Google Coral Flask Server

Expose deep learning models and receive data on the Coral usb accelerator and Edge TPU via a Flask app

Instructions

clone repository

git clone https://github.com/Danny-Dasilva/Google-Coral-Flask-Server.git

Path to folder

cd Google-Coral-Flask-Server

install dependencies

sh install.sh

run server

sudo python3 test.py

Running Models

in test.py

Empty Feed

Image = camera(Empty.AI())

returns a video feed with no model running

Image Classification

Image = camera(Classify.AI())

runs the image classify model, returns an image classification model

Object Detection

Image = camera(Detect.AI())

runs an object detection model detect model

Teachable Machine

Image = camera(Teachable.AI())

runs the teachable machine model

1, 2, 3, 4 add an image to each corresponding class

q reset current classes

Pose-Net

Image = camera(Pose.AI())

Anonymizer

Image = camera(anonymizer.AI())

example of the pose net model, when you move out of the frame it saves the image backround

Music Synthesizer

Image = camera(synthesizer.AI())

Three people are given control instrument and octave, the pitch is controlled with their right wrists and the volume with their left wrists.

Return values

in test.py

def my_function():
    while True:
        sleep(0.01)
        count = Image.numImages
        fps = Image.fps
        Inference = Image.inference
        Class = Image.Class
        Score = Image.Score

        print(fps, Inference, Class, Score, count)

For every model your run there are corresponding instance attributes that you can call from the camera class

fps returns the frames per second for the camera

inference returns the interference rate

numImages returns the number of examples for the teacheable machine

Class returns the class of the model

Score returns the percentage accuracy the model believes the class to be

e.g

teacheable returns 'One', 'Two', 'Three', or 'Four'

classify returns 'ping-pong ball' or 'spatula'

detect returns bounding box

You can modify my_function and add your own python operations

Args

All normal arguments that each model takes can be applied as long as the right class is being used

eg. 'python 3 test.py --model /'

Flask stuff

in order to access the server you need to go to the ip of the Coral and go to the port 5000

on the Coral

hostname -I

on browser on remote computer

<IP_Addr>:5000

Additional

If you receive a segmentation fault error simply rerun the code

If you encounter a gstreamer error 'must write bytes' then simply restart the program

If this problem persists power off plug in the camera again and change the usb C power cable