- Download kitti odometry (grayscale, color, velodyne, calibration, ground truth poses, development kit) from https://www.cvlibs.net/datasets/kitti/eval_odometry.php
- Download SemanticKITTI label (labels) from http://www.semantic-kitti.org/dataset.html
source data organization:
├── source folder
│ ├── data_odometry_color
│ ├── data_odometry_labels
│ ├── data_odometry_poses
│ ├── data_odometry_velodyne
| ├── data_odometry_calib
target data organization:
├── target folder (defaults as xxx/dataset/sequences)
│ ├── 00
│ │ ├── velodyne
│ │ | ├── 000000.bin
│ │ | ├── ....bin
│ | ├── labels
│ │ | ├── 000000.label
│ │ | ├── ....label
│ │ ├── image2
│ │ | ├── 000000.png
│ │ | ├── ....png
│ │ ├── image3 （default we do not use this but also store it in this folder）
│ │ | ├── 000000.png
│ │ | ├── ....png
│ ├── 01
│ ├── ...
│ ├── poses.txt
│ ├── calib.txt
│ ├── timestamps.txt
| ├── detectron_labels.csv
# >> set up detectron2 environment
conda create -n dhp python=3.9 -y
conda activate dhp
pip install torch==2.0.1 torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu118
python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
pip install natsort fire opencv-python pandas scipy open3d
# >> run the python script
python3 preprocess_kitti.py $source_folder $target_folder
# >> get lidar camera extrinsic calibration
python3 get_lidar_camera_extrinsic.py $data_root_folder
Your have finished the data process, and you can back to the main README.md to continue the next step.
The following is the optional step to check and visualize the data.
modify the launch file to set the correct data path and max_frames in the launch file
cd <workspace_dir>
catkin build panoptic_mapping_utils
source ./devel/setup.bash
# >>> using the segmentation results of detectron2 for experiments
roslaunch panoptic_mapping_utils play_kitti_image_label.launch
rviz -d $(rospack find panoptic_mapping_utils)/config/kitti_player_check.rviz
# >>> using the ground truth labels(from kitti label panoptic labels) for generate ground truth map
roslaunch panoptic_mapping_utils play_kitti_lidar_label.launch
rviz -d $(rospack find panoptic_mapping_utils)/config/kitti_player_check_lidar.rviz
# vis detectron segmentation
python3 vis_detectron_segmentation.py