💊 Real-time Object Counter

Real-time Object Counter is a computer-vision demo that runs straight in your browser: turn on the webcam, point it at objects, and the model boxes and counts them in real time (~60 FPS with WebGPU).

Highlights:

• Fully on-device — inference via onnxruntime-web + WebGPU, images never sent to a server (private + works offline once the model loads).
• A model I fine-tuned — a YOLO11 nano fine-tuned to one object class then exported to ONNX (~10MB), light enough to run in real time right in the browser.
• Class-agnostic pipeline — it reads the model’s output shape, so swapping in another model (e.g. COCO’s 80 classes) just works.

Open in Chrome / Edge for WebGPU. The first load fetches the model (a few seconds), then it’s smooth. Allow camera access when prompted.

The story behind training this model (data → fine-tune on a single GPU → real-time on the web) and the product decisions: read Training a vision model to count objects in real time.