WaymoPreprocessing

Preprocessing script for WaymoDynamic Dataset

We provide a preprocess script for the waymo v2 perception dataset for dynamic neural lidar simulations

To obtain the dataset, you need to first

1. Download the v2 perception dataset from waymo open dataset

make sure it has following structure, if you have one context with the name of 10061305430875486848_1080_000_1100_000

waymo
└── training
    ├── lidar
    │   └── 10061305430875486848_1080_000_1100_000.parquet
    ├── lidar_box
    │   └── 10061305430875486848_1080_000_1100_000.parquet
    ├── lidar_calibration
    │   └── 10061305430875486848_1080_000_1100_000.parquet
    ├── lidar_pose
    │   └── 10061305430875486848_1080_000_1100_000.parquet
    └── vehicle_pose
        └── 10061305430875486848_1080_000_1100_000.parquet

you can download as many contexts as you want, our preprocess handle one context at a time

2. install requirements

We use python 3.10.10, python version under 3.9 would not work afaik

conda create -n waymov2 python=3.10.10
conda activate waymov2
python3 -m pip install gcsfs waymo-open-dataset-tf-2-11-0==1.5.1
pip install torch==1.13.1 torchvision functorch --extra-index-url https://download.pytorch.org/whl/cu117
pip install open3d==0.16.0
pip install numba

3. launch the preprocess script

specify the context_name as well as the dataset_dir in preprocess_waymo.py, e.g.

dataset_dir = './waymo/training' ##where you download in the first step

context_name = '10061305430875486848_1080_000_1100_000'

Then launch the script

python preprocess_waymo.py

It will generates a folder called processed_data_dynamic in your PWD, which as following structure:

processed_data_dynamic
└── 10061305430875486848_1080_000_1100_000
    ├── normals.npy
    ├── object_ids_per_frame.npy
    ├── objects_id_2_anchors.npy
    ├── objects_id_2_corners.npy
    ├── objects_id_2_dynamic_flag.npy
    ├── objects_id_2_frameidx.npy
    ├── objects_id_2_tsfm.npy
    ├── objects_id_types_per_frame.npy
    ├── range_images1.npy
    ├── range_images2.npy
    ├── ray_dirs.npy
    ├── ray_object_indices.npy
    ├── ray_origins.npy
    └── valid_normal_flags.npy

Preprocessed data

We preprocessed 5 scenes in advance if you would like to skip running the script on your end.