Skip to content

Style transfer with Style Attentional Networks implementation.

Notifications You must be signed in to change notification settings

Aflutter-Craft/Network

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Style Transfer with Style Attentional networks

Image style transfer using a style attentional network.

Training

  • install requirements with pip install -r requirements.txt
  • get your kaggle API key from here
  • start training with python3 train.py

Arguments

  • content_dir directory to content images (default to data/content)
  • style_dir directory to style images (default to data/style)
  • vgg pretrained vgg file (default to pretrained/vgg.pth)
  • lr leaning rate
  • lr_decay leaning rate decay
  • max_iter maximum number of iterations (default to 160,000)
  • batch_size number of image pairs in a batch
  • style_weight style images weight (default to 3.0)
  • content_weight content images weight (default to 1.0)
  • n_threads number of threads used for the data loader (default to 2)
  • save_model_interval interval to save models (default to 1000)
  • start iter number of iteration to start training from (in case training was stopped)

Testing

  • run test.py with appropriate arguments

Arguments

  • content content image
  • style style image
  • alpha style content trade-off
  • vgg path to pretrained vgg model
  • decoder path to pretrained decoder
  • transformer path to pretrained transfomer model

More results

References

About

Style transfer with Style Attentional Networks implementation.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages