Discovering Human Interactions in Videos with Limited Data Labeling

TL;DR

This paper introduces an unsupervised method for discovering human interactions in videos using a maximum margin clustering approach that incorporates user feedback, achieving high-quality semantic clusters with minimal labeled data.

Contribution

It proposes a novel iterative maximum margin clustering framework that integrates user feedback for unsupervised discovery of human interactions in videos.

Findings

Successfully clusters human interactions with limited labeled data

Effective on multiple challenging datasets

Outperforms existing unsupervised methods

Abstract

We present a novel approach for discovering human interactions in videos. Activity understanding techniques usually require a large number of labeled examples, which are not available in many practical cases. Here, we focus on recovering semantically meaningful clusters of human-human and human-object interaction in an unsupervised fashion. A new iterative solution is introduced based on Maximum Margin Clustering (MMC), which also accepts user feedback to refine clusters. This is achieved by formulating the whole process as a unified constrained latent max-margin clustering problem. Extensive experiments have been carried out over three challenging datasets, Collective Activity, VIRAT, and UT-interaction. Empirical results demonstrate that the proposed algorithm can efficiently discover perfect semantic clusters of human interactions with only a small amount of labeling effort.