possibility for adopting other complex data augmentation strategies

[wangren97]

Dear authors,

Thanks for your interesting work.
I have a question about the data augmentation strategies used in multi-view object detection cases.
This paper discusses the geometric data augmentation.
Is it possible to extend this idea to other complex data augmentation strategies, such as mosaic, mix-up, and copy paste?
Based on my understanding, I think the above augmentation strategies cannot be adopted since they cannot preserve the geometric relations among the views. Is it correct?

Thanks