search

naoto0804 / pytorch-AdaIN

Unofficial pytorch implementation of 'Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization' [Huang+, ICCV2017]
MIT License
1.08k stars 208 forks source link

Why add a 3*3*1*1 conv layer on top of official vgg-19 structure? #22

Closed CharlesNord closed 4 years ago

CharlesNord commented 5 years ago

Hi, I found that different from official vgg-19, in this project you added a conv layer of 3*3*1*1 to the input of vgg-19. Could you please tell me the reason? Besides, how did you train this new version of vgg-19 and get the pretrained feature extractor?

naoto0804 commented 5 years ago

I adapt everything from the original implememtation, so I hope it should be the same..

diaodeyi commented 3 years ago

Hi, I found that different from official vgg-19, in this project you added a conv layer of 3*3*1*1 to the input of vgg-19. Could you please tell me the reason? Besides, how did you train this new version of vgg-19 and get the pretrained feature extractor?

Did you find out ?