data_preparation.md

Data Preparation

Datasets

Full Waymo v1.3 dataset

Download the data from the website of Waymo Open Dataset. Then prepare Waymo data in Kitti format by running

cd code
python tools/create_data.py waymo --root-path ./data/waymo --out-dir ./data/waymo --extra-tag waymo

Waymo mini dataset

We also provide a mini Waymo dataset in Kitti format which includes 1/5 Waymo v1.3 dataset. You can download the Waymo mini dataset from Google Drive or Baidu Disk, and extract the data following the folder structure. As for the extraction, you can use the commands below.

cat training.tar.gz* | tar -xzv

Pretrained Model

Download pretrained model from here.

Folder Structure

The folder structure should like this.

|── docs
|── bev_toolbox
├── waymo_playground/
|   ├── projects
│   |   ├── configs
│   |   ├── mmdet3d_plugin
|   ├── mmdetection3d
|   ├── nuscenes-devkit
|   ├── toolbox
|   ├── tools
|   ├── ckpts
│   |   ├── r101_dcn_fcos3d_pretrain.pth
|   ├── data
│   |   ├── waymo_mini
│   |   │   ├── training
│   |   │   ├── waymo_calibs.pkl
│   │   |   ├── waymo_mini_infos_train.pkl
│   │   |   ├── waymo_mini_infos_val.pkl
│   │   |   ├── gt.bin
│   │   |   ├── gt.bin.pkl
|   |   filter_waymo.txt
...