Download WAYMO dataset and organize it as follows:

└── WAYMO_DATASET_PATH 
       ├── tfrecord_training       
       ├── tfrecord_validation   
       ├── tfrecord_testing 

Remember to change the path in start.sh to the WAYMO_DATASET_PATH path above.

export PYTHONPATH="${PYTHONPATH}:/home/trainer/fmf/FMF"
export PYTHONPATH="${PYTHONPATH}:/home/trainer/fmf/nuscenes-devkit/python-sdk"

Prepare the dataset

Prepare train set

CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path '/home/trainer/fmf/FMF/data/Waymo/training/*.tfrecord'  --root_path '/home/trainer/fmf/FMF/data/Waymo/train/'

Prepare validation set

CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path '/home/trainer/fmf/FMF/data/Waymo/tfrecord_validation/*.tfrecord'  --root_path '/home/trainer/fmf/FMF/data/Waymo/val/'

Prepare testing set

CUDA_VISIBLE_DEVICES=-1 python det3d/datasets/waymo/waymo_converter.py --record_path '/home/trainer/fmf/FMF/data/Waymo/tfrecord_testing/*.tfrecord'  --root_path '/home/trainer/fmf/FMF/data/Waymo/test/'

Create 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

Training on WAYMO dataset:

For 1st stage distributed training use:

python -m torch.distributed.launch --nproc_per_node=2 ./tools/train.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/waymo_fmf_voxelnet_3x.py --work_dir waymo_exp/FMF-VoxelNet-Base --resume_from waymo_exp/FMF-VoxelNet-Base/latest.pth

For 2nd stage distributed training use:

python -m torch.distributed.launch --nproc_per_node=2 ./tools/train.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/two_stage/waymo_fmf_voxelnet_two_stage_bev_5point_ft_6epoch_freeze.py --work_dir waymo_exp/FMF-VoxelNet-Base-2nd-Stage

For single device training use:

CUDA_VISIBLE_DEVICES=1 python tools/train.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/waymo_fmf_voxelnet_3x.py --work_dir waymo_exp/FMF-VoxelNet-Base --resume_from waymo_exp/FMF-VoxelNet-Base/latest.pth

Validation on WAYMO dataset:

python tools/dist_test.py /home/trainer/fmf/FMF/configs/waymo/pp/waymo_fmf_pp_two_pfn_stride1_3x.py --work_dir waymo_exp/FMF-PointPillars-Base --checkpoint waymo_exp/FMF-PointPillars-Base/epoch_9.pth  --speed_test --gpus 1

python tools/dist_test.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/two_stage/waymo_fmf_voxelnet_two_stage_bev_5point_ft_6epoch_freeze.py --work_dir waymo_exp/FMF-VoxelNet-Base-2nd-Stage --checkpoint waymo_exp/FMF-VoxelNet-Base-2nd-Stage/epoch_6.pth  --speed_test --gpus 1

Testing on WAYMO dataset:

python tools/dist_test.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/waymo_centerpoint_voxelnet_3x_no_neck_1sweep.py --work_dir waymo_exp/CP-VoxelNet-No-Neck-1Sweep --checkpoint waymo_exp/CP-VoxelNet-No-Neck-1Sweep/epoch_36.pth  --speed_test --testset --gpus 1

Tracking

export CUDA_VISIBLE_DEVICES='0,1,2,3' # specify the GPU devices numbers for training
export num_gpus=4 # the number of used GPU devices
python -m torch.distributed.launch --nproc_per_node=4 ./tools/train.py /home/trainer/fmf/FMF/configs/waymo/pp/waymo_centerpoint_pp_two_pfn_stride1_3x_no_neck.py --work_dir waymo_exp/CP-PP-3Sweeps 

export CUDA_VISIBLE_DEVICES='0,1,2,3' # specify the GPU devices numbers for training
export num_gpus=4 # the number of used GPU devices
python -m torch.distributed.launch --nproc_per_node=4 ./tools/train.py /home/trainer/fmf/FMF/configs/waymo/voxelnet/two_stage/waymo_fmf_voxelnet_two_stage_bev_5point_ft_6epoch_freeze_no_neck.py --work_dir waymo_exp/CP-VoxelNet-3Sweeps-2nd-stage