Generate Faces

This repository shows an example implementation of Udacity's Deep Learning Nanodegree project on generative adversarial networks (GANs). The goal is to generate celebrity faces using the CelebA dataset in only one epoch, which means that the results would get better if you would train for more epochs.

Getting Started

The below instructions will show how I setup the environment to run this project on an Amazon GPU. It is assumed that you already have an AWS Account.

AWS Environment Setup

1. Sign in to Console. Then go to Services > EC2.
2. Click on Launch Instance button.
3. Choose Amazon Machine Image (AMI) by selecting AWS Marketplace: Search for Deep Learning AMI with Source Code (CUDA 8, Ubuntu) and select it.
4. Choose an Instance Type (filter by GPU compute): Select p2.xlarge.
5. Click on Review and Launch.
6. Configure Security Groups.
7. Generate .pem authentication key pair, if needed.
8. Click Launch Instances and then View Instances to come to the EC2 Dashboard.
9. When View Instances displays 2/2 checks passed under Status Checks, proceed to next step.

EC2 Instance Setup

1. Login to EC2 Instance:

ssh -i  <your key name>.pem ubuntu@<IPv4 public IP address displayed on EC2 Dashboard>

2. Clone the repository and navigate to the project folder:

git clone https://github.com/gro1m/generate_faces.git
cd generate_faces

3. Install requirements from requirements.txt:

sudo python3 -m pip install -r requirements.txt

4. Install cudnn:
   1. Download and untar cudnn version 6 for CUDA 8 (Copied from RISHAV KUMAR):

   CUDNN_TAR_FILE="cudnn-8.0-linux-x64-v6.0.tgz"
   wget http://developer.download.nvidia.com/compute/redist/cudnn/v6.0/${CUDNN_TAR_FILE}
   tar -xzvf ${CUDNN_TAR_FILE}
   

   2. Copy the files to appropriate locations and set the correct rights - see also NVIDIA Docs:

   sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include
   sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64
   sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
   

   3. Adjust cudnn library softlinks:

   cd /usr/local/cuda/lib64
   sudo ln -sf libcudnn.so.6 libcudnn.so
   sudo ln -sf libcudnn.so.6.0.21 libcudnn.so.6
   sudo ldconfig
   

NOTE: I tried to use cudnn6.5(v2), but this led to an import error.

Setup jupyter notebook

1. cd /home/ubuntu
2. jupyter notebook --ip=0.0.0.0 --no-browser
3. Need token by jupyter notebook token, so copy everything starting with :8888/?token
4. Access jupyter notebook index from your web-browser by visiting: <IPv4 public IP address displayed on EC2 Dashboard> :8888/?token=...

License

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

Acknowledgments

I acknowledge that this project is heavily based on the framework provided by Udacity on face_generation.