I have a model that resembles an autoencoder (convolution -> maxpool layers followed by convolution -> upsample layers). When I tried to compile it, I got the following error:
l03:~/workspace/movidius$ mvNCCompile ./frozen_v5_detector.pb -in input_map -on detection_output/Sigmoid -s12 -o ./movidius_v5_detector.graph
mvNCCompile v02.00, Copyright @ Movidius Ltd 2016
/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/ops.py:766: DeprecationWarning: builtin type EagerTensor has no module attribute
EagerTensor = c_api.TFE_Py_InitEagerTensor(_EagerTensorBase)
/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:45: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead
if d.decorator_argspec is not None), _inspect.getargspec(target))
[Error 5] Toolkit Error: Stage Details Not Supported: ResizeNearestNeighbor
It seems that the compiler does not support the upsampling layer which is implemented with TensorFlow ResizeNearestNeighbor. We had the same problem when trying to implement a similar model with TensorRT (the upsampling layer was not supported). However, TensorRT has an API for implementing unsupported layers and we were able to implement the model through the custom layer definition. First, am I reading this error correctly, and second, is there an upsampling layer or way to add custom layers that is supported by the NCS?
