arXiv:2611.05633v1 Announce Type: cross Abstract: Many temporal approaches for video recognition rely on explicit cross-channel aggregation projections, which have quadratic complexity in the number of channels, introducing significant computational overhead.
