3D point cloud instance segmentation

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

Tools used: Python, PyTorch, PyTorch3D, Open3D

Proposed a U-Net based deep learning architecture for instance segmentation of 3D point clouds, we devised an encoder that was capable of learning the global features of the point cloud.

Proposed block diagram

Result on S3DIS dataset