CUSTOM : Operation is working on an unsupported data type EDGETPU

1
votes

I am trying to retrain custom object detector model for Coral USB and follow coral ai tutorials from these link; https://coral.ai/docs/edgetpu/retrain-detection/#requirements

After retrained ssd_mobilenet_v2 model, converting edge tpu models with edge tpu compiler. Compiler result are these ;

Operator Count Status
CUSTOM 1 Operation is working on an unsupported data type
ADD 10 Mapped to Edge TPU
LOGISTIC 1 Mapped to Edge TPU
CONCATENATION 2 Mapped to Edge TPU
RESHAPE 13 Mapped to Edge TPU
CONV_2D 55 Mapped to Edge TPU
DEPTHWISE_CONV_2D 17 Mapped to Edge TPU

And visualize from netron ;

enter image description here

"Custom" operator not mapped. All operations are mapped and worked on tpu but "custom" is working on cpu. I saw same operator in ssd_mobilenet_v1

How i can convert all operators to edgetpu models? What is the custom operator ? ( you can find supported operators from here https://coral.ai/docs/edgetpu/models-intro/#supported-operations)

1
tensorflowgoogle-coralmobilenetedge-tpu
After running a TF Lite model through the Edge TPU Compiler, everything that is mapped to the Edge TPU is placed in a single op (edgetpu-custom-op). This means that if the model fully runs on the Edge TPU, the entire model is in this operation and input/output. Please see : coral.ai/docs/notes/build-coral/… - Manoj
@Manoj thanks, i'll check my versions - Cansel Oğuz
@Manoj still couldnt handle.... versions are ; ~$ edgetpu_compiler -v --->Edge TPU Compiler version 15.0.340273435 ~$ python3 -c "import pycoral.utils.edgetpu; print(pycoral.utils.edgetpu.get_runtime_version())" ----->BuildLabel(COMPILER=5.4.0 20160609,DATE=redacted,TIME=redacted), RuntimeVersion(13) - Cansel Oğuz

1 Answers

1
votes

This is the correct output for a SSD model. The TFLite_Detection_PostProcess is the custom op that is not run on the EdgeTPU. If you run netron on one of our default SSD models on https://coral.ai/models/, you'll see the PostProcess runs on CPU in that case.

In the case of your model, every part of the of the model has been successfully converted. The last stage (which takes the model output and converts it to various usable outputs) is a custom implementation in TFLite that is already optimized for speed but is generic compute, not TFLite ops that the EdgeTPU accelerates.