3D Point-Cloud Instance Segmentation

U-Net based deep learning architecture for instance segmentation of 3D point clouds

Method: A U-Net-based architecture with an encoder for global point-cloud features, evaluated for instance segmentation on S3DIS.
Stack: Python, PyTorch, PyTorch3D, Open3D

We proposed a U-Net-based deep learning architecture for instance segmentation of 3D point clouds. We devised an encoder capable of learning the global features of the point cloud.

Proposed instance-segmentation architecture — Proposed block diagram.

Instance-segmentation result on the S3DIS dataset — Result on the S3DIS dataset.