Multi Dimension Input

[advian123]

Question

Hi, I have tensorflow model pb file which is multidimensional input, how can I pass that as command line argument?

Further information

Input node name: TensorInput: 0
Input shape is [?, 1024, 1024, 3] , [?1024, 1024, 3]

I'm passing this as an argument:
python -m tf2onnx.convert --saved-model test_model.pb --inputs TensorInput:0[[-1,1024,1024,3],[-1,1024,1024,3] --output model.onnx --outputs decoder:0

-1 because in readme it is written to us -1 for unknown dimension

[advian123]

@fatcat-z please help

[fatcat-z]

-1 is designed for unknown (dynamic) dimensions. Not sure if this is what you need.

[advian123]

Yes, I got that part that we have to use -1 for unknown dimension.
I want to know how to specify in the command line for a multidimensional array which is like this: [-1, 1024, 1024, 3] , [-1,1024, 1024, 3].

I tried this: python -m tf2onnx.convert --saved-model test_model.pb --inputs TensorInput:0[[-1,1024,1024,3],[-1,1024,1024,3] --output model.onnx --outputs decoder:0

But it failed. @fatcat-z

[fatcat-z]

What kind of errors you met? From my perspective, seems like there is 1 more unexpected '[' in your command. Could you please try the command below and share us the error you met?

python -m tf2onnx.convert --saved-model test_model.pb --inputs TensorInput:0[-1,1024,1024,3] --output model.onnx --outputs decoder:0

There is an example here which might help you a little.