Multi-Modal Domain Fusion for Multi-Modal Aerial View Object Classification
 | Our solution to the CVPR Perception Beyond Visible Spectrum Workshop 2022: Multi-Modal Aerial View Object Classification challenge. We proposed a novel Multi-Modal Domain Fusion(MDF) network and a training scheme to learn the domain invariant features from long-tailed multi-modal data distribution by reducing the domain discrepancy using Wasserstein distance and use it to accurately classify the aerially viewed objects. To make the learning more efficient, class-conditional domain adaptation was achieved by passing the multi-modal data in pairs at a time to the two models in MDF, effectively making it a shared feature learner. Sumanth V Udupa, Aniruddh Sikdar, Suresh Sundaram paper code |
|---|
