itsgiving brings reactive meme overlays to video calls
Developed by Gazi Jarin, itsgiving is an open-source Python tool that displays reactive meme overlays during video calls by piping camera feeds through virtual camera backends such as OBS Virtual Camera and v4l2loopback. The application utilizes MediaPipe to track 478 facial landmarks, 52 blendshapes, hand positions, and upper-body poses to recognize 14 distinct expressions and gestures—including heart hands, time-outs, jaw drops, side-eyes, and dancing—and overlays corresponding animated GIFs scaled to track the user's face in real time. To overcome the limitation of fixed thresholds across varying facial geometries, version 2 implements a seven-second resting-face calibration that records baseline means and standard deviations across blendshape channels, scoring gestures using scale-invariant z-scores and face-width normalized coordinates.
While turning video meetings into reactive meme boards risks office chaos, itsgiving tackles one of computer vision's most persistent calibration challenges with a clean statistical approach.
- –Individual z-score normalization transforms MediaPipe blendshape outputs into robust triggers, overcoming arbitrary thresholds that fail on expressive or asymmetrical resting faces.
- –Distance-independent coordinate geometry, scaled to face width rather than absolute pixels, ensures reliable gesture detection regardless of how close the user sits to the camera.
- –The project demonstrates the utility of lightweight, entirely on-device computer vision pipelines that operate in real time without cloud API latency or costs.
- –Rigid dependency requirements showcase the packaging fragility in Python computer vision environments when balancing NumPy 1.x/2.x migrations across MediaPipe and OpenCV releases.
DISCOVERED
1h ago
2026-09-10
PUBLISHED
1h ago
2026-09-10
RELEVANCE
AUTHOR
Github Awesome