~ Edge TPU Server with Object Detection ~

The first time I put my hands on a Coral board I was pretty amazed by the demo that they make you run at the end of the set-up procedure. It's a simple object detector running on the board, but that's not the point. You can check the predictions made by the network comfortably with your web browser! I recognized that was a pretty nice trick and that could be helpful for some of my works either serious or playful. Let's say expecially playful :)) So, I looked at their code and I had to admit that they'd done a very good job at over complicate their code. It works just fine, but you can't really build on that shit :)

So, here we are! I put few line of codes, using the Flask library, to build a very simple server, copycat of the original one. It simply shows, on your localhost, frames coming from a thread where the detections are made. For this repository, I wanted to put the same object detector of the demo. So, it runs a Single-Shot-multiBox Detector (SSD) either on your CPU or TPU if you have an accellerator attached to your host. However, you can simply change the code in the detector and make your own class to show on your browser other magics 🔮

1.0 Getting Started

Clone this repository

git clone https://github.com/EscVM/Coral_Server_with_Object_Detection

1.1 Installations for the hosting device

Install on the hosting device the following libraries:

• opencv-python
• numpy
• Flask
• TensorFlow Lite Interpreter. If you're using the Coral USB Accelerator with the Raspberry download ARM32.

2.0 Run the Server

Open your terminal in the project folder and launch:

python3 server.py

or

 python3 server_login.py

if you want a very (un)protected version with a login page. (for the credentials think at your router)

N.B. as default it runs on the CPU. Use

 python3 server_login.py --tpu 1

if you have a Coral Accelerator.

Bonus: Random Projects

Here a little list of stupid projects I built using this repository:

• Person Detector @ Edge here
• Person Counter during Quarantine (lol) (A person is not immediately add to the counter!) here
• Pose Detection @ Edge here
• Person Detection in NIR @ Edge here
• Deep Sort @ Edge here