # 29
APM 58.6 # 23
Keypoint Detection COCO test-dev AlphaPose APL 81.5 # 3
Pose Estimation COCO test-dev RMPE++ AP 72.3 # 18
AP50 89.2 # 22
AP75 79.1 # 18
APL 78.6 # 13
APM 68.0 # 16
Multi-Person Pose Estimation COCO test-dev RMPE AP 61.8 # 13
APL 67.6 # 10
APM 58.6 # 9
AP50 83.7 # 8
AP75 69.8 # 7
Keypoint Detection MPII Multi-Person AlphaPose mAP@0.5 82.1% # 1
Multi-Person Pose Estimation MPII Multi-Person AlphaPose AP 82.1% # 1
Pose Estimation OCHuman RMPE Test AP 30.7 # 7
Validation AP 38.8 # 6
Keypoint Detection OCHuman RMPE Test AP 30.7 # 6
Validation AP 38.8 # 5
2D Human Pose Estimation OCHuman RMPE Test AP 30.7 # 6
Validation AP 38.8 # 5
Pose Estimation UAV-Human AlphaPose mAP 56.9 # 1

Methods