search

luanfujun / deep-painterly-harmonization

Code and data for paper "Deep Painterly Harmonization": https://arxiv.org/abs/1804.03189
6.08k stars 628 forks source link

Conflicts between paper and the code #29

Closed HenryJia closed 4 years ago

HenryJia commented 6 years ago

Hi I'm reading through your neural_paint.lua and I'm noticing a lot of conflicts At https://github.com/luanfujun/deep-painterly-harmonization/blob/master/neural_paint.lua#L151 You use conv5_1 as your reference layer, yet in the paper, you say that conv4_1 was the optimum

At line 155, you use the resized style image as your reference guide, but in the paper, you again say that conv4_1 is the reference layer

At line 168, it appears you are storing the un-matched style features and not the matched features and are completely ignoring the matched and then consistency refined features

Can you explain these conflicts?

Thanks

tombohub commented 4 years ago

@HenryJia do this conflict affect the quality of final image?

HenryJia commented 4 years ago

I've no idea, because I was originally using this as a reference code to write my own implementation, but I've been unable to do so due to the vagueness of certain parts in the paper, my unfamiliarity with torch and the conflicting parts here

luanfujun commented 4 years ago

line 63: local layers = string.format('relu1_1,relu2_1,relu3_1,relu4_1'):split(",") So conv5_1 wasn't used, it was just some comments I didn't clean up.

luanfujun commented 4 years ago

@HenryJia Maybe try some other people's implemetation? https://github.com/sarahmfrost/manumorph https://github.com/sgugger/Deep-Learning/blob/master/DeepPainterlyHarmonization.ipynb

HenryJia commented 4 years ago

Considering it's been 2 years since I opened the issue, I'm currently not trying to implement this anymore