Criminal-gan

A GAN created to generate faces based on records of criminals

Requirements

Linux or Mac
Python 3.6
Pip
imagemagick

Datasets

In this case we are going to use the following datasets:

NIST Special Database 18

NIST Mugshot Identification Database (MID)

https://www.nist.gov/srd/nist-special-database-18

AT&T Laboratories Cambridge hosted in conjunction with Cambridge University Computer Laboratory

The Database of Faces

http://www.cl.cam.ac.uk/research/dtg/attarchive/facedatabase.html

BioID Face Database - FaceDB

https://www.bioid.com/About/BioID-Face-Database

FGnet - IST-2000-26434 Face and Gesture Recognition Working group

Head Pose Image Database

http://www-prima.inrialpes.fr/FGnet/html/home.html

Project instalation

Clone this project and run the following

Setting up the virtual env

pip install virtualenv
virtualenv venv_cgan
source venv_cgan/bin/activate

Installing dependencies

pip install -r requirements.txt

Setting up datasets

Use the following script or follow all the commands below

cd util
sh install_dataset.sh

OR

Download the datasets

cd data/
wget https://s3.amazonaws.com/nist-srd/SD18/sd18.zip
wget http://www.cl.cam.ac.uk/Research/DTG/attarchive/pub/data/att_faces.zip
wget ftp://ftp.uni-erlangen.de/pub/facedb/BioID-FaceDatabase-V1.2.zip
wget http://www-prima.inrialpes.fr/perso/Gourier/Faces/HeadPoseImageDatabase.tar.gz

Create a folder to contain all the dataset

mkdir all_images

Unzip the folders, convert if necessary and move to a single folder

NIST Special Database 18

unzip sd18.zip
mv sd18/*/*/*_F.png all_images/
rm -rf sd18

BioID Face Database - FaceDB

unzip BioID-FaceDatabase-V1.2.zip -d bioid/
convert bioid/*.pgm all_images/bioid_%03d.png
rm -rf bioid/

AT&T - The Database of Faces

unzip att_faces.zip -d att/
convert att/*/*.pgm all_images/att_%03d.png
rm -rf att/

FGnet - Head Pose Image Database

tar xvz -f HeadPoseImageDatabase.tar.gz Front/
convert Front/*.jpg all_images/att_%03d.png
rm -rf Front/

mkdir resized/
mogrify -resize 64x64! -quality 100 -path resized/ all_images/* -type Grayscale
rm -rf all_images/

Start

$ python gan.py


	+-+-+-+-+-+-+-+-+ +-+-+-+
	|C|R|I|M|I|N|A|L| |G|A|N|
	+-+-+-+-+-+-+-+-+ +-+-+-+
	Created by Davidson Mizael
	


# Starting...
# Loading data...
# Starting generator and descriminator...
# Starting epochs (15)...
# Progress: [0/15][38/38] Loss_D: 1.5640 Loss_G: 0.0004
# Progress: [1/15][38/38] Loss_D: 1.4570 Loss_G: 0.0004
...

dbx0/criminal-gan