Tomas’s Status Report for 2/19/22

What did you personally accomplish this week on the project?

• I spent this week by finishing the initial setup of the Jetson Xavier AGX, which took me a few days. I ran into some challenges during the setup, which I will describe here in the hopes that it might help future students in a similar situation.
  • Host OS: The operating system on the host machine must be Ubuntu 16.04 / 18.04. Anything else will not allow you to flash an OS image onto the Jetson. I dual-booted an old laptop with Ubuntu 18.04 to use as the host machine.
  • Flashing OS to Target: This may take a while, especially if your host machine has a slower hard drive. You may get pop-up warnings from the SDK Manager that the current process has been taking a while, and if you want to continue flashing. Say yes. It took me about an hour and a half, but you can assure yourself that the installation hasn’t stalled by clicking on the terminal tab of the manager and analyzing the timestamp of the most recent operation.
  • Networking: After flashing the OS, you will likely get an error saying that some of the SDKs couldn’t be installed on the Jetson, and that you should run sudo apt-get update && sudo apt-get check on the Jetson. Before you can do this though, you must connect the Jetson to CMU’s wired network. Simply plugging in the ethernet cable likely won’t be enough, because you need to register the device with Computing Services. Follow the guide here, and use the instructions for step 3 labeled “For Faculty” and “To register devices located in all other academic or administrative buildings…”. Note that it can take 30-60 mins for the Jetson to connect after it’s registered, which you can test by repeatedly refreshing any webpage in the pre-installed Chromium browser.
    • The first time I booted up the Jetson after initially setting it up, it could not establish an Ethernet connection to the wired network. It seems that the networking ethernet port eth0 was not automatically being activated. I was able to fix the issue with the following:
      • Add these two lines to /etc/network/interfaces:
        • auto eth0
        • face eth0 int dhcp
      • Save and exit, and run sudo ifup eth0
• After completing the initial set up, I also installed the Jetson-Inference library, an AI Computer Vision provided by NVIDIA and designed specifically for the Jetson models’ hardware. We expect to use this library for our object detection requirements.
• I also put in a few hours to help my team create our design review presentation. I specifically worked on the two slides related to the CV pipeline using the Jetson, as well as helping out with the related portion of the block diagram.

Is your progress on schedule or behind?

• Our progress is on schedule. I had hoped to have gotten the set up completed earlier in the week, so that I could start familiarizing myself with the CV and DNN libraries we will use. However, we expected there would be issues with the set up, and that is why we budgeted a week for this task.

What deliverables do you hope to complete in the next week?

• Next week I expect to start learning and applying the object detection features of the Jetson-Inference library. I expect to start by understanding the demo programs that used the Single Shot Detector DNN, and to create a program that does this in real time.
• I also want to create a plan for eventually training our own models for the DNN. As mentioned in our design review presentation, we expect to start by using a pre-trained model, and then move onto training a model from a subset of the Open Images dataset, and then creating our own custom dataset. Each successive option gives us more control over the types/number of object classes we would like to detect, but also require a greater time investment. To ensure we reach our MVP with a working object detection component, we will explore these model options successively.