Learning Joint Top-Down and Bottom-up Processes for 3D Visual Inference | IEEE Conference Publication | IEEE Xplore