deploy_to_autoware

deploy_to_autoware

This tools aim to deploy for Autoware.

Update onnx metadata

This script injects extra meta information.

• Run

python tools/deploy_to_autoware/update_onnx_metadata.py {path to onnx file} --version 1 --doc_string "Example additional description"

By this update, logs in Autoware log files or /rosout topic can have model metadata.

[autoware_lidar_centerpoint_node-1] [I] [TRT] Input filename:   /home/autoware/autoware_data/lidar_centerpoint/pts_backbone_neck_head_centerpoint_tiny.onnx
[autoware_lidar_centerpoint_node-1] [I] [TRT] ONNX IR version:  0.0.6
[autoware_lidar_centerpoint_node-1] [I] [TRT] Opset version:    11
[autoware_lidar_centerpoint_node-1] [I] [TRT] Producer name:    AWML
[autoware_lidar_centerpoint_node-1] [I] [TRT] Producer version: v1.0.0
[autoware_lidar_centerpoint_node-1] [I] [TRT] Domain:           ai.onnx
[autoware_lidar_centerpoint_node-1] [I] [TRT] Model version:    1
[autoware_lidar_centerpoint_node-1] [I] [TRT] Doc string:       Example additional description

[TBD] Generate ROS parameter

This script makes the ROS parameter from config file of autoware-ml.

• Run the script

python tools/deploy_ros_parameter_file/deploy_ros_parameter_file.py {config file of autoware-ml} {config file of deployment} {target_param_config_file}

• Example

tools/deploy_ros_parameter_file/deploy_ros_parameter_file.py \
projects/TransFusion/config/t4dataset/90m-768grid/transfusion_lidar_90m-768grid-t4xx1.py \
projects/TransFusion/config/deploy/transfusion_lidar_tensorrt_dynamic-20x5.py \
projects/TransFusion/config/deploy/target_param_config_file.yaml

• The config file of projects/TransFusion/config/deploy/target_param_config_file.yaml is as below.
  • The script fulfill the ROS parameter value of autoware-ml

/**:
  ros__parameters:
    class_names: {class_names}
    voxels_num: [{backend_config.model_inputs.input_shapes.voxels.min_shape[0]}, {backend_config.model_inputs.input_shapes.voxels.min_shape[1]}, {backend_config.model_inputs.input_shapes.voxels.min_shape[2]}]
    point_cloud_range: {point_cloud_range}
    voxel_size: {vocel_size}
    num_proposals: {num_proposal}