search

google-coral / tflite

Examples using TensorFlow Lite API to run inference on Coral devices
https://coral.withgoogle.com
Apache License 2.0
181 stars 67 forks source link

abnormal inference speed on rpi 3b using usb accelerator #18

Closed natxopedreira closed 4 years ago

natxopedreira commented 4 years ago

Hello

Im running a raspbery pi3b with Buster, python 3.7 and fresh usb accelerator install https://coral.ai/docs/accelerator/get-started/

pip3 freeze result tflite-runtime==2.1.0.post1

Im running this example https://github.com/google-coral/tflite/tree/master/python/examples/detection

And in the page is mentioned this benchmark as reference

INFO: Initialized TensorFlow Lite runtime. ----INFERENCE TIME---- Note: The first inference is slow because it includes loading the model into Edge TPU memory. 33.92 ms 19.71 ms 19.91 ms 19.91 ms 19.90 ms -------RESULTS-------- tie id: 31 score: 0.83984375 bbox: BBox(xmin=228, ymin=421, xmax=293, ymax=545) person id: 0 score: 0.83984375 bbox: BBox(xmin=2, ymin=5, xmax=513, ymax=596)

But im getting a much slower inferece speed

----INFERENCE TIME---- Note: The first inference is slow because it includes loading the model into Edge TPU memory. 276.87 ms 75.67 ms 75.27 ms 74.61 ms 74.14 ms -------RESULTS-------- tie id: 31 score: 0.83984375 bbox: BBox(xmin=226, ymin=417, xmax=290, ymax=539) person id: 0 score: 0.83984375 bbox: BBox(xmin=2, ymin=5, xmax=507, ymax=590)

natxopedreira commented 4 years ago

i will post it on edgetpu repo seems more logical

Namburger commented 4 years ago

@natxopedreira which model did you run?

natxopedreira commented 4 years ago

Sorry i closed this as i posted on the other repo, i reply to you in the other one https://github.com/google-coral/edgetpu/issues/115

Thanks