bthavanish/mneme

Mneme is a browser app that points your webcam at the world and shows you what it sees. It can draw labeled boxes around everyday objects like people, dogs, cars, or cups, and it can recognize faces you have saved yourself. Everything happens inside your browser tab, nothing is sent to a server, and there are no accounts or API keys involved. The app offers two modes that can run separately or together. Object detection uses a model called COCO-SSD, which can identify 80 common types of objects and draw a box with a label around each one it spots. Face recognition works differently: you save a face under a name, and afterward the app tries to match new faces against your saved list, showing anyone it does not recognize as Unknown. You choose whether to run objects only, faces only, or both at once. Under the hood, all of the machine learning runs in the browser using a library called TensorFlow.js, which uses your graphics card through WebGL to keep things fast. Object detection runs faster on desktop computers than on phones, and the same is true for face recognition, which uses a separate library for detecting and comparing faces mathematically. Saved face data is stored locally in the browser's own storage, and the app shows a consent message before it saves anything. To try it, you can visit the hosted website directly, or run it yourself by installing Node.js, running a couple of npm commands, and opening the address it prints in a browser, then allowing camera access. There are also instructions for testing the app on a phone over the same Wi-Fi network, and for building a production version to deploy anywhere static files can be hosted. A settings panel lets you adjust things like confidence thresholds, camera mirroring, and dark mode, or delete all saved face data. The project is released under the Apache 2.0 license.