Zhile Ren | Research Statement

Figure 1: COG descriptor encodes orientation-invariant gradient feature for objects with different views. I develop new representations and algorithms for three-dimensional (3D) scene understanding from cluttered indoor RGB-D images and outdoor video sequences. I introduce novel representations for 3D object detection systems that localize objects with cuboids and describe room layouts by Manhattan structures. Using view-invariant 3D features, I capture 3D style-variations and design systems to detect small objects by modeling support surfaces. Finally, I develop cascaded prediction frameworks to model 3D contextual relationships and enable rapid understanding of scene properties including depth, motion, and segmentation.