Learning Multi-Human Optical Flow

TL;DR

This paper introduces a new dataset and training approach for multi-human optical flow, improving accuracy and generalization in human motion analysis using deep networks.

Contribution

The authors develop a synthetic dataset of multi-human optical flow and train deep networks specifically on this data, addressing the domain gap in existing methods.

Findings

Trained networks outperform existing top methods on test data.

Models generalize well to real image sequences.

Provides publicly available code, models, and dataset.

Abstract

The optical flow of humans is well known to be useful for the analysis of human action. Recent optical flow methods focus on training deep networks to approach the problem. However, the training data used by them does not cover the domain of human motion. Therefore, we develop a dataset of multi-human optical flow and train optical flow networks on this dataset. We use a 3D model of the human body and motion capture data to synthesize realistic flow fields in both single- and multi-person images. We then train optical flow networks to estimate human flow fields from pairs of images. We demonstrate that our trained networks are more accurate than a wide range of top methods on held-out test data and that they can generalize well to real image sequences. The code, trained models and the dataset are available for research.