RUN_MODEL.md

Training ProposalContrast on Waymo

Training Scripts.

• Use the following command to pretrain the 3D models. The pretrain_description and finetune_description specify the task names. The models will be saved to ~/ProposalContrast/output/waymo/pretrain. We would recommand distributed training with 8*32G GPUs for better results.

# pretrain and fientune VoxelNet
bash tools/scripts/ssl_pretrain_voxel.sh pretrain_description finetune_description

# pretrain and fientune PillarNet
bash tools/scripts/ssl_pretrain_pillar.sh pretrain_description finetune_description

Note that the fine-tuning command is also contained in above scripts.

• Use the following command to obtain the baseline results that are trained from random initialization. TASK_DESC indicates the task name.

# train VoxelNet baseline
bash tools/scripts/train_voxel_sup.sh TASK_DESC

# train PillarNet baseline
bash tools/scripts/train_pillar_sup.sh TASK_DESC

Evaluation Scripts.

Use this command to inference a model:

python ./tools/dist_test.py CONFIG_PATH --work_dir output/waymo/finetune/xxx --checkpoint output/waymo/finetune/xxx/latest.pth --speed_test 

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

If you want to do local evaluation, please follow the waymo instructions here.

bazel-bin/waymo_open_dataset/metrics/tools/compute_detection_metrics_main ~/output/waymo/finetune/xxx/detection_pred.bin gt.bin