Heterogeneous-Modal Unsupervised Domain Adaptation via Latent Space Bridging
The Latent Space Bridging (LSB) method addresses knowledge transfer between different data types (modalities), such as 2D images and 3D point clouds, using an unlabeled 'bridge domain.' LSB employs a dual-branch network and two specific losses—feature consistency and domain alignment—to map features into a shared space. This allows the model to accurately segment objects in the target modality using labeled source data.