Hi, can anybody tell me how to replace the batch norm layer in resnet50 with group norm layer using pytorch?
To start I would take a look at the existing reference implementations in torchvision torchvision.models.resnet — Torchvision main documentation
and see if you can simply modify the handling of the norm_layer argument to handle GroupNorm (since it requires specifying the number of groups in addition to the number of channels)l