Thought about Box corruption.

[imabackstabber]

Looks like only scaling operation is done in Box corruption for training stage, however by only scaling them, the centers of the boxes remain the same, but in inference stage, the boxes are generated totally randomly, which means that you have to adjust the center(maybe drastically) to make a box center right.
Is there any ablation studies about the Corruption method? (or they're already in the paper, i just missed it?)

[ShoufaChen]

Hi,

As illustrated in these lines,
both center coordinates and sizes (width and height) are scaled.

[imabackstabber]

Hi,

As illustrated in these lines, both center coordinates and sizes (width and height) are scaled.

Code is very clear.Thanks for the prompt reply! I'm gonna close the issue.