Extend ONNX Frontend with Shape-15 operator

[mitruska]

Context

Neural networks are graphs consisting of nodes called operators. Each operator corresponds to a mathematical function, usually described in framework's documentation or an AI standard, such as ONNX.

OpenVINO ONNX Frontend is a component responsible for working with ONNX graphs and requires implementation of different ONNX operators in order to use ONNX models.

This task requires extending OpenVINO ONNX Frontend with Shape-15 (ONNX Shape operator since ONNX Opset 15).
ONNX Shape operator has been extended with additional "start, end" attributes, what should be reflected in the conversion logic.
Currently those attributes are ignored by ONNX FE, and ONNX Shape is converted directly to ov::ShapeOf,
but it can be supported at the ONNX FE conversion step, with additional ov::Slice layer.

Necessary help will be provided by ONNX Fronted team.

Operator specification

Operator details can be found in ONNX Operators.

Todo list

1. Update .hpp and .cpp files:
   /openvino/src/frontends/onnx/frontend/src/op/shape.hpp
   /openvino/src/frontends/onnx/frontend/src/op/shape.cpp
2. Prepare an implementation of this operator in form of a function. It should be placed in set_15 namespace.
3. Register the next version of the Shape operator: REGISTER_OPERATOR("Shape", 15, shape); in ops_bridge.cpp while keeping alphabetical order
4. Create ONNX Shape-15 test model(s) in ONNX models directory. OpenVINO test infrastructure then converts prototxt files to ONNX models - you will use those models later in tests. Double check the ONNX opset version of the model that should be at least 15.
5. Add tests covering all use cases here

More details in adding operators to ONNX Frontend guide

Resources

• Contribution guide - start here!
• Guides on ONNX Frontend architecture and ONNX Frontend tests

Example PRs

• ONNX Unique op support #15076
• ONNX CastLike operator tests #14936
• [ONNX FE] Extend ONNX FE for operation GenerateProposals #12510

Tips

Proper implementation should enable the following tests:

openvino/src/frontends/onnx/tests/__init__.py

Line 138 in a6e5f6b

xfail_issue_63138 = xfail_test(reason="Missing ONNX Shape-15 support")

openvino/src/frontends/onnx/tests/tests_python/test_backend.py

Lines 322 to 329 in a6e5f6b

	xfail_issue_63138,
	"OnnxBackendNodeModelTest.test_shape_end_1_cpu",
	"OnnxBackendNodeModelTest.test_shape_end_negative_1_cpu",
	"OnnxBackendNodeModelTest.test_shape_start_1_cpu",
	"OnnxBackendNodeModelTest.test_shape_start_1_end_2_cpu",
	"OnnxBackendNodeModelTest.test_shape_start_1_end_negative_1_cpu",
	"OnnxBackendNodeModelTest.test_shape_start_negative_1_cpu",
	),

Also for compatibility API:

openvino/src/bindings/python/tests_compatibility/test_onnx/test_backend.py

Line 330 in a6e5f6b

xfail_issue_63138,

Contact points

@mitruska
@p-wysocki
@gkrivor
Don't hesitate to reach out, we're here to help!

[hotcheetofiend]

@mitruska If no one has been assigned yet, could I work on this issue?

[p-wysocki]

Hello @hotcheetofiend, thanks for taking a look! I assigned you.

[p-wysocki]

Hi @hotcheetofiend, are you still working on it? I'm updating the tasks' statuses.

[hotcheetofiend]

Hi @hotcheetofiend, are you still working on it? I'm updating the tasks' statuses.

Yes; sorry! I will try to get it done by Friday.

[mitruska]

Hello @hotcheetofiend, let us know if you find this task challenging.
We are here to help 💡

[p-wysocki]

Hi @hotcheetofiend, are you still working on this or can we reopen the issue for pickup by other contributors?

[BharathSatheeshKumar]

.take

[github-actions]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[BharathSatheeshKumar]

May I get guided by anyone. Like i just want a simple description on the issue so that i can work on it.
Even tho im new i can give results

[mitruska]

Hello @BharathSatheeshKumar, thank you for taking this issue, the idea is explained in the issue description above:

This task requires extending OpenVINO ONNX Frontend with Shape-15 (ONNX Shape operator since ONNX Opset 15).
ONNX Shape operator has been extended with additional "start, end" attributes, what should be reflected in the conversion logic.
Currently those attributes are ignored by ONNX FE, and ONNX Shape is converted directly to ov::ShapeOf,
but it can be supported at the ONNX FE conversion step, with additional ov::Slice layer.

In other words, the goal is to add support for the start and end attributes of the ONNX Shape-15, by conversion to Slice(ShapeOf(data), start, end, 1, 0) subgraph.
Also please take a look at the TODO steps and example PRs, where related files are mentioned.

[BharathSatheeshKumar]

Thank You.
And Do I Get a deadline for this so that i can give my best at anycost

[mitruska]

Happy to help 🚀
Currently there is no specific deadline for this task, but you can share your approximate time estimation and let us know about any troubles.

[BharathSatheeshKumar]

I've updated the hpp and cpp files but cant create a PR. Help me out
I'll let you know about the deadline within a few days of working with this issue

[mitruska]

Step by step contribution guide, including how to create a Pull Request [PR], can be found in the Resources section of the issue description, please follow:

blog post on contributing to OpenVINO

[p-wysocki]

Hello @BharathSatheeshKumar, do you need any help?

Just yesterday our CONTRIBUTING.md has been updated with a technical guide - I highly recommend checking it out. :)

[BharathSatheeshKumar]

Yea sure will check that out and will let you guys know soon.
And i still have problems in creating a pr for this base even tho I followed what @mitruska sent

[BharathSatheeshKumar]

PR link?

[p-wysocki]

Hello @BharathSatheeshKumar, could you please let us know what issues you're having?

[BharathSatheeshKumar]

Just now sorted the problem thanks....and will update within a day or so

[BharathSatheeshKumar]

i cant access the main file any ideas?

[p-wysocki]

Hi @BharathSatheeshKumar, I think we may need more details. Could you please join Intel DevHub Discord and ask your questions there? It will be easier for both sides.

[p-wysocki]

Due to long inactivity I'm unassigning the issue for other contributors. If you're still working on this task @BharathSatheeshKumar please let us know, you'll be reassigned.

[AlexFierro9]

.take

[github-actions]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[BharathSatheeshKumar]

sorry for the delay tho
let me work with @AlexFierro9 for the next project

[p-wysocki]

Hello @AlexFierro9, are you still working on that issue?

[MeeCreeps]

.take

[github-actions]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[MeeCreeps]

Hi @p-wysocki , I have raised PR #23492 for this issue, please let me know if any modifications are needed.

[mlukasze]

thanks @MeeCreeps!
@gkrivor will take a look soon

[mitruska]

Hello @MeeCreeps your PR has been reviewed, approved and merged 🚀
Thanks, Looking forward to your further contribution!