One of the works spotlighted is Mesh R-CNN, a method that’s able to predict three-dimensional shapes from images of cluttered and occluded objects. Above: Mesh R-CNN converts 2D objects to 3D ...