ONNX export: No conversion function for op type _arange

Hello everyone,

I’m trying to convert a trained yolov3 model to ONNX with onnx_mxnet.export_model.
At first I got an error with a slice_axis layer which I fixed thanks to https://github.com/apache/incubator-mxnet/issues/14875.
But now I get a new error : “AttributeError: No conversion function registered for op type _arange yet”.
I saw that other people were facing the same issue (see the previous link) but no solution has been given yet.
Has anyone successfully solved this problem?

I have Windows10, Python 3.6.5, MxNet 1.5.0 and ONNX 1.2.1.

Looks like the _arange is not supported yet: https://cwiki.apache.org/confluence/display/MXNET/ONNX+Operator+Coverage

You can write your own onnx arange operator and register it in mxnet/contrib/onnx/mx2onnx/_op_translations.py. BTW, if you want to convert the onnx to TensorRT engine, then maybe you have also to implement the plugin layer in tensorrt, which could be another time-consuming work.

It doesn’t work if only add the operator _arange in _op_translations.py, instead it said I have create a bad node. I wonder how to add the self-made op.