VGGSounder: Audio-Visual Evaluations for Foundation Models

Arxiv CS.SD (Sound/Audio ML)

Proposes VGGSounder, an evaluation methodology for audio-visual foundation models. It reveals that the VGGSound benchmark has significant labeling errors and ambiguities, affecting reliability of prior evaluations.

foundation-model

Arxiv CS.SD (Sound/Audio ML)·Daniil ZverevThadd\"aus WiedemerAmeya PrabhuMatthias BethgeWieland BrendelA. Sophia Koepke

7 days ago