Autoware-ML¶
Autoware-ML is a machine learning framework for autonomous vehicle perception tasks. Built on PyTorch Lightning and Hydra, it provides a streamlined path from training to deployment.
Early Alpha
This project (tier4/autoware-ml) is in Early Alpha and will replace tier4/AWML. We welcome your feedback!
Get Started¶
-
Installation
Choose your preferred installation method
-
Quick Start
First interaction with Autoware-ML
-
Design
Architecture overview
-
Contributing
How to contribute
Key Features¶
- Hydra Configuration - Hierarchical YAML configs with runtime overrides
- PyTorch Lightning Core - Scalable training with multi-GPU support
- Optuna Integration - Automated hyperparameter optimization
- MLflow Tracking - Experiment logging and comparison
- ONNX & TensorRT Export - Production deployment with dynamic shapes
Supported Models¶
| Task | Modality | Model | NuScenes | T4 Dataset | Training | ONNX | TensorRT |
|---|---|---|---|---|---|---|---|
| Calibration Status | Camera, LiDAR | ResNet18 | |||||
| Detection 3D | LiDAR | CenterPoint | |||||
| Detection 3D | LiDAR | TransFusion | |||||
| Detection 3D | Camera, LiDAR | BEVFusion | |||||
| Detection 3D | Camera, LiDAR | StreamPETR | |||||
| Segmentation 3D | LiDAR | FRNet | |||||
| Segmentation 3D | LiDAR | PointTransformerV3 |
Available | Coming Soon
License¶
Developed by TIER IV, Inc. under the Apache 2.0 License.