Dongliang Chang
Dongliang Chang
@auroua Did you find a solution?I also have this problem。
> Hello @dongliangchang , good work. May I ask what applications your model serve? Fine-grained birds/airplane/car classification. e.g., Identifying a bird as White_eyed_Vireo or Whip_poor_Will
@fengyuentau
> > > 感谢您的回复。 > 对于您说的“MCLoss的权重更小，造成的影响更小”，我对此的理解是使用pretrained模型时，由于最后有2048个通道，强行将类别对应的通道数放大到适配，但是对于某个类别并没有这么多的判别性区域，故导致了MCLoss的权重相对更小 你说的是其中一个原因。另一个原因是：MC-Loss会改变特征的分布，这与在pre-trained模型上进行fine-tune的思想违背。因此MC_loss在用pre-trained模型微调时，性能提升有限（在使用pretrained模型时，MCLoss的权重很小）。
Fix a typo.
Thank you very much for pointing out this error. You are absolutely correct, and we appreciate the accurate information you provided. We will promptly make the correction, and the order...