OmniPose: A Multi-Scale Framework for Multi-Person Pose Estimation

18 Mar 2021 Bruno Artacho Andreas Savakis

We propose OmniPose, a single-pass, end-to-end trainable framework, that achieves state-of-the-art results for multi-person pose estimation. Using a novel waterfall module, the OmniPose architecture leverages multi-scale feature representations that increase the effectiveness of backbone feature extractors, without the need for post-processing... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Multi-Person Pose Estimation COCO OmniPose AP 0.764 # 2
Validation AP 79.5 # 1
Multi-Person Pose Estimation COCO OmniPose-Lite Validation AP 71.4 # 4
Pose Estimation Leeds Sports Poses OmniPose PCK 99.5% # 1
Pose Estimation MPII Human Pose OmniPose (WASPv2) PCKh-0.5 92.3 # 7
Pose Estimation UPenn Action OmniPose Mean PCK@0.2 99.4 # 1

Methods used in the Paper


METHOD TYPE
Batch Normalization
Normalization
Residual Connection
Skip Connections
Convolution
Convolutions
ReLU
Activation Functions
HRNet
Convolutional Neural Networks
Heatmap
Output Functions