This repo is forked from https://github.com/datitran/face2face-demo. Thanks Dat Tran for awesome work!

This is a pix2pix demo that learns from facial landmarks and translates this into a face. A webcam-enabled application is also provided that translates your face to the trained face in real-time.

You just read original description, after this modification it is not working in real-time anymore, thanks to increased resolution to 1024x1024.

• Increased resolution to 1024x1024!
• Pix2pix as submodule
• Use OpenPose for face/skeleton detection

# Clone this repo
git clone https://github.com/karolmajek/face2face-demo.git --recursive

# Create the conda environment from file (Mac OSX)
conda env create -f environment.yml

python generate_train_data.py --file angela_merkel_speech.mp4 --num 400 --landmark-model shape_predictor_68_face_landmarks.dat

Input:

• file is the name of the video file from which you want to create the data set.
• num is the number of train data to be created.
• landmark-model is the facial landmark model that is used to detect the landmarks. A pre-trained facial landmark model is provided here.

Output:

• Two folders original and landmarks will be created.

If you want to download my dataset, here is also the video file that I used and the generated training dataset (400 images already split into training and validation).

# Clone the repo from Christopher Hesse's pix2pix TensorFlow implementation
git clone https://github.com/affinelayer/pix2pix-tensorflow.git

# Move the original and landmarks folder into the pix2pix-tensorflow folder
mv face2face-demo/landmarks face2face-demo/original pix2pix-tensorflow/photos

# Go into the pix2pix-tensorflow folder
cd pix2pix-tensorflow/

# Resize original images
python tools/process.py \
  --input_dir photos/original \
  --operation resize \
  --size 1024 \
  --output_dir photos/original_resized

# Resize landmark images
python tools/process.py \
  --input_dir photos/landmarks \
  --operation resize \
  --size 1024 \
  --output_dir photos/landmarks_resized

# Combine both resized original and landmark images
python tools/process.py \
  --input_dir photos/landmarks_resized \
  --b_dir photos/original_resized \
  --operation combine \
  --output_dir photos/combined

# Split into train/val set
python tools/split.py \
  --dir photos/combined

# Train the model on the data
python pix2pix.py \
  --mode train \
  --output_dir face2face-model \
  --max_epochs 200 \
  --input_dir photos/combined/train \
  --which_direction AtoB

For more information around training, have a look at Christopher Hesse's pix2pix-tensorflow implementation.

1. First, we need to reduce the trained model so that we can use an image tensor as input:

   python ../reduce_model.py --model-input face2face-model --model-output face2face-reduced-model
   

   Input:

   • model-input is the model folder to be imported.
   • model-output is the model (reduced) folder to be exported.

   Output:

   • It returns a reduced model with less weights file size than the original model.

2. Second, we freeze the reduced model to a single file.

   python ../freeze_model.py --model-folder face2face-reduced-model
   

   Input:

   • model-folder is the model folder of the reduced model.

   Output:

   • It returns a frozen model file frozen_model.pb in the model folder.

python run_webcam.py --source 0 --show 0 --landmark-model shape_predictor_68_face_landmarks.dat --tf-model face2face-reduced-model/frozen_model.pb

Input:

• source is the device index of the camera (default=0).
• show is an option to either display the normal input (0) or the facial landmark (1) alongside the generated image (default=0).
• landmark-model is the facial landmark model that is used to detect the landmarks.
• tf-model is the frozen model file.

THIS MODEL WILL NOT WORK WITH THIS CODE

Pre-trained frozen model here. This model is trained on 400 images with epoch 200.

Me, my face and output:

Video on youtube:

Frozen model can be downloaded from here (1.1GB)

• Anaconda / Python 3.5
• TensorFlow 1.2
• OpenCV 3.0
• Dlib 19.4

Thanks to Dat Tran for inspiration, code and model!

See LICENSE for details. Copyright (c) 2017 Dat Tran.