This repository contains a Gradio-based web application that detects whether an image is a deepfake. The application uses a pre-trained InceptionResnetV1 model from the facenet_pytorch library for face recognition, and pytorch-grad-cam for visual explainability.
- Face Detection: The application uses the
MTCNNmodel to detect faces in the input image. - Face Classification: The detected face is then passed through an InceptionResnetV1 model to classify it as real or fake.
- Explainability: The application uses Grad-CAM to generate a visual explanation of the model's decision.
The following libraries are required to run the application:
torchtorchvisionfacenet-pytorchnumpypillowopencv-pythonpytorch-grad-camgradio
You can install these dependencies using the following command:
pip install -r requirements.txt- app.py: The main application script.
- resnetinceptionv1_epoch_32.pth: The model checkpoint file. Download from here
or Download Via Pinata through this hash
QmUrarTy82uk2bUzty7Rhtc2XNCJHnsJy9UQhh7sMmAj78 - requirements.txt: List of dependencies.
- README.md: This file.
To run the application locally, execute the following command:
python app.py
This will launch the Gradio interface in your default web browser. You can then upload an image and get the prediction along with the explainability visualization.
Go over to this Hugging Face Space: Deep Fake Detection
- Upload an Image: Click on the "Upload" button and select an image file.
- Get Prediction: The app will process the image and display whether it is "real" or "fake" along with confidence scores.
- Explainability: The app will also display an image with visual explainability using Grad-CAM.
This project uses the following open-source libraries:
- Gradio
- PyTorch
- FaceNet-PyTorch
- PyTorch-Grad-CAM
This project is licensed under the MIT License.
Any Contribution is highly appreciated for this Project!