About the training parameters of DOTA-v2.0 #13
Description
Thank you for your open source work. I would like to ask about the training parameters of DOTA-v2.0, and what are the batch_size (42 or 22), epoch (12 or 36), and learning rating of the results reported by RHINO in the paper (59.26 and 60.72)?
In addition, for the result of 68mAP on DOTA-v1.0, what parameters were used to train Deformable DETR?