/DramaGAN

Long Image Sequence Generation from Text

Primary LanguagePython

DramaGAN

Long Image Sequence Generation from Text

Usage

  1. Download Dataset
  • Download Structure
  -- dataset
      |- Drama_Dataset
  		|- AnotherMissOh_QA
  		|- AnotherMissOh_images
  		|- AnotherMissOh_subtitles.json
  1. Set Config.py train_settings
  • "load_pretrained" should be None if training from beginning. If not, set {"epoch", "iteration"}, both zero-indexed. If pretrained does not exist, will train from beginning.
  1. Train
python train.py

StoryGAN Result

  • epoch: 59
  • iteration: 500
  • dataset mode: train

Input description

Dokyung is sitting on the chair. Dokyung texted a message to Haeyoung1. Haeyoung1 made surprised Dokyung.

Output

Real Images: Randomly selected n-number of frames from scene
Generated Images: Generated Images using Story GAN

Real Images Generated Images