Cloud Edge TPU Demo

Overview

This repository contains the sample script to run the image classification and object detection using Google USB Coral Accelerator Edge TPU in RaspberryPi 4.

Demo

Boat Classification

The different boat images are uploaded into Google's AutoML Vision portal and tagged into five different categories

• Sail Boat
• Kayak
• Ferry
• Cruise
• Gondola

Example images

Once the models are trained, the trained models are exported into both TFLite ML model and TFLite ML model optimized for Edge TPU.

Navigate to the classfication directory

cd classfication

CPU/GPU:

python3 classify_image_non_edge.py --model models/boat.tflite --label models/boat_labels.txt --input images/boat1.jpg

Initializing TF Lite interpreter...
INFO: Initialized TensorFlow Lite runtime.
----INFERENCE TIME----
111.8ms
110.0ms
107.0ms
112.0ms
110.8ms
-------RESULTS--------
b`kayak`: 0.7523

With Edge TPU - USB Accelerator

python3 classify_image.py --model models/boat_edge_tpu.tflite --label models/boat_labels.txt --input images/boat1.jpg

You should see results like this:

Initializing TF Lite interpreter...
INFO: Initialized TensorFlow Lite runtime.
----INFERENCE TIME----
Note: The first inference on Edge TPU is slow because it includes loading the model into Edge TPU memory.
11.8ms
3.0ms
2.8ms
2.9ms
2.9ms
-------RESULTS--------
b`kayak`: 0.76562

Object Detection

The object detection script runs the webcam from Raspberry Pi and run the object detection using the CocoSSD MobileNet

# Navigate to the object detection directory
cd objectdetection

python3 detection_webcam.py --modeldir model

This wil open the webcam and detect the object with bounding boxes.

Notes

• Connect through ethernet attached to pi

ssh pi@192.168.2.2