I have a Rock960, 2 GB RAM Version, running on the Stock Android 7.1 OS. This version of the board can’t run TensorFlow Lite Object Detection. It can run for about a couple frames but shuts down automatically. Just to clarify, the 4GB RAM Version of the board runs TF-Lite Object Detection without any hiccups, but I don’t have my 4GB RAM Version anymore.
My next option then was to look at the Ubuntu Distribution that 96Boards provided, and install the OAID AI Frameworks there and hope that it runs.
I then looked at OAID and it seems that they built their own version of Tensorflow (TF-HRT) using this guide. And this version seems to be on Ubuntu. I then followed the instructions to go install Ubuntu 16.04 on my board. And here are where the problems begin:
- The Ubuntu 16.04 Distro is a Server version, which means that we don’t have a GUI.
- It seems difficult to connect to the local Wi-Fi so that I have internet access?
So my thought process now is this: I have to install a GUI to help me program AI, and I need the internet connection for it.
I know I don’t have internet since when I hit
ping google.com, I get “unkown host google.com”. Upon looking for guides on how to connect to the internet using the terminal, all I had to do was to type in
iwconfig wlan0 essid name key s:password, replacing the essid with the id of my wifi network, and the password with the plaintext password. However, upon doing so, I get the following error message:
sudo: unable to resolve host rock960: Connection refused
Error for wireless request “Send Encode” (8B2A):
SET failed on device wlan0: Invalid argument
So I then used (this guide)[https://superuser.com/a/477393/572364] on how to connect to the internet, but it didn’t seem to work as well.
Now, without internet connection, I won’t be able to install the general dependencies, let alone install a GUI and help make programming easier.
Now, I don’t think that Tensorflow would run on Linaro, as it is based on Debian (if I’m not mistaken).
Has anyone actually made Tensorflow/OIAD Object Detection (NOT IMAGE CLASSIFICATION) work on a Rock960 board?
Are there any guides out there on how to run Object Detection on the Rock960? Because honestly, it isn’t well documented, and there seems to be a lot of steps that they expect the end users to just get.
Any leads or guides would be very much appreciated.