Quick Start¶

This guide gets you from zero to a trained model. We'll train a PTv3 3D semantic segmentation model using the NuScenes dataset.

1. Setup Dataset¶

Download the NuScenes full dataset (v1.0) from the official website after registration. After the download, confirm that the dataset is located at $AUTOWARE_ML_DATA_PATH/nuscenes.

2. Launch the Container¶

cd ~/autoware-ml
./docker/container.sh --run

3. Generate Dataset Info Files¶

Autoware-ML needs preprocessed info files that index the dataset:

autoware-ml create-dataset \
    --dataset nuscenes \
    --task segmentation3d \
    --root-path data/nuscenes \
    --out-dir data/nuscenes/info \
    --version v1.0-trainval

This creates pickle files for train/val splits.

4. Train the Model¶

autoware-ml train --config-name segmentation3d/ptv3/voxel005_102m_nuscenes

Training progress appears in your terminal. Checkpoints are saved automatically.

5. Monitor with MLflow¶

autoware-ml mlflow ui --port 5000

Open http://localhost:5000 to view loss curves, metrics, and hyperparameters.

6. Export for Deployment¶

autoware-ml deploy \
    --config-name segmentation3d/ptv3/voxel005_102m_nuscenes \
    --weights mlruns/segmentation3d/ptv3/voxel005_102m_nuscenes/<run_id>/artifacts/checkpoints/best.ckpt \
    deploy.tensorrt.enabled=false

This generates an ONNX file. TensorRT export is disabled because PTv3 requires a runtime with matching sparse convolution plugins.

To evaluate a trained checkpoint before deployment:

autoware-ml test \
    --config-name segmentation3d/ptv3/voxel005_102m_nuscenes \
    --weights mlruns/segmentation3d/ptv3/voxel005_102m_nuscenes/<run_id>/artifacts/checkpoints/best.ckpt