DFT bloack seems only can training on batchsize=1 ?

csxmli2016 / DFDNet

Blind Face Restoration via Deep Multi-scale Component Dictionaries (ECCV 2020)

913 stars 217 forks source link

DFT bloack seems only can training on batchsize=1 ? #65

Open miaoYuanyuan opened 3 years ago

miaoYuanyuan commented 3 years ago

such as : get the part location of the vggFeat. F.inteplate, adaptivate_instance_norm, the feature matches, these process only can run on batch size=1 ?

chengqianma commented 3 years ago

such as : get the part location of the vggFeat. F.inteplate, adaptivate_instance_norm, the feature matches, these process only can run on batch size=1 ?

Hi, I also found these problems. Have you ever tried to train the network with batch size 1. Thanks.