My First Jetson Nano Object Detection Inference Server

In this tutorial video, we'll guide you through the process of setting up your Jetson Nano board and running your first inference example.
We'll cover the hardware overview, writing the image to the microSD card, powering up the board, installing Ubuntu 18.04, and the first boot essentials.
We'll also show you how to access your Jetson Nano board via SSH, use Docker to spin up the object detection inference server, and demonstrate an inference example.
This tutorial video is a comprehensive guide for beginners to get started with Jetson Nano Developer Kit.
Chapters:
0:00 Introduction
00:57 Jetson Nano 4GB Board Overview
01:52 Write the Image to the microSD Card
04:22 Power Up the Board
05:15 Installing Ubuntu 18.04 on Jetson Nano
06:36 First Boot Essentials - Internet Connectivity, Docker, Installing JTOP
09:08 Accessing the Bord via SSH
10:06 Using Docker to Spin up Object Detection Inference Server
10:56 Inference Example
13:55 Conclusion
Resources:
🌏 Roboflow: roboflow.com
🌌 Roboflow Universe: universe.roboflow.com
📖 Getting Started with Jetson Nano Developer Kit instructions: developer.nvidia.com/embedded....
📄 Roboflow NVIDIA Jetson Inference Server documentation: docs.roboflow.com/inference/n...
⚽ Football Players Tracking video: • Football Players Track...
Stay updated with the projects I'm working on at github.com/roboflow and github.com/SkalskiP! ⭐

[@tomaszbazelczuk4987]

Awesome!!! I can’t wait for a next episode 🙂

[@Roboflow]

Stay tuned! We have a gigantic video backlog, but I'll do my best. :D

[@richardverheyen6373]

I'm loving this series so much, please do keep making them. I'm trying so hard to get one of my newly trained models running in Flutter. Will the Roboflow team ever have time to look into a tutorial for object detection and maybe even tracking in Flutter?

[@Roboflow]

Well… I hear that request for the first time. People were walking about swift, and Java or even Kotlin. But never Flutter. I’d need to do a research if that’s even possible. And btw, thanks a lot for the kind works 🙏🏻

[@IsiUbilla]

more videos with jetson please!

[@g.s.3389]

great more with jetson!!

[@Roboflow]

Comming soon! I hope :)

[@YigalBZ]

Thank you for the clear and good video. I got a little confused as you use pre made docker. Could you please advise if it is usable for the following application: I have a Jetson Nano 4GB, connected to a camera. The target is to identify a train toy on track (specifically locomotive among few), and get it coordinates, in real time. I assume not because the latency of the server. What do you think?

[@MichaMichalewski]

Pozdro ziomek!

[@kaisbedioui7456]

Interesting! I like to see more of the edge devices projects Question: is the roboflow inference server optimized in a way for Jetson devices?

[@Roboflow]

We have two servers. One implemented in node.js where we use tf.js as an engine. We have second one that is much more optimized for NVIDIA devices.

[@infocus2160]

can we use the same inference server for RTMDet model as well.

[@sergiirizvan4669]

It would be cool to see how you can run object detection + tracking on video stream on Jetson Nano

[@Roboflow]

This sounds awesome! Could you add that to our video ideas: github.com/roboflow/notebooks/discussions/categories/video-ideas ?

[@harperzappone5495]

hi! I was wondering how you would do something like this if you didn't train your model using roboflow? I use roboflow to annotate images, but I trained my models on a large server we use at my work. Can I upload those .pt weight files to the inference server instead of importing from roboflow?

[@Roboflow]

Not for now… but you can upload your trained model to Roboflow and then use it in our server.

[@sansy9568]

Hello, can u tell me which version of Yolo is capable of runninng on Jetson Nano with acceptable FPS ?

[@snehitvaddi]

Hey Peter! For my project I am thinking of using external camera like OAK-D or generic webcam for doing Object detection and Object tracking. What would you suggest? Respberry PI or Jetson Nano to do the AI stuff?

[@Roboflow]

Depends on FPS you want to get and the model that you plan to use :)

[@snehitvaddi]

@@Roboflow We're talking about a conveyer belt that can move at least 100 objects per minute and runs continuously for 2-3 hours. If I'm not mistaken, storing that much data and processing OpenCV, ML stuff will be impossible. What are your thoughts? Is that possible on an edge device?

[@Roboflow]

@@snehitvaddi it think it should be quite easy to do in real time :)

[@snehitvaddi]

@@Roboflow okay, let say I'm going to use Yolov8, and I might need 25 FPS. What would you suggest? Respberry PI or Jetson Nano or someother thing?

[@Roboflow]

@@snehitvaddi Jetson Nano ishould work. I think that Pi is to wake

[@carolenejoy9562]

I powered up my nano using both micro usb and barrel jack, but the green light is not working and not booting. But when I checked the green light seperately using a multimeter, the light turns on. Can someone help me out?

[@saintprado1266]

You need a jumper clip on J28 to disable USB power and active the power to the Barrel Jack.

[@CoolAdministrator]

Can you do a video on the Khadas VIM4?

[@Roboflow]

You can submit your video idea here: github.com/roboflow/notebooks/discussions/categories/video-ideas

[@gemlisboa3236]

Do you have youtube tutorial on how connect arducam to jetson nano

[@Roboflow]

Not yet... Is that something you'd be interested in?

[@gemlisboa3236]

I really interested because it helps it to me

[@Roboflow]

@@gemlisboa3236 I'll add it to the list, but I need to say it is pretty long already. There is a lot of interest in edge devices.

[@gemlisboa3236]

@@Roboflow its okay, Ill be waiting. Thank you so much

[@sainandhan1108]

Please explain how to visualise results tooooo

[@ishmaelchirwa3823]

v4l2-ctl --device=/dev/video0 --stream-mmap --stream-count=100 --stream-to=test.mp4

[@AlainPilon]

Wait a sec! You can connect both keyboard and mouse with a single dongle?

[@Roboflow]

Hahaha! You can if both are Logitech. Not sure if it works same with other manufacturers.

[@AlainPilon]

@@Roboflow I already have the logitech mouse, but they down have good wireless keyboards for coding (I have a wired one for gaming). Maybe I should check them again.

[@weistrass]

I agree with the Jetson criticism, but give us some alternatives...

[@MinaSchloch]

Bro you did apt update but not apt upgrade! Your system, at least the desktop, looks extremely old. Like, 2016 old. Like that, also with the GUI and all, you may not want to open any ports for that system to the internet. Really interesting board, funny that NVIDIA with their hater proprietary driver and all, used Linux for their own hardware! They also use RISC-V in their GPUs afaik. Poorly I guess RISC-V is not copylefted, so people can make technology based on it proprietary. I guess using that board for AI should require a better cooling solution. Nice GUI warning "System is throttling" in the upper right corner, and it throtteled A LOT. You may be able to overclock the components and have it never throttle, when using a better cooler. Your installation guide was... "click on the first link in Google" is something you should never say, often its Ads. Please link that official documentation and all Downloads in the description. Nice video!

[@Roboflow]

Well… thanks a lot for that comment. You gave me a lot of motivation to do better job next time ;)