ViSAGE wins NTIRE 2026 video saliency prediction challenge with multi-expert ensemble

Researchers unveiled ViSAGE, a multi-expert framework that won the NTIRE 2026 Challenge on Video Saliency Prediction at CVPR 2026. The model uses adaptive gating across specialized decoders to capture complex spatio-temporal patterns in videos, ranking first on half the evaluation metrics and outperforming competitors on others.