Reproduce “MSAF: Multimodal Split Attention Fusion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lang Su, Chuqing Hu, Guofa Li, Dongpu Cao
The authors propose MSAF, a lightweight multimodal fusion method that lets audio and video features inform which parts of each stream matter. It splits each modality into channel blocks, uses shared multimodal context to weight them, then returns them to their streams. On RAVDESS, MSAF reaches 74.86% accuracy versus MMTM’s 73.12%, with 19% fewer parameters, across six actor-based folds.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.