Multi-View Action Recognition via Directed Gromov-Wasserstein Discrepancy

This work proposes a multi-view attention consistency method using directed Gromov-Wasserstein discrepancy to ensure that the action recognition model focuses on the proper action subject across different camera views.