Skip to content

Latest commit

 

History

History
107 lines (76 loc) · 3.82 KB

WAYMO.md

File metadata and controls

107 lines (76 loc) · 3.82 KB
Raw

Getting Started with CenterPoint on Waymo

PointPillars model

We now support two models using PointPillars architecture.

Split mAPH@L1 mAPH@L2
PointPillars-512 Val 71.2 62.6
CenterPoint-Pillar-512 Val 72.7 64.2

We have pretrained models available for download which can be requested by filling out this form.

Prerequisite

  • Follow INSTALL.md to install all required libraries. spconv is optional
  • 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

Convert the tfrecord data to pickle files.

# train set 
python det3d/datasets/waymo/waymo_converter.py --input_file_pattern='WAYMO_DATASET_ROOT/tfrecord_training/segment-*.tfrecord'  --output_filebase='WAYMO_DATASET_ROOT/train/'

# validation set 
python det3d/datasets/waymo/waymo_converter.py --input_file_pattern='WAYMO_DATASET_ROOT/tfrecord_validation/segment-*.tfrecord'  --output_filebase='WAYMO_DATASET_ROOT/val/'

Create a symlink to the dataset root

mkdir data && cd data
ln -s WAYMO_DATASET_ROOT 
mv WAYMO_DATASET_ROOT Waymo # rename to Waymo

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

Create info files

# train set 
python tools/create_data.py waymo_data_prep --root_path=data/Waymo --split train --nsweeps=1

# val set 
python tools/create_data.py waymo_data_prep --root_path=data/Waymo --split val --nsweeps=1

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

└── CenterPoint
       └── data    
              └── Waymo 
                     ├── tfrecord_training       
                     ├── tfrecord_validation
                     ├── train <-- all training frames and annotations 
                     ├── val   <-- all validation frames and annotations 
                     ├── infos_train_01sweeps_filter_zero_gt.pkl
                     ├── infos_val_01sweeps_filter_zero_gt.pkl

Train & Evaluate in Command Line

Now we only support training and evaluation with gpu. Cpu only mode is not supported.

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  --gt