Computer vision: objects, people and AR
YOLOX finds and labels objects in a photo or on the camera and counts the people. In AR, MediaPipe finds the hair and the iris and changes their color live. Everything runs in your browser.
or an example:
0people
Each box is a person YOLOX-Nano found. In a crowd it undercounts: the model is small (0.9 M parameters) and people far away or covered are missed.
Nothing leaves your browser: the photo and the camera are processed here, with YOLOX on ONNX Runtime (WebAssembly) and MediaPipe, and the models are served by this site. The examples are CC0 photos from Wikimedia Commons.
Models and photos
- YOLOX-Nano · Megvii, Apache-2.0
- MediaPipe Hair Segmentation e Face Landmarker · Google, Apache-2.0
- ONNX Runtime Web · Microsoft, MIT
- Street · Gabriel Santiago (Unsplash), CC0
- Crowd · Mx. Granger, CC0
- Cat · Gerda Arendt, CC0
- Coffee · Katerina Pavlickova (Unsplash), CC0
- Portrait · Christopher Campbell (Unsplash), CC0