Deepstream or Darknet?


I’m an engineer student working on my final internship on real-time object detection/recognition. I use a jetson nano.
I am a beginner in the domain of computer vision and AI.

I want to implement a program on the jetson nano who can do object detection/recognition in real-time AND use the data (objects founds) to send order to motors (control a prototype).

I trained myself with darknet and use YOLOv3 for object detection/recognition. But I just saw that YOLO can be available on deepstream, who works on the jetson nano.

Knowing that I’m a beginner and that I want to modify the program to use the datas, what is the best solution for me?

Is it better/faster/easier to use Darknet or Deepstream for YOLO on the jetson nano?

I also have a laptop with a Nvidia Quadro RTX3000, if I use deepstream on the jetson nano, is it possible to train my YOLO on the laptop then transfer the weights and cfg on the jetson nano?


1 Like


It’s recommended to use Deepstream.
Deepstream is optimized for multimedia pipeline and it brings you an optimized performance on Jetson.

A general workflow is like this:

  1. Train your model with Darkenet on desktop.

  2. Update your YOLO parameter based on this document:

  3. Run your YOLO model with Deepstream:


1 Like