Add transposition to the tflite model

[pipipyau]

Issue Type

Feature Request

OS

OSX

onnx2tf version number

1.25.12

onnx version number

1.16.2

onnxruntime version number

1.19.2

onnxsim (onnx_simplifier) version number

0.4.36

tensorflow version number

2.17.0

Download URL for ONNX

https://drive.google.com/file/d/18IkWTktKG2wTvtjnrjsNodD9ri4jmLZE/view?usp=drive_link

Parameter Replacement JSON

{}

Description

1. I am studying model conversion as part of my Research development at the university.
2. Need to leave the option to use the same input data as for the onnx model for tflite model. simple example is (1,3,224,224) (NCHW), but my obtained (from onnx) tflite model have NHWC input and no transpose layer.
3. I tried to use flags when converting -k
4. This task takes into account that when replacing the model, it will not be possible to change the input data

[PINTO0309]

I'm sorry, but I don't understand what you're expecting.

onnx2tf -i conv_model.onnx

onnx2tf -i conv_model.onnx -kat input

[pipipyau]

Thank you for your reply! I used the -k flag incorrectly. And what about the output - is it possible to keep the shape there by adding a transposition (1, 16, 224, 224)?

[PINTO0309]

WIP: https://github.com/PINTO0309/onnx2tf/tree/rep_param_conv WIP: https://github.com/PINTO0309/onnx2tf/commit/1ab3f1177b35db34e44463ee6d02c64140679404

• replace_conv.json

  {
  "format_version": 1,
  "operations": [
    {
      "op_name": "wa/conv/Conv",
      "param_target": "outputs",
      "param_name": "output",
      "post_process_transpose_perm": [0,3,1,2]
    }
  ]
  }

• test

  onnx2tf -i model_conv.onnx -kat input -prf replace_conv.json

  

[pipipyau]

Thanks for your work✨

[PINTO0309]

Fix: https://github.com/PINTO0309/onnx2tf/releases/tag/1.26.0