EdgeTPU-FaceNet

Implement SSD and FaceNet on Edge TPU Accelerator.

Class can be increased immediately.

You can use 'q' to quit and 'a' to add new class.

Requirement

tensorflow - 1.15.0

Ubuntu - 16.04

Edge TPU compilier - 2.0.267

USB Camera

Demo

about 25 FPS

https://youtu.be/I9F_GT_quFs

Usage

1. Download SSD and FaceNet weight

(1). Download ssd weight and put in weight/SSD

(2). Download facenet weight and put in weight/FacaNet

Both weights have already been compiled and quantized.

2. Run demo.py

$ git clone https://github.com/Kao1126/EdgeTPU-FaceNet.git
$ cd EdgeTPU-FaceNet
$ python3 demo.py

Valitading on FLW

1. Create lfw folder

   $ mkdir lfw

2. Download LFW datasets and put in lfw

   $ python3 validate_lfw.py

Reference

• coral: https://coral.withgoogle.com/docs/edgetpu/models-intro/

• tensorflow: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/quantize

• face net: https://github.com/LeslieZhoa/tensorflow-facenet