Tensor initializers aren't getting converted to constants

[s1ddok]

Even though onnx contain needed input as tensor initializer:

The model yields an error while converting:

Underlying exception message was: Error compiling model: "Error reading protobuf spec. validator error: Layer 'Pad_0' consumes an input named 'initial_block' which is not present in this network.".
  RuntimeWarning)

is there a way to support constant inputs to layers?

[bhushan23]

@s1ddok you can call load_input_constants before converting node! see here
Which node is this? we might not be calling load_input_constants for this operator

[bhushan23]

also, you can give a try to new pytorch converter https://coremltools.readme.io/docs/pytorch-conversion

[s1ddok]

@bhushan23 it happens with constant input to padding and then conv. I manually patched converter to load that constant