Computer Graphics

We develop a deep neural network that automatically extracts contextual features from patches in sketches, trained with 3D models rendered with non-photorealistic techniques. Our method is able to find dense correspondences between real world sketches!