Gaze Correction for Live Video

I've been experimenting with gaze correction using TensorFlow, OpenCV, and face detection libraries. The goal is to create a more natural eye contact experience during video calls.

Overhead

Running on macbook with apple silicon (and mediapipe backend), these are the numbers from activity monitor:

• OBS: ~800MB
• python: ~600mb

Performance

I'm getting ~60fps, which is sufficient for live video. There is no noticeable lag or delay in the video feed with audio.