Action2Vec: A Crossmodal Embedding Approach to Action Learning

TL;DR

Action2Vec introduces a cross-modal embedding space combining video features and linguistic cues, enabling zero-shot action recognition and semantic analogy tests, advancing action understanding in videos.

Contribution

It presents the first joint embedding of verbs and action videos, integrating hierarchical recurrent networks with semantic loss for improved action recognition.

Findings

Achieved state-of-the-art zero-shot action recognition results.

Developed novel analogy tests for semantic evaluation.

First to evaluate distributional semantics in action-video embeddings.

Abstract

We describe a novel cross-modal embedding space for actions, named Action2Vec, which combines linguistic cues from class labels with spatio-temporal features derived from video clips. Our approach uses a hierarchical recurrent network to capture the temporal structure of video features. We train our embedding using a joint loss that combines classification accuracy with similarity to Word2Vec semantics. We evaluate Action2Vec by performing zero shot action recognition and obtain state of the art results on three standard datasets. In addition, we present two novel analogy tests which quantify the extent to which our joint embedding captures distributional semantics. This is the first joint embedding space to combine verbs and action videos, and the first to be thoroughly evaluated with respect to its distributional semantics.