I was trying to implement universal style transfer in Pytorch based on this repository . This repository uses pre-trained torch models and then changes them to Pytorch models. But on some research I found out that torch models can be directly used in Pytorch by calling
model.forward function manually. So replaced all the calls to forward function with the torch’s pre-trained models. But I’m getting different results. The style is barely transfered from style to content image. The output is just a bit distorted version of the input and there’s no style transfer taking place.
What could be the reason for this behaviour ? Is it natural of torch model to misbehave in pytorch?