Webcam AI Filter (WIP)

A small Python app that captures a live webcam stream and runs real-time face detection using a pre-trained MediaPipe model.
This repository is being built step-by-step toward a simple face filter overlay and optional background segmentation.

Features (so far)

• Live webcam capture (OpenCV)
• Real-time face detection (MediaPipe Face Detection, pre-trained)
• Draws a bounding box around the detected face
• PNG overlay filter (transparent PNG blended onto the webcam feed)
• Background segmentation (MediaPipe Selfie Segmentation, pre-trained)
• Background replacement modes: blur / solid color / image
• Modular project structure (src/ package)

Assets

Place a transparent PNG overlay here:

• assets/glasses.png
• background.jpg — optional background image for mode 3 (any JPG/PNG)

The PNG must contain an alpha channel (transparency). The overlay is resized relative to the detected face box and alpha-blended onto the frame.

Setup (Conda)

1) Create and activate the environment

From the project root:

conda env create -f environment.yml
conda activate webcam-ai-filter
conda activate webcam-ai-filter

2) Run the app

python -m src.main

Controls

• q — quit the app
• b — toggle background effect (none ↔ blur)
• 1 — solid background
• 2 — blur background
• 3 — background image (requires assets/background.jpg)

Notes

• OpenCV frames are BGR by default; MediaPipe expects RGB (conversion is done in code).
• Background segmentation is real-time but may reduce FPS depending on camera resolution. The app uses downscaling and mask smoothing for performance/quality.

Next Steps

• Add overlay filter (PNG with alpha blending) based on direction of view
• Add simple face tracking / smoothing
• Add background segmentation and replacement