Tinysign: sign language recognition in low resolution settings

Existing sign language recognition (SLR) methods mostly rely on high-quality videos that include clear hand and body movements. However, these approaches often fall short in addressing the challenges of real-world sign language communication scenarios. In this work, we tackle the problem of SLR in low resolution settings. To this end, we propose a novel approach by effectively encoding the spatial and temporal features. Our approach includes a sign-specific super-resolution module that improves discriminability between classes and a sign classifier that learns high-level spatio-temporal features. In order to evaluate the effectiveness of our approach, we introduce the first low-resolution sign language recognition benchmark and present the experimental results together with detailed analysis.