support best Low-shot Object Counting model geco

[talebolano]

Search before asking

• I have searched the X-AnyLabeling issues and found no similar feature requests.

Description

In the FSC147 dataset for the Low-shot Object Counting task, A Novel Unified Architecture for Low-Shot Counting by Detection and Segmentation has achieved the best performance. By introducing this model, it is possible to achieve an effect similar to TReXLabel. I have completed the ONNX conversion of this model and implemented its backend code. However, I am not sure how to pull request. Because I see that you usually place the ONNX model in the release for publication.

Use case

No response

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[CVHub520]

Hi @talebolano, 😄

Thank you for your contribution and for sharing your impressive results with the "geco" model! It's great to hear about your success with ONNX conversion.

To discuss this further and help you with the pull request process, please add me on WeChat: ww10847. When you add me, please include the tag "X-AnyLabeling #863" so I know it's you. We can then chat about the best way to integrate your model.

Thanks again!

[talebolano]

@CVHub520 I tried searching for ww10847 on WeChat, but it showed that the user does not exist.

[CVHub520]

My apologies, I made a typo. I meant to say ww10874.