PointHop: An Explainable Machine Learning Method for Point Cloud Classification
This addresses the problem of efficient and interpretable point cloud classification for computer vision applications, though it is incremental as it builds on existing concepts like space partitioning and Saab transforms.
The authors tackled point cloud classification by proposing PointHop, an explainable method that uses iterative local-to-global attribute building and ensembles, achieving performance comparable to state-of-the-art methods with much lower training complexity.
An explainable machine learning method for point cloud classification, called the PointHop method, is proposed in this work. The PointHop method consists of two stages: 1) local-to-global attribute building through iterative one-hop information exchange, and 2) classification and ensembles. In the attribute building stage, we address the problem of unordered point cloud data using a space partitioning procedure and developing a robust descriptor that characterizes the relationship between a point and its one-hop neighbor in a PointHop unit. When we put multiple PointHop units in cascade, the attributes of a point will grow by taking its relationship with one-hop neighbor points into account iteratively. Furthermore, to control the rapid dimension growth of the attribute vector associated with a point, we use the Saab transform to reduce the attribute dimension in each PointHop unit. In the classification and ensemble stage, we feed the feature vector obtained from multiple PointHop units to a classifier. We explore ensemble methods to improve the classification performance furthermore. It is shown by experimental results that the PointHop method offers classification performance that is comparable with state-of-the-art methods while demanding much lower training complexity.