Signing Outside the Studio: Benchmarking Background Robustness for Continuous Sign Language Recognition

TL;DR

This paper evaluates the robustness of continuous sign language recognition models against background variations, introduces a new diverse background dataset, and proposes training techniques to improve model generalization in real-world scenarios.

Contribution

It creates a benchmark dataset with diverse backgrounds and proposes training methods to enhance CSLR model robustness to background shifts.

Findings

State-of-the-art CSLR models perform poorly on varied backgrounds.

Background randomization and feature disentanglement improve generalization.

The proposed methods achieve better accuracy on the new diverse background dataset.

Abstract

The goal of this work is background-robust continuous sign language recognition. Most existing Continuous Sign Language Recognition (CSLR) benchmarks have fixed backgrounds and are filmed in studios with a static monochromatic background. However, signing is not limited only to studios in the real world. In order to analyze the robustness of CSLR models under background shifts, we first evaluate existing state-of-the-art CSLR models on diverse backgrounds. To synthesize the sign videos with a variety of backgrounds, we propose a pipeline to automatically generate a benchmark dataset utilizing existing CSLR benchmarks. Our newly constructed benchmark dataset consists of diverse scenes to simulate a real-world environment. We observe even the most recent CSLR method cannot recognize glosses well on our new dataset with changed backgrounds. In this regard, we also propose a simple yet effective training scheme including (1) background randomization and (2) feature disentanglement for CSLR models. The experimental results on our dataset demonstrate that our method generalizes well to other unseen background data with minimal additional training images.