/Face-Generation-using-GAN

In this project,generative adversarial networks is used to generate new images of faces.

Primary LanguageHTMLMIT LicenseMIT

Face-Generation-using-GAN

This project was developed as a part of Udacity's Deep Learning Nanodegree. In this project, i have created a General Advarsarial Networks from scratch using pytorch.

Getting Started

Just run all the cells in the ipynb notebook. Tune the hyper parameters for better accuracy. Get Data from here

Prerequisites

  • Python 3
  • Numpy
  • MatPlotLib
  • OpenCv
  • Pytorch

Install

  1. Download the project materials from our GitHub repository. You can get download the repository with
git clone https://github.com/vickipedia6/Face-Generation-using-GAN.git

  1. Download anaconda or miniconda based on the instructions in the Anaconda documentation.

  2. Create a new conda environment:

conda create --name deep-learning python=3
  1. Enter your new environment:
  • Mac/Linux: >> source activate deep-learning
  • Windows: >> activate deep-learning
  1. Ensure you have numpy, matplotlib, pandas, and jupyter notebook installed by doing the following:
conda install numpy matplotlib jupyter notebook
  1. Run the following to open up the notebook server:
jupyter notebook dlnd_face_generation.ipynb
  1. Execute all the cells in the code.

Project results

This project has met the following specifications:

  • The submission includes all required, complete notebook files.
  • All the unit tests in project have passed.
  • The function get_dataloader should transform image data into resized, Tensor image types and return a DataLoader that batches all the training data into an appropriate size.
  • Pre-process the images by creating a scale function that scales images into a given pixel range. This function should be used later, in the training loop.
  • The Discriminator class is implemented correctly; it outputs one value that will determine whether an image is real or fake.
  • The Generator class is implemented correctly; it outputs an image of the same shape as the processed training data.
  • This function should initialize the weights of any convolutional or linear layer with weights taken from a normal distribution with a mean = 0 and standard deviation = 0.02.
  • The loss functions take in the outputs from a discriminator and return the real or fake loss.
  • There are optimizers for updating the weights of the discriminator and generator. These optimizers should have appropriate hyperparameters.
  • Real training images should be scaled appropriately. The training loop should alternate between training the discriminator and generator networks..
  • There is not an exact answer here, but the models should be deep enough to recognize facial features and the optimizers should have parameters that help wth model convergence.
  • The project generates realistic faces. It should be obvious that generated sample images look like faces.
  • The question about model improvement is answered.

Losses

Generator vs Discriminator Loss :

Generated Faces

Built With

  • Python 3

License

This project is licensed under the MIT License - see the LICENSE file for details

Acknowledgments

  • The data comes from here