/ArtistGAN

ArtistGAN

MIT LicenseMIT

ArtistGAN

ArtistGAN is a tool used to generate human face images from their descriptions given in textual format.

FlowChart

Flow Chart ArtistGAN

Data Set

Images:

  1. Labelled Faces in the Wild (LFW)
  2. CelebA

Descriptions:

Face2Text for image descriptions

Architecture

Architecture ArtistGAN

  1. Text Encoder

  2. Conditioning Augmentation

  3. Generative Adversarial Networks (GANs)

  4. Feature Tuning

ArtistGAN Outputs

Outputs ArtistGAN

Experimental Setup

  • Programming Language: Python(PyTorch)

  • Libraries: Numpy, Pillow, Torch, TorchVision, etc.

  • System Configuration:

    • Processor: Intel Xeon @ 2.30 GHz
    • RAM: 16gb
    • GPU: Google Colab and Kaggle - Tesla T4 and K80

For more details, refer to this presentation and thesis report.