Deepfake Detection App (Image + Video)

This project provides deepfake detection for images and videos using:

• Face detection: MTCNN (facenet-pytorch)
• UI: Gradio

For videos, the app samples every Nth frame, detects faces, classifies each face, and aggregates one final verdict.

Features

• Image deepfake detection with face bounding boxes
• Video deepfake detection with frame sampling
• Per-face predictions and overall verdict
• Explicit No face detected result when no valid faces are found
• Warning banner in UI for safe usage

Project Structure

• app.py: Gradio app entrypoint
• src/model_service.py: model loading and inference
• src/face_service.py: MTCNN detection and face crops
• src/image_pipeline.py: image processing flow
• src/video_pipeline.py: sampled-frame video processing
• src/rendering.py: box rendering helpers
• src/aggregation.py: overall decision logic

Setup (Windows PowerShell)

1. Create and activate a virtual environment:

python -m venv .venv
.\.venv\Scripts\Activate.ps1

2. Install dependencies:

pip install -r requirements.txt

3. Run the app:

python app.py

Then open the local Gradio URL shown in the terminal.

Notes

• The classifier is image-based. Video detection is an approximation through sampled frame analysis.
• First run downloads model weights, so startup can take longer.
• If you get codec issues on videos, install FFmpeg and retry.

Tuning

You can tune defaults in src/config.py:

• FACE_CONFIDENCE_THRESHOLD
• FAKE_SCORE_THRESHOLD
• VIDEO_SAMPLE_RATE
• MAX_FACES_PER_FRAME
• MAX_VIDEO_FRAMES_TO_PROCESS