MobileNetV2 is slow on EdgeTPU

[hychiang-git]

Description

Hi,

I am trying to run MobileNetV2 on the Edge TPU with a Dev Board Mini. I follow the instructions and run the classification example code on my board, but I only get around 20 ms per inference. I am wondering how I could reach 2.6 ms per inference speed as the benchmark table reports.

The tflite runtime version is 14 and the edge tpu compiler version is 16 on my dev board mini.

Thanks!

Click to expand!

Issue Type

Performance

Operating System

Mendel Linux

Coral Device

Dev Board Mini

Other Devices

No response

Programming Language

Python 3.7

Relevant Log Output

mendel@mocha-horse:~/coral/pycoral$ python3 examples/classify_image.py \
> --model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \
> --labels test_data/inat_bird_labels.txt \
> --input test_data/parrot.jpg
----INFERENCE TIME----
Note: The first inference on Edge TPU is slow because it includes loading the model into Edge TPU memory.
164.7ms
19.3ms
19.0ms
19.0ms
19.1ms
-------RESULTS--------
Ara: 0.75781