Understanding Pixel-level 2D Image Semantics with 3D Keypoint Knowledge Engine - 42Papers