Why use nearest neighbor to match two histograms in neural style transfer with histogram loss.

gabrieltseng / datascience-projects

A collection of personal data science projects

57 stars 22 forks source link

Why use nearest neighbor to match two histograms in neural style transfer with histogram loss. #1

Open baldFemale opened 5 years ago

baldFemale commented 5 years ago

It seems that nearest neighbor brings more fluctuation. Did you try other methods such as bilinear interpolation to match the histograms?

gabrieltseng commented 5 years ago

Hi!

I haven't tried other methods, but I was having trouble making the histogram matching work well; alternative methods definitely make sense to experiment with.