AI Object Detector

What the AI Object Detector does

Upload a photo and this tool finds the objects in it, draws a box around each one and labels what it is — people, cars, animals, furniture, food and dozens of other everyday categories. It runs the DETR (Detection Transformer) model from Facebook Research directly in your browser via WebAssembly, so your images are analysed locally and never sent to a server.

It is a quick way to see what a computer-vision model 'sees' in an image, to count or spot objects in a scene, to generate a starting point for image tagging and accessibility descriptions, or simply to explore how modern object detection behaves on your own photos.

How detection transformers work

Traditional detectors scanned an image with many candidate boxes and filtered them down. DETR takes a different approach: a convolutional backbone turns the image into a set of features, and a Transformer encoder-decoder then reasons about the whole picture at once to predict a fixed set of objects, each with a label and a bounding box. This lets it consider the relationships between objects rather than judging each region in isolation.

The model used here is trained on the COCO dataset, which defines 80 common object classes. That means it recognises a wide range of everyday things very well, but it will not detect categories it was never trained on — specialised or unusual objects may be missed or mislabelled.

How to use it and what to expect

Select or drop in an image and wait for the model to run; as with all the in-browser AI tools, the model downloads once on first use and is then cached for instant reuse. Each detected object appears with a labelled box and a confidence score. Clear, well-lit photos where objects are not heavily overlapping give the most reliable results, while crowded or low-light scenes are harder and may produce a few misses.