Understanding Human Gaze Communication by Spatio-Temporal Graph Reasoning

TL;DR

This paper introduces VACATION, a large-scale dataset and a novel spatio-temporal graph neural network to understand and predict human gaze communication in social videos at both atomic and event levels.

Contribution

It provides the first comprehensive dataset for gaze communication and proposes a new graph-based model for detailed analysis of gaze interactions in social scenes.

Findings

Model significantly outperforms baselines in predicting gaze communication.

VACATION dataset covers diverse social scenes with detailed annotations.

Proposed network effectively captures both atomic and event-level gaze behaviors.

Abstract

This paper addresses a new problem of understanding human gaze communication in social videos from both atomic-level and event-level, which is significant for studying human social interactions. To tackle this novel and challenging problem, we contribute a large-scale video dataset, VACATION, which covers diverse daily social scenes and gaze communication behaviors with complete annotations of objects and human faces, human attention, and communication structures and labels in both atomic-level and event-level. Together with VACATION, we propose a spatio-temporal graph neural network to explicitly represent the diverse gaze interactions in the social scenes and to infer atomic-level gaze communication by message passing. We further propose an event network with encoder-decoder structure to predict the event-level gaze communication. Our experiments demonstrate that the proposed model improves various baselines significantly in predicting the atomic-level and event-level gaze