Update ONNX submodule to support opset 18

[thiagocrepaldi]

This PR has the goal of updating ONNX to a particular commit and allow ORT to execute ONNX graphs with opset 18

PyTorch is about to merge the same change. However, because PyTorch depends on ORT for unit testing, ORT also must update its ONNX submodule for their pipelines to succeed. Otherwise, ORT will not recognize the new ops PyTorch ONNX converter adds, such as Col2Im

[thiagocrepaldi]

@skottmckay could you look into this one?

[thiagocrepaldi]

@snnn This PR changes ONNX from the latest stable release to a commit. No code/dataset change, but several tests consistently failed with

[E:onnxruntime:Default, provider_test_utils.cc:1045 Run] Resolve failed with status: Node (node1) Op (Resize) [ShapeInferenceError] Either `sizes` or `scales` must be provided, but not both of them
2022-10-03T18:03:46.8850993Z /onnxruntime_src/onnxruntime/test/providers/provider_test_utils.cc:1047: Failure
2022-10-03T18:03:46.8851477Z Value of: status.IsOK()
2022-10-03T18:03:46.8851741Z   Actual: false
2022-10-03T18:03:46.8851984Z Expected: true
2022-10-03T18:03:46.8852340Z Node (node1) Op (Resize) [ShapeInferenceError] Either `sizes` or `scales` must be provided, but not both of them

I wonder if the reason is because of the step 4 at onnxruntime/How_To_Update_ONNX_Dev_Notes.md, which states:

If you are updating ONNX from a released tag to a new commit, please ask Changming (@snnn) to deploy the new test data along with other test models to our CI build machines. This is to ensure that our tests cover every ONNX opset.

If that is the case, could you help updating the test data?

[snnn]

If you are updating ONNX from a released tag to a new commit, please ask Changming (@snnn) to deploy the new test data along with other test models to our CI build machines. This is to ensure that our tests cover every ONNX opset.

If that is the case, could you help updating the test data?

It needs something like this: #13062

And your PR needs an approval from the admins of this repo. I'm not one of the admins.

[skottmckay]

Looks like the ONNX type/shape inferencing started enforcing the Resize spec more strictly for opset 13 which is causing some of the test failures. Tests should be updated to only provide either 'scales' or 'sizes'

onnx/onnx#4448

[  FAILED  ] ResizeOpTest.ResizeOpLinearDownSampleTest_tf_crop_and_resize
[  FAILED  ] ResizeOpTest.ResizeOpLinearDownSampleTest_4DBilinear1_WithSizes
[  FAILED  ] ResizeOpTest.ResizeOpLinearDownSampleTest_2DBilinear_pytorch_half_pixel
[  FAILED  ] ResizeOpTest.NhwcResizeOpLinearDownSampleTest_4DBilinear_pytorch_half_pixel_uint8
[  FAILED  ] ResizeOpTest.NhwcResizeOpLinearDownSampleTest_4DBilinear_pytorch_half_pixel_int8
[  FAILED  ] ResizeOpTest.ResizeOpLinearDownSampleTest_3DTrilinear_pytorch_half_pixel
[  FAILED  ] ResizeOpTest.ResizeOpNearestDownSampleTest_WithSizes
[  FAILED  ] ResizeOpTest.ResizeOpNearestUpSampleTest_WithSizes_CeilMode
[  FAILED  ] ResizeOpTest.ResizeOpNearestUpSample5dTest_WithSizes_CeilMode
[  FAILED  ] ResizeOpTest.ResizeOpNearestUpSample_Nearest2xOptimization_Sizes
[  FAILED  ] ResizeOpTest.ResizeOpCubicUpSampleTest_MultiChannel

[snnn]

It needs an approval from @onnxruntime-admin

[thiagocrepaldi]

It needs an approval from @onnxruntime-admin

Yes, @faxu mentioned it was ok to merge this post 1.13 release.

[faxu]

This should be updated to the latest released version of onnx (1.13) once it's available.

[thiagocrepaldi]

Duplicate of #14279