Convolutional Point Transformer

We present CpT: Convolutional point Transformer – a novel neural network layer for dealing with the unstructured nature of 3D point cloud data. CpT is an improvement over existing MLP and convolution layers for point cloud processing, as well as existing 3D point cloud processing transformer layers. It achieves this feat due to its effectiveness in creating a novel and robust attention-based point set embedding through a convolutional projection layer crafted for processing dynamically local point set neighbourhoods. The resultant point set embedding is robust to the permutations of the input points. Our novel layer builds over local neighbourhoods of points obtained via a dynamic graph computation at each layer of the network’s structure. It is fully differentiable and can be stacked just like convolutional layers to learn intrinsic properties of the points. Further, we propose a novel Adaptive Global Feature layer that learns to aggregate features from different representations into a better global representation of the point cloud. We evaluate our models on standard benchmark ModelNet40 classification and ShapeNet part segmentation datasets to show that our layer can serve as an effective addition for various point cloud processing tasks while effortlessly integrating into existing point cloud processing architectures to provide significant performance boosts.