Closed UnderController closed 4 years ago
hello, a greate work~~~ I have a question why you fix encoder bn when training ? have you do some comparion
Hi, I have been told to fix bn when the batch size is small (<=16), especially when a imagenet-pretrained model is used.
thank you for reply, i may do as your suggestion
hello, a greate work~~~ I have a question why you fix encoder bn when training ? have you do some comparion