Skip to content

Latest commit

 

History

History
executable file
·
141 lines (95 loc) · 5.56 KB

WAYMO.md

File metadata and controls

executable file
·
141 lines (95 loc) · 5.56 KB
Raw

Getting Started with CenterPoint on Waymo

Prerequisite

  • Follow INSTALL.md to install all required libraries.
  • Tensorflow
  • Waymo-open-dataset devkit
conda activate centerpoint 
pip install waymo-open-dataset-tf-1-15-0==1.2.0

Prepare data

Download data and organise as follows

# For Waymo Dataset         
└── WAYMO_DATASET_ROOT
       ├── tfrecord_training       
       ├── tfrecord_validation   
       ├── tfrecord_testing

Convert the tfrecord data to pickle files.

# train set 
CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path 'WAYMO_DATASET_ROOT/tfrecord_training/*.tfrecord'  --root_path 'WAYMO_DATASET_ROOT/train/'

# validation set 
CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path 'WAYMO_DATASET_ROOT/tfrecord_validation/*.tfrecord'  --root_path 'WAYMO_DATASET_ROOT/val/'

# testing set 
CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path 'WAYMO_DATASET_ROOT/tfrecord_testing/*.tfrecord'  --root_path 'WAYMO_DATASET_ROOT/test/'

Create a symlink to the dataset root

mkdir data && cd data
ln -s WAYMO_DATASET_ROOT Waymo

Remember to change the WAYMO_DATASET_ROOT to the actual path in your system.

Dense Object Set

When you have downloaded the dense object set, you should unzip it and put it under the train folder.

Create info files

Note that we modified the waymo_common.py and added name of objects into the info files.

# One Sweep Infos 
python tools/create_data.py waymo_data_prep --root_path=data/waymo --split train --nsweeps=1

python tools/create_data.py waymo_data_prep --root_path=data/waymo --split val --nsweeps=1

python tools/create_data.py waymo_data_prep --root_path=data/waymo --split test --nsweeps=1

# Two Sweep Infos (for two sweep detection and tracking models)
python tools/create_data.py waymo_data_prep --root_path=data/waymo --split train --nsweeps=2

python tools/create_data.py waymo_data_prep --root_path=data/waymo --split val --nsweeps=2

python tools/create_data.py waymo_data_prep --root_path=data/waymo --split test --nsweeps=2

In the end, the data and info files should be organized as follows

└── Sparse2Dense
       └── data    
              └── Waymo 
                     ├── tfrecord_training       
                     ├── tfrecord_validation
                     ├── train <-- all training frames and annotations 
                            ├── lidar
                            ├── annos
                            ├── gt <-- dense object set
                     ├── val   <-- all validation frames and annotations 
                     ├── test   <-- all testing frames and annotations 
                     ├── infos_train_01sweeps_filter_zero_gt.pkl
                     ├── infos_train_02sweeps_filter_zero_gt.pkl
                     ├── infos_val_01sweeps_filter_zero_gt.pkl
                     ├── infos_val_02sweeps_filter_zero_gt.pkl
                     ├── infos_test_01sweeps_filter_zero_gt.pkl
                     ├── infos_test_02sweeps_filter_zero_gt.pkl

Train & Evaluate in Command Line

Use the following command to start a distributed training using 4 GPUs. The models and logs will be saved to work_dirs/CONFIG_NAME.

python -m torch.distributed.launch --nproc_per_node=4 ./tools/train.py CONFIG_PATH

For distributed testing with 4 gpus,

python -m torch.distributed.launch --nproc_per_node=4 ./tools/dist_test.py CONFIG_PATH --work_dir work_dirs/CONFIG_NAME --checkpoint work_dirs/CONFIG_NAME/latest.pth

For testing with one gpu and see the inference time,

python ./tools/dist_test.py CONFIG_PATH --work_dir work_dirs/CONFIG_NAME --checkpoint work_dirs/CONFIG_NAME/latest.pth --speed_test

This will generate a my_preds.bin file in the work_dir. You can create submission to Waymo server using waymo-open-dataset code by following the instructions here.

If you want to do local evaluation (e.g. for a subset), generate the gt prediction bin files using the script below and follow the waymo instructions here.

python det3d/datasets/waymo/waymo_common.py --info_path data/waymo/infos_val_01sweeps_filter_zero_gt.pkl --result_path data/waymo/ --gt

All pretrained models and configurations are in MODEL ZOO.

Second-stage Training

Our final model follows a two-stage training process. For example, to train the two-stage CenterPoint-Voxel model, you first need to train the one stage model using ONE_STAGE and then train the second stage module using TWO_STAGE. You can also contact us to access the pretrained models, see details here.

Visualization

Please refer to visual.py. It will take a prediction file generated by simple_inference_waymo.py and visualize the point cloud and detections.

Test Set

Add the --testset flag to the end.

python ./tools/dist_test.py CONFIG_PATH --work_dir work_dirs/CONFIG_NAME --checkpoint work_dirs/CONFIG_NAME/latest.pth --testset