Reasoning

DAVE

DAVE tests whether audio-visual models can link an overlaid sound to the corresponding event in an egocentric video. This import covers the released multimodal judgments of eight end-to-end models, with complete question text and original video/audio assets.