Generating Images from text

In this, we have worked on generating bird images conditioned on the binary segmentation masks and text embeddings

Dependencies

  • Tensorflow >= 1.0
  • Scipy
  • Scikit-image
  • torchfile

Preprocessing

  • Download the birds caption data in Torch format

  • Download the birds image data

  • Extract and place the images and segmentation masks in a directory CUB

  • Run the script to create tfrecords for training the model:

    python preprocess_data.py
    

Training

  • For training the model:

    python main.py --batch_size=64 --output_size=128