New version model conversion error

[psiydown]

New version model: https://storage.googleapis.com/mediapipe-assets/face_landmark.tflite
tflite2onnx face_landmark.tflite face_landmark.onnx
Error prompt:
Data type float16 not supported/tested yet, the generated model may contain error
Data type float16 not supported/tested yet, the generated model may contain error
Data type float16 not supported/tested yet, the generated model may contain error
Data type float16 not supported/tested yet, the generated model may contain error
Data type float16 not supported/tested yet, the generated model may contain error
Traceback (most recent call last):
File "N:\anaconda3\lib\runpy.py", line 197, in _run_module_as_main
return _run_code(code, main_globals, None,
File "N:\anaconda3\lib\runpy.py", line 87, in run_code
exec(code, run_globals)
File "N:\anaconda3\Scripts\tflite2onnx.exe_main.py", line 7, in
File "N:\anaconda3\lib\site-packages\tflite2onnx\convert.py", line 58, in cmd_convert
convert(args.tflite_path, args.onnx_path)
File "N:\anaconda3\lib\site-packages\tflite2onnx\convert.py", line 44, in convert
model.convert(explicit_layouts)
File "N:\anaconda3\lib\site-packages\tflite2onnx\model.py", line 42, in convert
g.convert(explicit_layouts)
File "N:\anaconda3\lib\site-packages\tflite2onnx\graph.py", line 102, in convert
self._propagateLayout()
File "N:\anaconda3\lib\site-packages\tflite2onnx\graph.py", line 173, in _propagateLayout
t.transform()
File "N:\anaconda3\lib\site-packages\tflite2onnx\tensor.py", line 126, in transform
self.shape = self.layout.transform(self.shape)
File "N:\anaconda3\lib\site-packages\tflite2onnx\layout.py", line 27, in transform
output = transform(input, self.source, self.target)
File "N:\anaconda3\lib\site-packages\tflite2onnx\layout.py", line 16, in transform
transfrom_axis = [input[p] for p in perm]
File "N:\anaconda3\lib\site-packages\tflite2onnx\layout.py", line 16, in
transfrom_axis = [input[p] for p in perm]
IndexError: list index out of range

@keijiro Can you give me any help? Thank you

[keijiro]

That's basically an issue with tflite2onnx. Could you ask them first?

[psiydown]

@keijiro I have already asked tflite2onnx but have not replied yet. Additionally, I have successfully converted onnx through another method:
python - m tf2onnx.convert --opset 9 --tflite face_Landmark.tflite --output face_Landmark.onnx
But there was an error running your Python script:

Traceback (most recent call last):
File "D:\python\scripts\1\face_landmark_barracuda.py", line 73, in
replace_pad_ops(model)
File "D:\python\scripts\1\face_landmark_barracuda.py", line 28, in replace_pad_ops
data = next(n for n in model.graph.value_info if n.name == node.input[0])
StopIteration

[keijiro]

Then how about using Colab instead of your local Python environment?

[psiydown]

@keijiro I tried, it's not an environmental issue, the result is the same. Local use of old version face_landmark.onnx model is fine, but the new version cannot generate Barracuda after tf2onnx.convert.

[keijiro]

Thanks for the clarification. The model design (or its architecture) might be changed. I think you don't only need updating the Python script but also modifying the C# code. In other words, you can't try the new model with a quick hack.