Google Edge TPU
Rouyun Pan
Google Edge TPU
https://coral.withgoogle.com/products/
Dev Board datasheet
4 TOPS, 1.5W
System Of Module (SOM)
USB Accelerator
https://www.seeedstudio.com/Coral-USB-Accelerator-p-2899.html
Neural network accelerator
https://patents.google.com/patent/US20190050717A1/
Tensorflow lite
The basic workflow to create
a model for the Edge TPU
The compiler creates a single custom
op for all Edge TPU compatible ops
Inference Engines
Run multiple models with
multiple Edge TPUs
Imprinting Engine
• Benefits:

• Transfer-learning happens on-device, at
near-realtime speed.

• You don't need to recompile the model.

• Drawbacks:

• Training data size is limited to a max of
200 images per class.

• It is most suitable only for datasets that
have a small inner class variation

• The last fully-connected layer runs on
the CPU, not the Edge TPU. So it will
be slightly less efficient than running a
pre-compiled on Edge TPU.
https://coral.withgoogle.com/docs/edgetpu/retrain-classification-ondevice/
Performance
c
c
Perfermance
Comparing with Others
https://medium.com/@aallan/benchmarking-edge-computing-ce3f13942245
Demo
https://www.youtube.com/watch?v=AokyF2HWnqk

Google edge tpu