Want to try SpinePose? Pick an image, video, or camera, load a model, and run it here in your browser. Your images and video frames stay on your device.
01 Choose a model
The first load downloads model files. Processing then runs in this browser tab.
02 Choose an input
Camera access is requested only when you select Camera and enable it.
03 Run inference
Preview
Your selected image or tracked video appears here.
The colored lines show estimated joints. Results are not intended for clinical decisions.
For best results: Keep one person near the center and fairly large in the frame. To enable fast inference, the browser demo skips person detection and treats the whole frame as one person's crop. With several people or someone far away, the points may be off. Use our Python library for more robust person detection and tracking.
For setup and API details, see spinepose-js. The SpinePose library chapter has the model results and citations.