BackEnd of the FaceApp application developed as part of a study on facial recognition and Deep Fake.
- Celebrity classification API
- Face morphing API
- Face manipulation API
- Mongodb
- Docker-compose
- Technologies
Python version > 3.8
This API return the most similar celebrity to the user. It uses FaceNet model and a saved SVM classifier. You can refer to the "celeb_class_api" directory if you want to test. You can also retrain the model on your own dataset using the notebook "facenet.ipynb" in the "notebooks" directory.
In addition, You can create and run Docker image using the Dockerfile in the "celeb_class_api" directory. You can use these commands:
$ cd celeb_class_api
$ sudo docker build -t celeb_class_api:2.0 .
$ sudo docker run -it --rm -p 5002:5002 -v ../../../mnt/d/big_dataset/Train:/app/volume -e NUMBER_WORKERS=10 -e NUMBER_THREADS=4 celeb_class_api:2.0
Replace "../../../mnt/d/big_dataset/Train" with your own path in this command lines as well as in the docker-compose.
Python version = 3.7 or 3.6
By employing the SimSwap model, this API manages to merge two faces. To use it, you can refer to the "simswap_api" directory. You should at first clone the SimSwap github. Next, you must prepare the evironment and install the different pre-trained models. All the steps are given in this link. Finally, add the file given in the "simswap_api" folder and install requirements. In addition, you must install other modules related to the API implementation: flask, uwsgi.
You can start the API through the "uwsgi_simswap.sh" script. If you find problems with the dependicies, you can run the API as a Docker image through the following commands:
$ cd simswap_api
$ sudo docker build -t simswap:1.0 .
$ sudo docker run -it --rm -p 5001:5001 -e NUMBER_PROCESSES=10 -e NUMBER_THREADS=1 simswap:1.0
Python version = 3.7
This API generates face manipulation videos using First Order Model for Image Animation github. After cloning the previous github, you must download checkpoints using this link. In this project, we used the checkpoint nammed "vox-cpk.pth.tar". Then, install all the requirements as well as face-allignement library as follows:
$ git clone https://github.com/1adrianb/face-alignment
$ cd face-alignment
$ pip install -r requirements.txt
$ python setup.py install
You can launch the API using either "gunicorn_fom.sh" script or Docker as follows:
$ cd fom_api
$ sudo docker build fom_api_url:4.0 .
$ docker run -it --rm --gpus all -p 5000:5000 -e NUMBER_WORKERS=10 -e NUMBER_THREADS=2 fom_api_url:4.0
The Dockerfile inside the "mongodb" folder aims to create a database called "ingeniance" that contains all the image urls that will be needed in the face manipulation task. It is important that this data was not used to train the classification model. You can create this docker image as follows:
$ cd mongodb
$ sudo docker build -t mongo:latest .
In order to run this docker file that start the entire application at one, you must at first build att the previous images as well as the image of the app front-end given in this github. In addition, it is important to mention that this docker-compose must be executed inside the "webcam" folder created after clonning the font-end github. Therefore, you must copy it to the "webcam" folder.
The commands for running docker-compose are:
$ cd webcam
$ sudo docker-compose up
-
Except face manipulation API, all the APIs can be run using the normal version of Docker. But for face manipulation API, you have to install docker-nvidia if you are working on linux. For Windows 11 users, you can refer to this link.
-
For each API, it is better to use virtual environment. You can either use pycharm to create it directly or command lines as follows:
$ python -m venv c:\path\to\myenv
To activate it on windows, use:
$ myenv\Scripts\activate.bat
To activate it on Unix or MacOS, run:
$ source myenv/bin/activate
To deactivate it, run:
$ deactivate