Change the size of the region features

[GabrieleFerrario]

Hi, Is it possible to change the region feature size from 2048 to 500 for finetuning captioning model? If yes, what should I change?
Alternatively is it ok to use padding with 0?

I would like to finetune the captioning model by changing the object detection model and using one that returns region features with size smaller than 2048.
@xjli @pzzhang @xiaoweihu

Thanks!