Object detection in the browser

A detector can run on the visitor's own device, inside the web page. No server, no upload.

5 min read

How it works

The model is exported to ONNX, converted to half precision to make it smaller, and loaded by ONNX Runtime Web, a JavaScript build of the same runtime used on servers. The page draws the image onto a canvas, runs the model and draws the boxes.

WebGPU or WASM

WebGPU runs the model on the graphics card. It is the fast path, in browsers that support it.
WASM (WebAssembly) runs on the CPU and works in every modern browser. Slower, but always there.

Private by design

Photos, video and camera frames are processed on the device. Nothing is sent to a server, which matters for cameras in homes, workplaces and anything with faces.

Why fotonet suits the browser

A small model: fotonetn downloads as 5.8 MB.
No NMS: the browser does not need a post-processing step reimplemented in JavaScript; the model's output is the result.

Try it

The live demo runs fotonetn exactly this way, on your images, video or camera.