What have we learned from deep representations for action recognition?

TL;DR

This paper investigates deep spatiotemporal representations in action recognition models, revealing how local detectors, fusion, and hierarchy contribute to understanding actions, and using visualizations to interpret learned features and training data quirks.

Contribution

It provides insights into the internal representations of two-stream models for action recognition, highlighting the roles of local detectors, fusion, and feature hierarchy, and demonstrates visualization as a tool for interpretation.

Findings

Cross-stream fusion learns true spatiotemporal features.

Networks develop class-specific and generic local representations.

Features become more abstract and invariant through hierarchy.

Abstract

As the success of deep models has led to their deployment in all areas of computer vision, it is increasingly important to understand how these representations work and what they are capturing. In this paper, we shed light on deep spatiotemporal representations by visualizing what two-stream models have learned in order to recognize actions in video. We show that local detectors for appearance and motion objects arise to form distributed representations for recognizing human actions. Key observations include the following. First, cross-stream fusion enables the learning of true spatiotemporal features rather than simply separate appearance and motion features. Second, the networks can learn local representations that are highly class specific, but also generic representations that can serve a range of classes. Third, throughout the hierarchy of the network, features become more abstract and show increasing invariance to aspects of the data that are unimportant to desired distinctions (e.g. motion patterns across various speeds). Fourth, visualizations can be used not only to shed light on learned representations, but also to reveal idiosyncracies of training data and to explain failure cases of the system.