Scene-Level Supervision

This research presents a Multimodal Interlaced Transformer (MIT) for weakly supervised point cloud segmentation that jointly considers 2D and 3D data. Current methods require extra 2D annotations to achieve 2D-3D…