Feature Request: Support for GECO2 (Generalized Object Counting & Segmentation) model in X-AnyLabeling

[dongri-liao]

Hello,

First, thank you for building such a powerful annotation tool — it has significantly improved the efficiency of labeling workflows, especially for detection and segmentation tasks.

I would like to propose support for the GECO2 model:
https://github.com/jerpelhan/GECO2

GECO2 introduces a strong interactive capability where users can select one object and automatically propagate annotations to similar instances. This is extremely valuable for scenarios involving:

• Dense object layouts
• Industrial inspection datasets
• Repetitive structures
• Few-shot annotation workflows

Compared to traditional detection-assisted labeling, this approach can greatly reduce manual effort.

Thank you again for your work — X-AnyLabeling is becoming one of the most practical annotation tools available today.

Looking forward to your thoughts!

[CVHub520]

Hi @dongri-liao, thanks for the feature request and the detailed use cases.

We did consider integrating GECO2 earlier, but in our internal evaluations (accuracy + generalization across diverse datasets), GECO2 performed noticeably worse than SAM3. Also, SAM3 already supports fine-tuning on custom data, so we’re currently not fully convinced what extra value GECO2 would bring to X-AnyLabeling.

If you can share one or two concrete examples where GECO2 provides a clear advantage, we can re-evaluate the integration priority and the best way to support that capability (either via GECO2 or via improving SAM3-based interaction).