Cannot convert alignment hourglass model to onnx

Hi! So I trained the alignment hourglass model using the [train script](https://github.com/deepinsight/insightface/tree/master/alignment/heatmap/train.py).

Then I used the [`mxnet_to_onnx.py` script](https://github.com/deepinsight/insightface/tree/master/alignment/heatmap) and I got the following error:
```plaintext
Traceback (most recent call last):
File "./mxnet_to_onnx.py", line 109, in
converted_model_path = onnx_mxnet.export_model(sym, all_args, [input_shape], np.float32, args.output)
File "/home/stf/.local/lib/python3.8/site-packages/mxnet/contrib/onnx/__init__.py", line 53, in export_model
return export_model_(*args, **kwargs)
File "/home/stf/.local/lib/python3.8/site-packages/mxnet/onnx/mx2onnx/_export_model.py", line 127, in export_model
onnx_graph = converter.create_onnx_graph_proto(sym, params, in_shapes,
File "/home/stf/.local/lib/python3.8/site-packages/mxnet/onnx/mx2onnx/_export_onnx.py", line 335, in create_onnx_graph_proto
converted, dtypes = MXNetGraph.convert_layer(
File "/home/stf/.local/lib/python3.8/site-packages/mxnet/onnx/mx2onnx/_export_onnx.py", line 96, in convert_layer
ret = convert_func(node, **kwargs)
File "/home/stf/.local/lib/python3.8/site-packages/mxnet/onnx/mx2onnx/_op_translations/_op_translations_opset12.py", line 299, in convert_deconvolution
assert len(input_nodes) == 3, 'Deconvolution takes 3 input if no_bias==False'
AssertionError: Deconvolution takes 3 input if no_bias==False
```

I tried a lot of things but I did not manage to solve this. How can I get past this error? I made no changes to the mxnet models.

I saw that @nttstar is the one who created that script. Do you have any insights? Thanks.