Skip to content

Latest commit

 

History

History
45 lines (37 loc) · 2.28 KB

notes.md

File metadata and controls

45 lines (37 loc) · 2.28 KB

long-term ideas:

try to use autoencoder with embedding on individual anime, then PCA and the widget like in https://www.youtube.com/watch?v=wXWKWyALxYM?t=286

try to go through tips in: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/docs/tips.md try the StyleGAN paper https://arxiv.org/abs/1812.04948 and also try the original neural style paper take a look at DiscoGAN take a look at https://www.cs.ru.nl/bachelors-theses/2018/Robin_Elbers___4225678___On_the_replication_of_CycleGAN.pdf take a look at CartoonGAN take a look at Deep Analogy (Visual Attribute Transfer through Deep Image Analogy) take a look at CNNMRF https://arxiv.org/pdf/1601.04589.pdf

ideas:

todo:

  • add the hyper-parameter tuning and orchestration
  • use https://github.com/automl/ConfigSpace and SMAC
  • integrate TFGAN's inception score and FID
  • create original neural style transfer - should work and learn more quickly
  • compare PCA, autoencoder and variational autoencoder for individual animes
  • compare random sampling style image from anime with selecting components from dimension reduction as style image

lecture notes

video for CartoonGAN https://www.youtube.com/watch?v=r4zzhN8Aibw&ab_channel=ZivZone to lecture - faces from it mention in lecture https://github.com/luanfujun/deep-painterly-harmonization (the most funny of them) definitely describe https://www.gwern.net/Faces, and https://www.thiswaifudoesnotexist.net/ try something with trained network from that blog mention waifu2x mention makegirlsmoe

for installing lua - did not get that working: install lua from https://github.com/rjpcomputing/luaforwindows/releases for installing torch: install cmake manually by conda install -c anaconda cmake clone https://github.com/BTNC/distro-win.git then follow win instructions, but for beware problems with outputting things to files download zips manually, there is problem with them