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).

[xian-stack]

In very dense scenes (>200 objects), SAM3 fails completely! In contrast, GECO2 works very well in this setting, as we tested. So while GECO2 is truly much less general, it clearly outperforms SAM3 in dense-object regimes, which is the gap it could fill.

[CVHub520]

@xian-stack done.

[CVHub520]

Hi @dongri-liao, the latest version now supports GeCo2. Please refer to the example documentation here: https://github.com/CVHub520/X-AnyLabeling/blob/main/examples/counting/geco2/README.md