Preprint
Aug 2026
Hear to See: Discerning Stateful Listening for Audio-Visual Instance Segmentation
H2S addresses challenges of audio-visual instance segmentation through the Acoustic-Semantic Projector and the Asynchronous Dynamics Modulator, prioritizing current information during dynamic variations and maintaining continuity in stable periods.
Leiye Liu, Miao Zhang, Jia-Hong Jiang et al.
· 0 citations