Training your own DOPE model

Overview

The DOPE network architecture is intended to be trained on objects of a specific class, which means that using DOPE for pose estimation of a custom object class requires training a custom model for that class.

NVIDIA Isaac Sim offers a convenient workflow for training a custom DOPE model using synthetic data generation (SDG).

Tutorial Walkthrough

  1. Clone the Isaac Sim DOPE Training repository and follow the training instructions to prepare a custom DOPE model.

  2. Using the Isaac Sim DOPE inference script, test the custom DOPE model’s inference capability and ensure that the quality is acceptable for your use case.

  3. Complete until Run Launch File of the quickstart here.

  4. Move the prepared .pth model output from the Isaac Sim DOPE Training script into the ${ISAAC_ROS_WS}/isaac_ros_assets/models/dope path inside the Docker container.

    docker cp custom_model.pth isaac_ros_dev-x86_64-container:${ISAAC_ROS_WS}/isaac_ros_assets/models/dope

  5. Continuing with the DOPE quickstart’s Run Launch File section, run the dope_converter.py script with the custom model instead:

    ros2 run isaac_ros_dope dope_converter.py --format onnx \
   --input ${ISAAC_ROS_WS}/isaac_ros_assets/models/dope/custom_model.pth --output ${ISAAC_ROS_WS}/models/dope/custom_model.onnx

  6. Next, launch the ROS 2 launch file with the custom model.

    ros2 launch isaac_ros_dope isaac_ros_dope_tensor_rt.launch.py model_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/models/dope/custom_model.onnx engine_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/models/dope/custom_model.plan

  7. Bring up a data source containing your custom object class, such as a camera feed or a rosbag with the custom object.

    The data source should publish to the /image and /camera_info topics, and it should provide the appropriate tf frame representing the optical frame of the camera.

  8. Continue with the rest of the quickstart. You should now be able to detect poses of custom objects.