RevealNet: Seeing Behind Objects in RGB-D Scans

Figure 1 : RevealNet takes an RGB-D scan as input and learns to "see behind objects": from the scan's color images and geometry (encoded as a TSDF), objects in the observed scene are detected (as 3D bounding boxes and class labels) and for each object, the complete geometry of that object is predicted as per-instance masks (in both seen and unseen regions).