MultiScan: Scalable RGBD scanning for 3D environments with articulated objects

Figure 1 : MultiScan offers 3D reconstructions of indoor scenes with multiple scans, capturing objects in multiple states, and with multiple levels of annotations. For each scene, we provide multiple scans (leftmost column) with objects in varied states (second column). Each scan is densely annotated with semantic object instances (third column) that are corresponded across scans. Each object is annotated with parts and part articulation parameters (fourth column). Different scans capture objects in different articulation states (e.g., the cabinet door is opened in the bottom row). The MultiScan pipeline and dataset enable investigation of holistic 3D scene understanding with articulated objects by supporting tasks that reason at multiple levels of a 3D scene (rightmost column).