Teaching your webcam to forget people exist

What it does

The project runs in a web browser and uses TensorFlow.js to strip humans out of a live video stream. Instead of asking for a clean background shot upfront, it watches the scene over time to learn what the backdrop looks like, then attempts to remove people in real time. All processing happens locally in the browser.

The interesting bit

The clever part is the temporal approach: the code watches the scene over time to learn which pixels belong to the background, then uses that learned backdrop to blot out humans as they move. It is a tidy trick for a browser tab.

Key highlights

• Runs fully client-side in the browser via TensorFlow.js
• Learns the background over time instead of requiring an empty-room reference frame
• Live CodePen and Glitch demos available to try immediately
• Released under Apache license for reuse
• Explicitly framed as an experiment by the author

Caveats

• The author warns it is an experiment and may not work perfectly in all scenarios
• Specific accuracy, performance, or browser compatibility details are not quantified in the README

Verdict

Worth a look if you’re building browser-based computer vision toys or teaching TensorFlow.js in the classroom. If you need guaranteed person removal for a production pipeline, this is explicitly not that.