The implement code of PointSeg,PointSeg: Real-Time Semantic Segmentation Based on 3D LiDAR Point Cloud if you think this work is useful for your research, pleace cite

@article{wang2018pointseg,
  title={PointSeg: Real-Time Semantic Segmentation Based on 3D LiDAR Point Cloud},
  author={Wang, Yuan and Shi, Tianyue and Yun, Peng and Tai, Lei and Liu, Ming},
  journal={arXiv preprint arXiv:1807.06288},
  year={2018}
}

Training/Validation

cd data/
Download data from the link  https://hkustconnect-my.sharepoint.com/:u:/g/personal/ywangeq_connect_ust_hk/EZoRWzAwcc1Ak6n4gnnHuswBnn-EP5rUzXwajzaU_8zHNw?e=UZ26cQ
tar -xzvf lidar_2d.tgz
rm lidar_2d.tgz

Also, in the src/test_demo.py .It has shown the tranformation process. 
If you are trying to train your own data. You can get some useful code from it.
I still manage this generation code to release which can be used easily.

Train

./train.sh -gpu 0 -image_set train -log_dir ./log/

eval

./eval.sh -gpu 0 -image_set val -log_dir ./log/

requirement

easydict==1.6
joblib==0.10.3
numpy==1.12.0
Pillow==4.0.0
tensorflow-gpu>1.5

Demo

This part is prepared for application on the velodyne with ros inference (test_demo.py) and visualization of results in rviz(demo.py)

Additional requirement

ros-kinetic (ubuntu 16.04)
python-pcl (1.7)

Run the demo

• test_demo This file is used for real demo in ros and has written the projection in the file. To run this demo, you need to download the dataset from KITTI. And transfer it to the rosbag by yourself. If you have the velodyne, you can just pass this way and change the lidar_topic to the your device.
• To transfer dataset into rosbag, you can check the https://github.com/tomas789/kitti2bag and follow the steps directly. (i am sorry i can not upload this for too big size)

cd src
python test_demo.py
rosbag play <data>.bag  ---- the bag which you transferrd from Kitti.

lidar_topic: /kitti/velo/pointcloud  # can be changed

publish:  1 points_raw
          2 points_raw 1-3  #car person cyclist
          

• demo
  • This file is used for the visualization, and do the segmentation on the file in data/samples, so if you want to see the result in rviz, just put the related dataset which you download from link, and run the demo.

cd src
python demo.py

if you put more than one file in ''samples'' to see the results, please comment the ''while not rospy.is_shutdown():''