/image-dataset-generator

Algorithmically generated image dataset

Primary LanguagePythonMIT LicenseMIT

Algorithmically generate images containing specific invariances

Inspired by: Albert Sanchez Lafuente

Create image datasets of arbitrary size (compatible with Julia's Flux) containing any user-specified objects and backgrounds. These objects printed on the background will have user-specified symmetries in-place. Made for teseting GDL frameworks.

Description of the project files:

  • backgrounds: folder containing background images
  • characters: folder containing character images
  • generated: folder where generated images are saved
  • data.csv: example of a generated dataset containing the characteristics of each generated image.
  • main.py: project code