most citedA cross-modal fusion network based on self-attention and residual structure for multimodal emotion recognition

28 citations · 33 across the 4 of their papers we have counts for

collaborators

4 papers