Pose Embeddings: A Deep Architecture for Learning to Match Human Poses

TL;DR

This paper introduces a deep learning-based pose embedding method that directly compares human poses in images, bypassing joint estimation, and shows promising results in pose matching and retrieval tasks.

Contribution

It proposes a novel deep architecture for learning pose embeddings using a triplet-based criterion, improving pose comparison without joint detection.

Findings

Effective pose matching and retrieval demonstrated on video data

Embedding captures pose similarities accurately

Avoids challenges of joint position estimation

Abstract

We present a method for learning an embedding that places images of humans in similar poses nearby. This embedding can be used as a direct method of comparing images based on human pose, avoiding potential challenges of estimating body joint positions. Pose embedding learning is formulated under a triplet-based distance criterion. A deep architecture is used to allow learning of a representation capable of making distinctions between different poses. Experiments on human pose matching and retrieval from video data demonstrate the potential of the method.