Real Fake Face Image Classifier - "ArtificialFaces"

This Project has been created to classify the faces which are Real or Fake (generated by GAN). To be precise, the StyleGAN2 architecture generates the faces at an exceptional level producing these results indistinguishable from real photographs. But when I observed lots of very finely generated faces and observed some indications

Some key indications these pictures were generated are:

• Out-of-focus, non-descript backgrounds
• Only a single person in frame
• Odd artefacts, such as earrings merging with ears
• The identical positioning of the eyesThis is primary reason to

This is my inspiration to create this project. This Classifier can be used in differentiating Fake Account Photo on LinkedIn and other social media platfoms.

For this project work and with the kaggle dataset, a Real-Fake face classifier is developed using several approaches: Convolutional Neural Networks (CNN) and Transfer Learning with and without face cropping using the Multi-Task Cascaded Convolutional Neural Networks (MTCNN).

This is a real-fake face classifier built with convolutional neural networks. The classifier was trained on a data set comprised of 1400 images (700 of each class) and tested on 600 images (300 per class). The classifier achieved an accuracy of 83%.

To extend this Project, and in order to leverage the created models, a Web Application is developed using the streamlit framework. For more information about the models or the project, check out the Project Notebook in the root directory of this repository. The application source code is in the app folder.

View the web app live

Click here to go to the app page.

demo.mp4

Run the app locally

Open a terminal and start by cloning the project repository

  https://github.com/Prasantkumar987/Real-FakeFaces.git

Go to the project directory

  cd Real-FakeFaces

Install dependencies

  pip install -r requirements.txt

Start the server (Linux / MacOS)

  streamlit run app/app.py

Start the server (Windows)

  streamlit run app\app.py

If the browser window does not open automatically when the streamlit run command is executed, you can manually navigate to localhost:8501 To terminate the application, go back to the terminal and shutdown the server by pressing CTRL + C.

App Snapshot