Author: Sreenivasa Hikkal Venugopala
Contact: hvsreenivasa93@gmail.com
-
Clone the repository Scripts.
-
Requirements: Install following libraries
- Tensorflow 1.15.0
- waymo-open-dataset
- OpenCV-python
-
Download the Waymo open dataset and extract the content into a folder.
-
Run the following command - "python Waymo_to_kitti.py --source=path/to/tfrecord_data_folder --dest=path/to/extract/data --all"
- Provided with more command line options to generate
"--velo" - velodyne lidar points with option with camera calibration and labels "--img" - camera images with option with camera calibration and labels "--all" - to generate lidar, images, camera calibration and labels. "--oclu" - to generate lidar, images, camera calibration and labels with basic occlusion information. -
Output folders will have following sructure under destination folder.
.
Output
├── Calibration
│ └── Calib
│ ├── 0
│ ├── 1
│ ├── 2
│ ├── 3
│ └── 4
│ └── Calib_all
├── Camera
│ ├── Front
│ ├── Front_left
│ ├── Front_right
│ ├── Side_left
│ └── Side_right
├── Labels
│ └── Label
│ ├── 0
│ ├── 1
│ ├── 2
│ ├── 3
│ └── 4
│ └── Labels_all
└── Velodyne
Waymo has 5 cameras - Front, Front_left, Front_right, Side_left, Side_right. Each is assigned with values from 0 to 4.
Under construction
Consists of two sub folders.
- Labels - consists of labels for each individual cameras and folder names denotes what camera the label belongs. The number is mentiond in camera section.
- Labels_all - consists of all label in single file.
Labels in kitti format with basic occlusion information as follows:
- occlusion level 1 - for occluded objects.
- occlusion level 0 - for non occluded objects.
All in vehicle frame.
Consists of two sub folders.
- Calib - consists of calibration for each individual cameras and folder names denotes what camera the label belongs. The number is mentiond in camera section.
- Calib_all - consists of all calibrations in single file.
P0-P4 : intrinsic matrix for each camera
R0_rect : rectify matrix
Tr_velo_to_cam_0 - Tr_velo_to_cam_4 : transformation matrix from vehicle frame to camera frame
Point cloud in vehicle frame.
x y z reflectance