/style-transfer

Implementations of style transfer methods using torch

Primary LanguageLua

Style transfer

Two style transfer methods, based on:

  1. Leon A. Gatys, Alexander S. Ecker, Matthias Bethge
  2. Justin Johnson, Alexandre Alahi, Li Fei-Fei

  • The implementations are created in torch, and use a pre-trained vgg network, taken via the method described in the Model Zoo

  • In order to run an experiment, run:

    • run_experiment_set.py <experiment.csv> [--show-images] [--noterm]
    • --show-images renders the images at each iteration of optimization
    • --noterm runs the experiments in the file in the background, independent of the terminal they were started from.

  • Each line in the experiment.csv file should have the following format:

    • train_method,experiment_label,regularize_term1,regularize_term2,regularize_term3,content_image,style_image,results_folder,save_every_count,content_convolution_layer,gpu_number,style_convolution_layers