/DL_Generate_Faces

Project from Deep Learning Nanodegree of Udacity

Primary LanguageHTML

Face Generation

Project Overview

In this project, I defined and trained a DCGAN on a dataset of faces. The goal of this project is to get a generator network to generate new images of faces that look as realistic as possble. The image below is a result of the training:

example

Project Instruction

Instruction

  1. Clone the repository and navigage to the downloaded folder. 
    	git clone https://github.com/choonghee-lee/Face-Generation
	cd Face-Generation
  2. Open the dlnd_face_generation.ipynb file. Of course, you can find HTML version of the file. 
    	jupyter notebook dlnd_face_generation.ipynb
  3. Read and follow the instructions! This repository does not include the dataset of faces. You can find and download it in the notebook.

Project Information

Contents

  • Pre-processed Data
  • Create a DataLoader
  • Define the Model
    • Discriminator
    • Generator
    • Initialize the weights of your network
    • Build complete network
  • Discriminator and Generator Losses
  • Optimizers
  • Training
  • Training Loss
  • Generator samples from training

Model - Discriminator

Layer Input Dimension Output Dimension Batch Normalization
Conv1 3 64 False
Conv2 64 128 True
Conv3 128 256 True
Conv4 256 512 True
FC 2048 1 False

Model - Generator

Layer Input Dimension Output Dimension Batch Normalization
FC 100 2048 False
Deconv1 512 256 True
Deconv2 256 128 True
Deconv3 128 64 True
Deconv4 64 3 False

Libraries

The list below represents main libraries and its objects for the project.

  • PyTorch (Generator and Discriminator)

Accelerating the Training Process

In the training phase, it takes your time too much so I recommend you to use a GPU to train the dataset of faces.

Amazon Web Services

You can use Amazon Web Services to launch an EC2 GPU instance. (This costs money!)