Point Cloud Segmentation using PointNet

This repo contains the implementation of the PointNet model for Semantic Segmentation of LiDAR scans on the KITTI dataset.

Contents

• Requirements
• Demo
• Dataset

Requirements

pip install opencv-python numpy pandas torch torchvision h5py redis matplotlib pyyaml open3d tqdm 

Demo

• Set the path of the environment variable KITTI_ROOT to the path of your dataset

python viewpoints.py

Dataset

• Dataset used for this project is the KITTI Dataset with KITTI Odometry Benchmark Velodyne Point Clouds, Calibration data, Color RGB Dataset and SemanticKITTI label data.
• Ensure the file structure is similar to this:

.
└── Dataset/
    └── Sequences/
        ├── 00/
        │   ├── Velodyne/
        │   │   ├── 000000.bin
        │   │   └── .
        │   ├── labels/
        │   │   ├── 000000.label
        │   │   └── .
        │   ├── image_2/
        │   │   ├── 000000.png
        │   │   └── .
        │   ├── image_3 /
        │   │   └── 000000.png
        │   ├── calib.txt
        │   ├── poses.txt
        │   └── time.txt
        ├── 01
        ├── .
        ├── .
        ├── .
        └── 21