We have recently updated the base SD card image on the Jetson Nano, so there are different versions in circulation. If you have just received a Jetson Nano unit, it will be preinstalled with the AAU image version B, based on JetPack version 4.6. You can find version information in the README USB device. This page has been updated to reflect the new image B. For information related to the previous version A, see the bottom of the page.
Congratulations with you borrowing a NVIDIA Jetson Nano (NJN) Developer Kit. We hope it can be a source of inspiration in your studies and work.
We will start by using the NJN in a mode called headless, where you use your computer/laptop to connect to your NJN. If you want you can also connect a screen and keyboard+mouse, but the following guide assumes you will use headless operation.
Your NJN comes with pre-installed software on a micro SD card, including a selection of pre-trained deep neural networks for AI applications. If you want a different setup, you can download the base image from Nvidia and flash it to the SD card using 1) the USB-micro SD card dongle or 2) the with the micro SD to SD adapter (if you have a device with a SD card reader).
To get started, do the following
- Insert USB wifi dongle
- Insert micro USB cable in the NJN and the USB into your computer.
- Insert the camera USB cable in the NJN.
- Connect the power-supply in the 5V DC plug.
- After a bit of time you should see a device called AAU-README connected to your computer.
In the following you will need your username/password. For the AAU NJN the default is:
Default username: aaunano
Default password: aaunano
To get started, follow the below three steps:
- We will work with a remote desktop, and will be using the remote desktop program called "Nomachine", which can be downloaded for your platform here
- Install and open Nomachine and connect to your NJN using the IP 192.168.55.1 and your default username/password: aaunano/aaunano when prompted. You will now see the desktop of NJN in your Nomachine window. With Linux/Mac you can also connect using $ ssh email@example.com
- For security reasons, its recommend to update your password. Open a terminal (e.g. the "black screen" on the left), and type passwd - type in the current password and the new password.
You can configure Nomachine for headless mode (Virtual desktop and physical desktop are independent) or screenshare mode (Virtual dekstop mirrors physical desktop) by changing the file /usr/NX/etc/server.cfg or by running one of the scripts in a terminal:
Initially, when you receive the Jetson Nano, it is configured for headless mode.
If the Dock (The icons on the left side of the screen) disappears, you can use Alt+F2 and enter the command "r" to restart Gnome.
Try a demo from jetson-inference
- In the terminal type "cd jetson-inference/build/aarch64/bin/"
- Theres a number of demos. Try e.g. the object detection program using the camera with (It takes a few minutes the first time so please be patient):
./detectnet-camera --camera=/dev/video0 --width=640 --height=480
or image recognition
./imagenet-camera --camera=/dev/video0 --width=640 --height=480
For both cases, point the cursor to the upper bar, and the x for terminating the window+application is located in the upper left corner.
Note: scripts from jetson-inference do not function well together with nomachine. There are three options for running these scripts.
Disable graphical output with the --headless option
Plug in a screen to the DP or HDMI port. You can still use nomachine in screenshare mode by running the script "sudo ./screenshare_mode.sh"
Stream the output to another computer via rtp. Note that this does introduce some delay. See the documentation of jetson-infererence
JUPYTER NOTEBOOK DEMOS
The introductory course from Nvidia contains some demos. You can follow the course or simply look at the Jupyter Notebooks. If you follow the course, you can skip the step "Introduction and Setup".
Open the terminal and run the following command to start the introductory course Jupyter Notebook server
sudo docker run --runtime nvidia -it --rm --network host --volume ~/nvdli-data:/nvdli-nano/data --device /dev/video0 nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1
This downloads the course material from Nvidias container cloud (NGC)
Open your browser on your own computer, point to 192.168.55.1:8888 and type the password dlinano.
Go to dlinano/classification, and double click classification_interactive.ipynb. In the setting right now, it is build to classify into two classes - thumbs-up and thumbs-down (or two other classes of your choice). Run all the cells, either by using Run->Run all cells or by selecting each code cell and click Shift-Enter (notice the numbering on the left side of the cell which is your execution order).
In the widget in the last cell, try to a number of thumbs up and press add to add these images to the training set. Select category thumbs_down, and do some thumbs-down in front of the camera and some of these images to the training set. Select a number of epochs, maybe a couple and click "train". Be a bit patient and you should see a blue progress bar. In the end, try go to live mode, and shift between thumbs up/thumbs down. Does the system classify?
Remember to read the text in the green box at the top. To release the camera resources in the end, you need to click Kernel and one of the Restart options.
You can stop the container running Jupyter Lab again with a 'Ctrl-D' or type ´exit'[Enter]
You can also look at the video walkthrough of the course demos.
To get WiFi @AAU, follow the this guide. (For some reason, connection may be a bit unstable but please just be persistent and follow the below guide and it should work)
- In the upper-right corner on you NJN nomachine window, select network (Two up-down arrows). When on the AAU network you can select AAU.
- This setup is temporary to setup a network configuration. Next well make a persistent solutions. On the NJN within your nomachine screen, open the Chromium browser on the desktop and go to net.aau.dk. Login with you AAU credentials and create a device. Give it a name like NJN, and click "Create". Under Automatic installation, click Ubuntu Linux, and download/click the configuration script. Click Keep.
- Open a terminal on the NJN by clicking on the upper left search bar, and type "terminal" and Enter. Run the script by typing "python3 Downloads/AAU-network-setup.py" and Enter.
To turn off the Nvidia Jetson Nano, click in the top right corner to open the power menu and select "Power Off".
You can also connect your screen+keyboard+mouse and work directly on the NJN. Since AAU image version B, you no longer need to configure the display resolution manually.
To conclude: The NVIDIA Jetson Nano developer kit offers an easy way to get started with AI applications, including a completete environment and material. Read and learn more at:
Information for AAU image version A
If your Jetson Nano is installed with version A of the AAU image, most of the information above is still valid. The main differences include the following:
- username/password are by default dlinano/dlinano.
- Due to technical decisions, if you connect a physical display, the possible resolutions are lower than 1280x1024. If you wish to use a higher resolution you need to manually adjust /etc/X11/xorg.conf to fit your usecase and monitor setup.
- The Jupyter Notebook server automatically starts at 192.168.55.1:8888
- Version A does not include scripts to change nomachine settings.