Report
StreamGaze benchmark accepted to NeurIPS 2026 E&D track
A developer calls it the first benchmark for evaluating how multimodal AI models use human gaze for temporal reasoning in streaming first-person video.
TLDR
A developer announced that StreamGaze, developed in collaboration with Adobe Research, was accepted to the NeurIPS 2026 E&D track. They describe it as the first benchmark for evaluating how well multimodal large language models use human gaze signals for temporal reasoning in streaming first-person videos.
Combined views
4
1 Source, first seen ago