Boneyard Tools

Object Detector

Find and label every object in a photo. Drop in an image, pick a model, and the detector locates each object, draws a labeled bounding box around it, gives a count of what it found, and lets you download the annotated image. Two quality tiers: Fast is YOLO11n, a 10 MB model that loads instantly, and Best is D-FINE-S, an Apache-2.0 model that is noticeably more accurate for a bigger download. Drag the confidence slider to hide weak guesses; it re-filters instantly without re-running the model. Both run entirely in your browser (WebGPU accelerated when available), so nothing is uploaded. The model downloads once on first use, then is cached.

How to detect objects in a photo

  1. Drop an image in, or click to browse.
  2. Wait a few seconds while the AI scans the photo for objects.
  3. Adjust the confidence slider if you like, then download the annotated image.

Examples

Photo to labeled objects

A photo of two cats on a couch with two TV remotes
Boxes labeled cat, cat, remote, remote with confidence scores

Frequently asked questions

Is my image uploaded anywhere?

No. Whichever model you pick runs entirely in your browser, using your GPU via WebGPU when available and WebAssembly otherwise. The image is processed on your device and never uploaded. Only the model itself is downloaded, once, then cached for instant reuse.

What kinds of objects can it detect?

It detects 80 common object types from the COCO dataset: people, animals (cats, dogs, birds, horses), vehicles (cars, bikes, buses, planes), and everyday objects (chairs, laptops, bottles, cups, phones, TV remotes, and more).

Which model powers the detection?

You pick the tier. Fast is YOLO11n, a state-of-the-art real-time detector that loads instantly as a roughly 10 MB ONNX model. Best is D-FINE-S, an Apache-2.0 transformer that is noticeably more accurate (about 48.5 COCO AP versus 39.5) for a similar, roughly 11 MB download. Both predict a labeled box and a confidence score per object and run locally in your browser (WebGPU accelerated when available, WebAssembly otherwise).

What does the confidence slider do?

Each detection has a score from 0 to 1 for how sure the model is. The slider hides any box below the threshold you pick, so you can trim weak or spurious guesses. It re-filters the existing results instantly without re-running the model.

Why does the first run take a moment?

The model you pick (roughly 10 MB for Fast, 11 MB for Best) downloads the first time you use it, then is cached for instant reuse. Larger images also take a little longer because more pixels are analyzed.

Related tools