ROS wrapper for TensorRT YOLOv4

Referenced repositiory

Please check this repository for detail implementation. The trained files are provided by the following repository. The trained files are automatically downloaded when you build.

https://github.com/lewes6369/TensorRT-Yolov3

https://github.com/wang-xinyu/tensorrtx

https://github.com/tier4/AutowareArchitectureProposal.git

Trained model

Please note that above repository is under MIT or Apache 2.0 license.

Dependecies

  • Ubuntu 18.04
  • ros melodic
  • cuda 10.2
  • cudnn 7.6.5
  • tensorrt 7.0.0

How to use

  1. install ros and colcon.
  2. mkdir -p workspace/src
  3. cd workspace/src
  4. git clone https://github.com/wep21/tensorrt_yolov4_ros.git
  5. cd tensorrt_yolov4_ros && mkdir data
  6. Place trained models under data/.
  7. copy msgs under src/ from https://github.com/tier4/AutowareArchitectureProposal/tree/master/src/common/msgs.
  8. cd workspace
  9. colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release --packages-up-to tensorrt_yolo4
  10. source install/setup.bash
  11. roslaunch tensorrt_yolo4 tensorrt_yolo4.launch
  12. Publish /image_raw by real camera or rosbag.
  13. Check /rois/debug/image by rqt_image_view. detection

Interface

Input topic type

sensor_msgs::Image

Output topic type

autoware_perception_msgs::DynamicObjectWithFeatureArray