RoCoISLR: A Romanian Corpus for Isolated Sign Language Recognition

2025-11-16

AI Summary

The scarcity of large-scale, standardized datasets hinders progress in Romanian Isolated Sign Language Recognition (RoISLR). Method: This work introduces RoCoISLR—the first high-quality, standardized corpus for RoISLR—comprising over 9,000 video samples and nearly 6,000 lexical items. It proposes a systematic methodology for constructing low-resource sign language corpora, analyzes the critical impact of long-tailed label distribution on recognition performance, and establishes the first dedicated RoISLR benchmark. Contribution/Results: Under a unified experimental protocol, seven state-of-the-art architectures—including I3D, SlowFast, Swin Transformer, and TimeSformer—are rigorously evaluated. Results demonstrate the clear superiority of Transformer-based models over CNNs: Swin Transformer achieves 34.1% Top-1 accuracy, confirming the dataset’s substantial difficulty and practical utility. RoCoISLR effectively bridges the data gap for under-resourced sign languages and advances non-dominant sign language recognition research.

Technology Category

Intelligent Robots: Multimodal Perception & Sensor FusionMachine Learning: Large Multimodal Models (LMMs)Computer Vision: Adversarial Attacks & Robustness

Application Category

Search and Retrieval-Augmented AI: Web evaluation methodologies and metricsSecurity and Privacy: Large-scale security measurementsEconomics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Automatic sign language recognition plays a crucial role in bridging the communication gap between deaf communities and hearing individuals; however, most available datasets focus on American Sign Language. For Romanian Isolated Sign Language Recognition (RoISLR), no large-scale, standardized dataset exists, which limits research progress. In this work, we introduce a new corpus for RoISLR, named RoCoISLR, comprising over 9,000 video samples that span nearly 6,000 standardized glosses from multiple sources. We establish benchmark results by evaluating seven state-of-the-art video recognition models-I3D, SlowFast, Swin Transformer, TimeSformer, Uniformer, VideoMAE, and PoseConv3D-under consistent experimental setups, and compare their performance with that of the widely used WLASL2000 corpus. According to the results, transformer-based architectures outperform convolutional baselines; Swin Transformer achieved a Top-1 accuracy of 34.1%. Our benchmarks highlight the challenges associated with long-tail class distributions in low-resource sign languages, and RoCoISLR provides the initial foundation for systematic RoISLR research.

Problem

• Addressing the lack of large-scale Romanian sign language datasets
• Establishing benchmark results with seven video recognition models
• Highlighting challenges of long-tail class distributions in low-resource languages

Innovation

• Created Romanian sign language corpus with 9000 videos
• Benchmarked seven video recognition models consistently
• Transformer models outperformed convolutional baselines in accuracy