Final Project for CSCI 1470 Deep Learning - Face-Aging with Identity Preserved Conditional GANs
We are implementing the Face Aging With Identity-Preserved Conditional Generative Adversarial Networks paper by Xu Tang, Zongwei Wang, Weixin Luo and Shenghua Gao. We will use GANs to extract low-level features from faces of different ages and encode the features to our images. This is a generative model with a face-aging objective.
Face-aging is a new and exciting field in Deep Learning. With the rising popularity of FaceApp and people’s concerns about privacy, we were drawn to implement our version of an aging-app and think over issues of data privacy, deep fakes and identity theft.
- Run
python download_data.py
to download the the Cross-Age Celebrity Dataset metadata(817K) and Face images(3.5G). These will be downloaded and unzipped in thedata\
directory. - To train the model from scratch, run
python3 main.py
- To test the model and get results from a saved model, run
python3 main.py mode=='test'
We have used Google's Deep Learning VM with Pytorch framework to train with GPU. To run locally, follow the directions in setup conda section.
python
from zipfile import ZipFile
zip_file = ZipFile('path_to_file/t.zip', 'r')
zip_file.extractall('path_to_extract_folder')
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
rm Miniconda3-latest-Linux-x86_64.sh
source .bashrc
conda env create -f environment_gpu.yml
conda env update -f environment_gpu.yml
conda activate clsgan
conda deactivate