Towards Zero-shot Sign Language Recognition

TL;DR

This paper introduces a zero-shot sign language recognition framework utilizing textual descriptions and attributes, along with new datasets, to recognize unseen signs by transferring knowledge from seen classes.

Contribution

It proposes a novel zero-shot learning approach for sign language recognition using semantic descriptions and attributes, supported by new benchmark datasets.

Findings

Textual and attribute embeddings improve recognition of unseen signs.

The approach effectively transfers knowledge from seen to unseen sign classes.

Analysis of attribute influence enhances understanding of zero-shot predictions.

Abstract

This paper tackles the problem of zero-shot sign language recognition (ZSSLR), where the goal is to leverage models learned over the seen sign classes to recognize the instances of unseen sign classes. In this context, readily available textual sign descriptions and attributes collected from sign language dictionaries are utilized as semantic class representations for knowledge transfer. For this novel problem setup, we introduce three benchmark datasets with their accompanying textual and attribute descriptions to analyze the problem in detail. Our proposed approach builds spatiotemporal models of body and hand regions. By leveraging the descriptive text and attribute embeddings along with these visual representations within a zero-shot learning framework, we show that textual and attribute based class definitions can provide effective knowledge for the recognition of previously unseen sign classes. We additionally introduce techniques to analyze the influence of binary attributes in correct and incorrect zero-shot predictions. We anticipate that the introduced approaches and the accompanying datasets will provide a basis for further exploration of zero-shot learning in sign language recognition.