A pretrained autoencoder for femalefaces

Run

Install dependencies [dlib], [cv2], [pygame], [matplotlib], [keras], [tensorflow]:

sudo pip install dlib opencv-python pygame matplotlib tensorflow keras

git clone [email protected]:BSolut/faceautoencoder.git

Execute the editor

python editor.py

Train your own dataset

Friendly note

If you still want to sleep peacefully at night, make sure that there are only two eyes in one picture in your training data.

Building training data

• Build prepare working dir:

  mdir ~/working_dir
  cd ~/working_dir
  mkdir clean
  mkdir raw
  mkdir ignore
  wget https://github.com/davisking/dlib-models/raw/master/shape_predictor_5_face_landmarks.dat.bz2
  bzip2 -d shape_predictor_5_face_landmarks.dat.bz2
  wget https://raw.githubusercontent.com/opencv/opencv/master/data/haarcascades/haarcascade_frontalface_default.xml

• Generate a textfile with links for training images. Acceptable sources are pinterest.com, famousbirthdays.com or any source of your linking
• Download images into a directory

  python data.py get --source [links.txt]

• Auto process images (croping/face aligment)

  python data.py process

• Remove any outliers (e.g. not a face, black and white images)

  python data.py check

  Once started, you can use: r - removes that image Left/Right arrow key - move inside the dataset ESC - exits
• Build train_data.npy

  python data.py build

Execute training

python train.py

Once you are happy with the results, build stats

python stats.py