Abstract: In Continuous Sign Language Recognition (CSLR), frame-wise features are often affected by occlusion and motion blur, leading to unstable predictions. To address this issue, we propose ...