Cross-View Isolated Sign Language Recognition Challenge: Design, Results and Future Research

The performance of Isolated Sign Language Recognition (ISLR) systems has improved considerably in recent years. However, existing ISLR models often struggle to generalize across different viewpoints, limiting their practical usability in real-world applications. To highlight this limitation, we introduce the Cross-View ISLR Challenge (CV-ISLR), a workshop focused on improving cross-view generalization in ISLR task. Our challenge fosters research on robust ISLR models capable of recognizing signs from various viewpoints. It also encourages the development of novel multi-modal learning techniques and helps bridge the gap between academic research and real-world deployment. We present an overview of the CV-ISLR Challenge, organized as part of the ACM Web Conference 2025 (WWW 2025). Participants are evaluated on MM-WLAuslan, a newly introduced large-scale multi-modal Australian Sign Language (Auslan) dataset. It contains 3,215 sign glosses and over 282K isolated sign video samples, performed by 73 different signers and recorded from three distinct RGB-D camera viewpoints. The challenge attracted 23 participants on the RGB track and 11 on the RGB+Depth track, leading to more than 120 submissions in total. We analyze and discuss the challenge design, top-performing solutions, and insights for future CV-ISLR research. For more details about the challenge, please visit our official website.