Implemented Conv2D, Depth2Space, and Resize. Is anyone interested in merging these changes back?

[linzj]

Describe the issue

I have partially implemented Conv2D, Depth2Space, and Resize, not fully tested. These implementations have been tested on an Android device and are running my model without issues. What should be the next steps? Is anyone interested in collaborating to improve these implementations and eventually merge them into the main branch?
My fork is https://github.com/linzj/onnxruntime, branch is feature/ops-svr.

To reproduce

No content.

Urgency

No response

Platform

Android

OS Version

15

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

afd3e81

ONNX Runtime API

C++

Architecture

ARM64

Execution Provider

Other / Unknown

Execution Provider Library Version

No response

[snnn]

Welcome to submit PRs to us.

[linzj]

Welcome to submit PRs to us.

Thanks for the feedback! Should I complete the full implementation of the Conv operator (including all functionalities and test cases) before submitting a PR, or would you prefer an early review of the current partial implementation?

[snnn]

Is it for webgpu ?

[linzj]

Is it for webgpu ?

yes.

[xhcao]

Hi, @linzj There was an conflict that I had already implemented the resize operator on webgpu backend. #23780
Sorry for the conflict, and I took this task before the Chinese Spring Festival, you could review my code if possible, thanks.

[linzj]

no problem. Mine version has passed the test and run our video super resolution model well. xhcao ***@***.***> 于 2025年2月21日周五 下午5:05写道：

…

Hi, @linzj <https://github.com/linzj> There was an conflict that I had already implemented the resize operator on webgpu backend. #23780 <#23780> Sorry for the conflict, and I took this task before the Chinese Spring Festival, you could review my code if possible, thanks. — Reply to this email directly, view it on GitHub <#23471 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNJCFMF6PKY5LRVQ7WMYKD2Q3T55AVCNFSM6AAAAABVW3I5BCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNZTHE4DEMBZGU> . You are receiving this because you were mentioned.Message ID: ***@***.***> [image: xhcao]*xhcao* left a comment (microsoft/onnxruntime#23471) <#23471 (comment)> Hi, @linzj <https://github.com/linzj> There was an conflict that I had already implemented the resize operator on webgpu backend. #23780 <#23780> Sorry for the conflict, and I took this task before the Chinese Spring Festival, you could review my code if possible, thanks. — Reply to this email directly, view it on GitHub <#23471 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNJCFMF6PKY5LRVQ7WMYKD2Q3T55AVCNFSM6AAAAABVW3I5BCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNZTHE4DEMBZGU> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[linzj]

Oh. I noted that you disabled the nhwc support. That's unexpected. Check mine version please. Mine has the nhwc supported. We must support nhwc to make conv2d vectorization available.