Ubuntu setup

Immediately after writing the Ubuntu image file to the SD card, there is only a little free space left on the SD card.

This time, I expanded the free space by gparted.

sudo apt install gparted

However, a running Process occurred, so I stopped it with the kill command and ran the installation again. I do not know if it will occur every time, but just in case, I will post the screen below, so please refer to it when it occurs.

After installing gparted, start it with sudo gparted.

Select the yellow part in the left figure below and execute Resize by right-clicking. Then, the screen in the middle below will appear, so drag and pull it to the maximum value. After that, it is completed by reflecting it with Edit → Apply All Operations (right figure below)

Next, just in case, I ran sudo apt update and upgrade.

(Upgrading takes a long time...)

Install HailoRT

Next, we will install the Hailo environment.

A brief description of the tools to be installed this time is given below.

HailoRT---Hailo Runtime. A collection of APIs for accessing Hailo's HW and performing inference via HailoRT
Tappas---Demo application for checking various network models

Each can be downloaded from the Hailo Website below, but please note that you will not be able to download it just by registering as a user.

ご興味ございましたら弊社までお問い合わせをお願いします。

https://hailo.ai/developer-zone/sw-downloads/

This time, HailoRT uses Ver4.5.0 and Tappas uses Ver3.16.0.

Now let's move on to installing HailoRT.

Basically, the HailoRT manual is prepared, so it's not difficult if you follow it. I am using the Ubuntu Installer this time.

The following are specified as required packages, so first install them with apt.

• build-essential package (needed to compile the PCIe driver)
• (Optional) bison, flex, libelf-dev and dkms packages (needed to regsiter the PCIe driver using DKMS)
• (Optional) cmake (needed to compile the HailoRT examples)
• (Optional) pip and virtualenv (needed for pyhailort)

A shell script called install.sh is prepared in the download file, so you can easily install it just by executing this.
(The installation screen is shown below on the left)

Reboot after installation and you're done.
As an operation check, when I entered the virtual environment of HailoRT and executed scan, the HW of Hailo was successfully recognized (right figure below).

Install Tappas

Next, Tappas installation is not difficult if you install it according to the manual.

First of all, it is necessary to install Docker, but since it was already installed, I only registered it in the Docker group just in case.

sudo usermod -aG docker $USER

This time, we installed with "Build Docker image by your own" without using pre-built Docker image.

The file to download is small, but it takes time to build the Docker image yourself (actually, I regretted it later...).

In this case, it is necessary to copy the entire HailoRT installation folder to the release folder, and it will look like the following.

After that, we will build and run Docker, but shell scripts (build_docker.sh, run_docker.sh) are prepared for both, so you can easily install without having to type commands yourself.

However, it took about 5 hours to build, so it is recommended to leave it after executing the shell script.

(Since it is an ARM-based CPU, there is no way around this. I would like to consider cross-compiling from the next time.)

When it is finished, the container starts up as shown in the left figure below, and when you actually run Tappas, you can check the operation as shown in the right figure below.