How to Run Yolov9 Real Time Object Detection on MILBOARD-AGX

WHAT YOU WILL LEARN?

1- Download the requirement package via terminal

2- Download the Deepstream-YOLO repo

3- Compile the Library

4- Run

ENVIRONMENT

Hardware: MILBOARD-AGX

OS:Ubuntu 20.04

In this blog post, you will learnhow to run Yolov9 Object Detection in real time. The GitHub repo hasbeen taken as a reference for the entire process.

1- Firstly download the requirement package via terminal


$ sudo apt install libgstrtspserver-1.0-dev

2- Check cuda and deepstream-app version if this both package is not installed please install via NVIDIA SDK manager.


$ nvcc --version
$ deepstream-app --version

3- Download the Deepstream-YOLO repo


$ git clone https://github.com/marcoslucianops/DeepStream-Yolo.git
$ cd DeepStream-Yolo

4- Compile the Library

a. Set the CUDA_VER according to your DeepStream version.

That tutorial we will use 11.4 version of CUDA


$ export CUDA_VER=11.4

b. Make the Lib


$ make -C nvdsinfer_custom_impl_Yolo clean && make -C nvdsinfer_custom_impl_Yolo

5- Edit the config_infer_primary_yoloV9.txt file according to your model

In this tutorial we will use yolov9-s-converted.pt.onnx model

6- Edit the deepstream_app_config.txt file according to your model.

Change the config_file parameter your yolo model

7- Run

NOTE: The TensorRT engine file may take a very long time to generate (sometimes more than 10 minutes).

Press ‘q’ to stop the program.

Looking to deploy high-performance AI at the edge? Explore our collection of Jetson carrier boards designed for seamless integration with Jetson SoMs — built for demanding vision, autonomy, and robotics workloads.

This tutorial is powered by the MILBOARD-AGX, a rugged, high-bandwidth carrier board built for Jetson AGX Orin. It features multiple camera inputs, Gigabit Ethernet, and extended I/O for real-time processing — making it ideal for object detection, multi-camera vision systems, and AI-enabled surveillance or inspection setups.

Thank you for reading our blog post.