Semantic2D: A Semantic Dataset for 2D Lidar Semantic Segmentation

Implementation code for our paper "Semantic2D: A Semantic Dataset for 2D Lidar Semantic Segmentation". Video demos can be found at multimedia demonstrations. The Semantic2D dataset can be found and downloaded at: https://doi.org/10.5281/zenodo.13730200.

Requirements:

• Ubuntu 20.04
• ROS-Noetic
• Python 3.8
• Labelme

Installation:

Install Labelme package:

git clone https://github.com/TempleRAIL/semantic2d.git
cd  semantic2d
pip install labelme
cp manually_labeling/.labelmerc ~/.labelmerc

Semantic2D Dataset Description:

• ``scans_lidar": each *.npy file contains a 1081x1 array 2D lidar range data
• ``intensities_lidar": each *.npy file contains 1081x1 array 2D lidar intensity data
• ``line_segments": each *.npy file contains a point list of line segments in the 2D lidar data
• ``positions": each *.npy file contains a 3x1 robot position data (x, y, z)
• ``velocities": each *.npy file contains a 2x1 robot velocity data (Vx, Wz)

Semi-Automatic Labeling Usage:

• Dataset Collection: collect and save the data from the rosbag file: assume you have already created the environment map using a mapping package like amcl and collected the raw rosbag data "xx.bag"

# complie the workspace:
cd manually_labeling/semantic_data_collection_ws
catkin_make
source devel/setup.sh
# read the rosbag and collect and save the data:
roslaunch laser_line_extraction example.launch
rosbag play xx.bag
cd manually_labeling
python dataset_collection.py

• Manually Labeling: use the Labelme tool to manually label the environment map and save the *.json file, as shown in

Then, export the labeled map images:

labelme_export_json *.json -o labelme_output

You will then get an environment map image and its semantically labeled map image, as shown in labelme_output.

• Automatic Labeling: Please modify the following configuration in the semi_automated_labeling_framework.py according to your robot and environment configuration:

# dataset: 
DATASET_ODIR = "/home/xzt/data/semantic_lidar_v2/2024-04-04-12-16-41"  # the directory path of the raw data
DATASET_NAME = "train" # select the train, dev, and test 
# map: parameters from the map configuration file
MAP_ORIGIN = np.array([-21.200000, -34.800000, 0.000000]) 
MAP_RESOLUTION = 0.025000
# labeling:
MAP_LABEL_PATH = '../manually_labeling/labelme_output/label.png'
MAP_PATH = '../manually_labeling/labelme_output/img.png'
# lidar sensor:
POINTS = 1081 # the number of lidar points
LIDAR_BASE_DIS = -0.12 # the distance from lidar_mount to base_link

Start the automatic labeling:

cd ../automatic_labeling
python semi_automated_labeling_framework.py

Plot the labeled semantic lidar data: you also need to modify the similar configuration as semi_automated_labeling_framework.py in the draw_semantic_label_sample.py

python draw_semantic_label_sample.py

Citation

@article{xie2024semantic2d,
  title={Semantic2D: A Semantic Dataset for 2D Lidar Semantic Segmentation},
  author={Xie, Zhanteng and Dames, Philip},
  journal={arXiv preprint arXiv:2409.09899},
  year={2024}
}