AlbertSid / image-search-raspberry-pi

Image Search by Tensorflow lite in Raspberry Pi (with Google Edge TPU)

image-search-raspberry-pi

This is an online image search platform running on the Raspberry Pi via Object detection through Tensorflow lite.

• Search image via URL link
• Search image via upload an image
• Search similar images

Requirements

• numpy==1.16.2
• Flask>=1.0.0
• requests==2.22.0
• Pillow>=6.2.0
• Google Edge TPU or Tensorflow Lite (Follow the instructions here)

Usage

It uses the flask in python, just one command to run the web server simply:

$ python3 server.py

Then visit the website: 127.0.0.1:5000 (port 5000 on the Raspberry Pi).

Here‘s screenshots of running on the Raspberry Pi 3B+ with Coral USB Accelerator:

How does image search work on the Raspberry Pi?

• Raspberry Pi 3 Model B+

• Coral USB Accelerator (Optional)

Important Things：This program does not have to use the Coral USB Accelerator, but I used it for accelerating the inference process in Tensorflow lite.

Thanks for object detection source code in the coral example, this image search platform is based on this. I used the MobileNet SSD v2 (COCO) model by default, you can modify it in the file detect_image.py .

If you don't have a usb accelerator, you can compile the entire Tensorflow lite on the Raspberry Pi and learn about its object detection here.