I want to know that does nn.GroupNorm have a big benefit for semantic segmentation and the comparison of the cost for memory with nn.BatchNorm.
it uses BN to implement, so similar in terms of memory
oh thank you.But what is the result on segmentation
I don’t know. You can try yourself.
For me it uses about 30% more memory than BatchNorm. Can anybody confirm that? Is that supposed to be the case?