I'm new to deep learning and I want to export a MXNet model to ONNX. Running the follow python program:
sym = 'Experiment401_0_out-symbol.json'
params = 'Experiment401_0_out-0000.params'
input_shape = (1,3,640,832)
onnx_file = './Experiment401.onnx'
mx.npx.reset_np()
with mx.util.np_shape(False):
converted_model_path = onnx_mxnet.export_model(sym, params, [input_shape], np.float32, onnx_file)
It fails with the following error:
raise AttributeError("No conversion function registered for op type %s yet." % op)
AttributeError: No conversion function registered for op type pick yet.
I'm trying to add a 'pick' conversion function in: python/mxnet/contrib/onnx/mx2onnx/_op_translations.py to convert the 'pick' operator to ONNX tensors and nodes.
Can anybody suggest how to write it, or beter yet have this function implemented?
I would gladly appreciate any help or suggestions.
