Skip to content

AbhinavS99/Generative-Adversarial-Networks-Based-Text-to-Image-Generation

Repository files navigation

Generative-Adversarial-Networks-Based-Text-to-Image-Generation


To view the literature survey, the model architecture details, the background mathematics and the outputs - Report.pdf


To test the code

  1. Open cmd/terminal in cwd
  2. run the command 'pip install -r requirements.txt'
  3. run the command 'python downloader.py' to download the dataset
    • If this does not work then download the dataset from the link and add the dataset in 'dataset' directory.
    • The size of dataset is 1.87GB.
  4. run the command 'python tester.py' to get results for both baseline models

Directory Structure

  1. dataset → has the flower dataset in hdf5 format
  2. modelState → Stores the trained models
    • dcgan-cls.pth → DCGAN-CLS model
    • gan-cls-int.pth → GAN-CLS-INT model
  3. results → stores sample results i.e. expected v/s generated images
  4. src → the training process

Note :

  • This code uses pytorch for deep learning based implementations.
  • Please install pytorch according to the specification of your machine.
  • The code will only run on device which has 'cuda' enabled.
  • Do not alter directory structure as paths are relative for downloading and extraction of dataset.

Releases

No releases published

Packages

No packages published

Languages