Clarification on the Implementation of Object-Aware Self-Attention (OASA)

[hjblearning]

Hi jiaqi,

First of all, congratulations on the acceptance of your excellent work "An Incremental Unified Framework for Small Defect Inspection" at ECCV 2024! The proposed Object-Aware Self-Attention (OASA) module is particularly inspiring, and I believe it addresses a key challenge in small defect detection.

I’m currently trying to dive deeper into your code to understand the implementation. However, I couldn’t locate the specific code for the Object-Aware Self-Attention module (e.g., in the model/ or layers/ directories). Could you clarify whether this part is included in the current release, or if it will be added in future updates?

Any guidance or pointers to the relevant code sections would be greatly appreciated! 😊

Thanks for open-sourcing your work—it’s a valuable contribution to the community!

Best regards.

[jqtangust]

please see
https://github.com/jqtangust/IUF/blob/master/models/reconstructions/ViT.py

Its outputs include two parts: middle features and classification vector.