I am using the up-squared-computer vision kit: https://up-board.org/upkits/up-squared-ai-vision-kit/
The kit comes preinstalled with ubuntu 16.04 and openvino 2018.5.455.
I trained a model using keras and tensorflow 1.14, the frozen graph from this model is converted using the model converter to both FP16 and FP32 without problems.
when doing inference using the inference engine python API: the following line causes an error
exec_net = plugin.load(network=model)
inference works fine when the device is CPU or GPU but not when using MYRIAD, sample models work with the MYRIAD so the cause is most likely the model converter. (both FP16 and FP32 work on GPU, FP16 fails on MYRIAD)
error output when running inference script: https://pastebin.com/jDXt0Gcr
keras_model summary: https://pastebin.com/rjC746EV
model optimizer output log when: https://pastebin.com/N22k1nyd
the error is not really helpful so i was wondering what steps i can take to get the myriad working
any advice is welcome
Dear Delnoije, Kevin,
Your model actually looks trivially simple so it should work on the MYRIAD. I don't see any odd, unsupported layers in your model. This is dumb but have you tried doing a power reset on your up-squared ?
Also are you stuck with 2018.5.455 ? Because that is a really old OpenVino version. We are on OpenVino 20192.01 now and many issues related to MYRIAD have been fixed. Is there a way to upgrade to the latest OpenVino version ?