Using TransposeOptimizer breaks model with split optimizer opset >= 13 error

[CarlPoirier]

Describe the bug
I have an onnx model on which I want to run the onnx-optimize.py script. All the optimizers work fine except the TransposeOptimizer. When I use it, my onnx model is not working anymore, with onnxruntime at least. I get this error :

[ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Split node. Name:'Split_12' Status Message: D:\a\_work\1\s\onnxruntime\core\providers\cpu\tensor\split.cc:160 onnxruntime::Split::ComputeImpl split_tensor->Shape().NumDimensions() == 1 was false. An split tensor must be a vector tensor.

Urgency
None, for now I deactivated this optimizer and I'm using all the others.

System information

• OS Platform and Distribution: Windows 10
• Tf2onnx version: 1.10.1
• Tensorflow Version: 2.5.2
• Python version: 3.9.9
• ONNXRuntime: 1.9.3

To Reproduce
test_models.zip

The above zip contains four files:

• test-run.py : script to run the model using onnxruntime
• test_model2_small.onnx : starting model which works
• test_model2_small_broken.onnx : optimized model which does not work anymore
• test_model2_small_ok.onnx : optimized model for which I have deactivated only the transpose optimizer and it still works

I'm running the onnx-optimize script this way:

python onnx-optimize-orig.py --input test_model2_small.onnx --output test_model2_small_broken.onnx 2022-05-17 19:13:24.283642: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll 2022-05-17 19:13:27,704 - INFO - Optimizing ONNX model 2022-05-17 19:13:27,732 - INFO - After optimization: Const -1 (7->6), Identity -1 (1->0)

Then I run the model with any image, this way:

python test-run.py

If you need any more information, let me know!

[hwangdeyu]

@CarlPoirier The latest tf2onnx verison 1.12 has been released. I think your model can be optimized as expected. Thanks!