Fix #23553: [OP CONFORMANCE][TEMPLATE] Interpolate test fails in op c…

[runzhangDL]

The test case failed because it uses a channel-last input pattern, while the linear_onnx_func() assume that all input data have the batch and channel in their first two dimensions. I've added extra condition check in the channel-last case and made it valid to continue the following operations.

Note: The implementation for channel-last input should be different from current one. However, after the input check has been changed, the test case passed already, so I didn't change the following implementation. Please let me know if there are concerns with this and I am happy to add extra adjustment.

[praasz]

build_jenkins

[mlukasze]

hey @runzhangDL please correct formatting to bass clang tests

[runzhangDL]

hey @runzhangDL please correct formatting to bass clang tests

Sure, will update soon. Do I need to fix other failed checks?

[t-jankowski]

hey @runzhangDL please correct formatting to bass clang tests

Sure, will update soon. Do I need to fix other failed checks?

I looks like these changes cause below failures - copied from Linux Static CC (Ubuntu 22.04, Python 3.11, Clang) logs. To reproduce it locally use ./bin/intel64/Debug/ov_onnx_frontend_tests

C++ exception with description "Exception from src/inference/src/cpp/infer_request.cpp:223:
Exception from reference/interpolate.hpp:430:
Axes are not correct!

[  FAILED  ] 11 tests, listed below:
[  FAILED  ] INTERPRETER.onnx_resize11_empty_constant_as_input
[  FAILED  ] INTERPRETER.onnx_resize10_down_scales_const_linear
[  FAILED  ] INTERPRETER.onnx_resize10_up_scales_const_linear
[  FAILED  ] INTERPRETER.onnx_resize11_down_scales_linear_asymmetric
[  FAILED  ] INTERPRETER.onnx_resize11_up_scales_linear_asymmetric
[  FAILED  ] INTERPRETER.onnx_resize11_up_sizes_linear_asymmetric
[  FAILED  ] INTERPRETER.onnx_resize11_down_sizes_linear_pytorch_half_pixel
[  FAILED  ] INTERPRETER.onnx_upsample6_bilinear_infer
[  FAILED  ] INTERPRETER.onnx_upsample8_linear_infer
[  FAILED  ] INTERPRETER.onnx_upsample9_scales_const_linear_infer
[  FAILED  ] INTERPRETER.onnx_empty_initializers_handling

[runzhangDL]

NTERPRETER.onnx_resize11_up_scales_linear_asymmetric

Ok I will try to look into it. It might be easier if I can see the test cases detail to debug why they failed.

[iefode]

@mitruska Could you please take a look?

[runzhangDL]

Still working on this, will require a bit more time. It will help if someone can confirm whether we need to add support for batch-last case on interpolation, which might cause failure of other test cases? Or we just remove the interpolation case where batch was put in the last dimension?

[runzhangDL]

Sorry for the delay, I was distracted by my courseworks, and will look into this issue soon.

[mitruska]

Hello @iefode and @runzhangDL, I can see the work is in progress;)
For any changes in the Interpolate reference implementation, please add corresponding unit test to:

openvino/src/plugins/template/tests/functional/op_reference/interpolate.cpp

Line 564 in a5cf8b3

std::vector<InterpolateV4TestParams> generateParamsForInterpolate_v4_linear_onnx() {

[mlukasze]

build_jenkins

[runzhangDL]

I am still working on the interpolation implementation, should be done soon

[iefode]

@mitruska Hi Mitrus, I've added the channel-last test case at the bottom. It is basically the same to the previous test case but move the spatial axis from (2,3) to (1,2) and channel at the last. The input and expected output should be the same, because only spatial dimension was changed. The test case run successfully in my local machine. Please take a look if it is convenient and let me know any other edits are needed

@mitruska Could you please take a look?

[mitruska]

This test case is okay, but it would be good to add few more tests,
especially with the last dimension not equal to 1.

[mlukasze]

hey @runzhangDL, will you find a time to finish this PR?

[github-actions]

This PR will be closed in a week because of 2 weeks of no activity.

[mlukasze]

@t-jankowski & @mitruska
could you take a look, please?

[runzhangDL]

hey @runzhangDL, will you find a time to finish this PR?

Apology for the late reply. I will try to finish what I left now. I am currently not sure whether my implementation is correct, because the test code didn't work out to me.

[mlukasze]

thanks! let us check it

[github-actions]

This PR will be closed in a week because of 2 weeks of no activity.

[mlukasze]

build_jenkins

[github-actions]

This PR will be closed in a week because of 2 weeks of no activity.

[github-actions]

This PR was closed because it has been stalled for 2 week with no activity.