HumanGPS: Geodesic PreServing Feature for Dense Human Correspondences

by   Feitong Tan, et al.

In this paper, we address the problem of building dense correspondences between human images under arbitrary camera viewpoints and body poses. Prior art either assumes small motion between frames or relies on local descriptors, which cannot handle large motion or visually ambiguous body parts, e.g., left vs. right hand. In contrast, we propose a deep learning framework that maps each pixel to a feature space, where the feature distances reflect the geodesic distances among pixels as if they were projected onto the surface of a 3D human scan. To this end, we introduce novel loss functions to push features apart according to their geodesic distances on the surface. Without any semantic annotation, the proposed embeddings automatically learn to differentiate visually similar parts and align different subjects into an unified feature space. Extensive experiments show that the learned embeddings can produce accurate correspondences between images with remarkable generalization capabilities on both intra and inter subjects.


page 1

page 5

page 7

page 13

page 16

page 17

page 19

page 21


Dense Human Body Correspondences Using Convolutional Networks

We propose a deep learning approach for finding dense correspondences be...

Learning Body Shape and Pose from Dense Correspondences

In this paper, we address the problem of learning 3D human pose and body...

4D Human Body Correspondences from Panoramic Depth Maps

The availability of affordable 3D full body reconstruction systems has g...

BodyMap: Learning Full-Body Dense Correspondence Map

Dense correspondence between humans carries powerful semantic informatio...

UltraPose: Synthesizing Dense Pose with 1 Billion Points by Human-body Decoupling 3D Model

Recovering dense human poses from images plays a critical role in establ...

Universal Correspondence Network

We present a deep learning framework for accurate visual correspondences...

Body sway responses to pseudorandom support surface translations of vestibular loss subjects resemble those of vestibular able subjects

Body sway responses evoked by a horizontal acceleration of a level and f...

Code Repositories