Point Cloud Oversegmentation with Graph-Structured Deep Metric Learning
This addresses the problem of efficient and accurate 3D point cloud segmentation for applications like robotics and autonomous driving, representing a significant advance over prior methods.
The paper tackles point cloud oversegmentation by learning deep embeddings for local geometry and radiometry, formulating it as a graph partition problem, resulting in a new state-of-the-art with over five times fewer superpoints needed for similar performance on S3DIS and improvements in semantic segmentation.
We propose a new supervized learning framework for oversegmenting 3D point clouds into superpoints. We cast this problem as learning deep embeddings of the local geometry and radiometry of 3D points, such that the border of objects presents high contrasts. The embeddings are computed using a lightweight neural network operating on the points' local neighborhood. Finally, we formulate point cloud oversegmentation as a graph partition problem with respect to the learned embeddings. This new approach allows us to set a new state-of-the-art in point cloud oversegmentation by a significant margin, on a dense indoor dataset (S3DIS) and a sparse outdoor one (vKITTI). Our best solution requires over five times fewer superpoints to reach similar performance than previously published methods on S3DIS. Furthermore, we show that our framework can be used to improve superpoint-based semantic segmentation algorithms, setting a new state-of-the-art for this task as well.