Convert trained Gluoncv model to ONNX

Gluon
1

Hi everyone,

Currently I’m trying to convert a finetune model trained with Gluoncv to ONNX but I got the error:
“No conversion function registered for op type slice_like yet”
Seems like the conversion is not fully developed yet.

I just followed a simple finetune example and got stuck
https://gluon-cv.mxnet.io/build/examples_detection/finetune_detection.html

DO you have any advices or comments?
Thanks in advance.

2

If the slice_operation is at the end of your model, then you can apply it as a post processing operation in e.g. python or C++. For this you can reduce the NN symbol architecture to have the last layer before the slice operation as the new output of your model:

def get_symbol_without_slice(symbol):
    new_output = symbol.get_internals()[<name_of_the_layer_before_slice>]
    return new_output

Depending on the model complexity it might still be more convenient to stay within MXNet for inference.