使用 onnx_coreml 转换节点类型切片时出错

Question

I have converted my dar.net YOLOv3-SPP model into a PyTorch.pt model. I then converted the.pt to a.onnx. My end goal is to get to a CoreML model. I tried to use this GitHub repository. However when converting my model I am getting an error like this...

...    
145/229: Converting Node Type LeakyRelu
146/229: Converting Node Type Conv
147/229: Converting Node Type Reshape
148/229: Converting Node Type Transpose
149/229: Converting Node Type Reshape
150/229: Converting Node Type Slice
Traceback (most recent call last):
  File "convert2.py", line 11, in <module>
    coreml_model = convert(model_proto, image_input_names=['inputImage'], image_output_names=['outputImage'], minimum_ios_deployment_target='13')
  File "/usr/local/lib/python3.6/dist-packages/onnx_coreml/converter.py", line 626, in convert
    _convert_node_nd(builder, node, graph, err)
  File "/usr/local/lib/python3.6/dist-packages/onnx_coreml/_operators_nd.py", line 2387, in _convert_node_nd
    return converter_fn(builder, node, graph, err)
  File "/usr/local/lib/python3.6/dist-packages/onnx_coreml/_operators_nd.py", line 2011, in _convert_slice
    end_masks=end_masks
  File "/usr/local/lib/python3.6/dist-packages/coremltools/models/neural_network/builder.py", line 4220, in add_slice_static
    assert len(strides) == rank
AssertionError

The script I am using is this...

import sys
from onnx import onnx_pb
from onnx_coreml import convert

model_in = sys.argv[1]
model_out = sys.argv[2]

model_file = open(model_in, 'rb')
model_proto = onnx_pb.ModelProto()
model_proto.ParseFromString(model_file.read())
coreml_model = convert(model_proto, image_input_names=['inputImage'], image_output_names=['outputImage'], minimum_ios_deployment_target='13')
coreml_model.save(model_out)

This simple python script should work, but I don't know why I am getting this error. I am very new to Machine Learning, so I do not understand how I can even begin to try to solve this issue. What should I do in order to convert my.onnx model to CoreML successfully?

Answer 1

Looks like rank mis-match between input tensor rank and slice parameter Could you please file bug at onnx-coreml

As @matthijs-hollemans commented, try installing latest onnx-coreml

pip install onnx-coreml==1.2

Few other concerns:

1. What is the version of onnx model you are working with? with Operator-9 slice behavior is changed and that could be potential failure point from converter.

2. Could you please attach ONNX model as well?