🔒 Vision runs on your device — images never uploaded. No cost, no sign-up.
More tools →

Find & label the objects in a photo

Drop an image and AI draws a labelled box around everything it recognises — tune the confidence, toggle classes, save the result. All on your device.

⚙ on-device🔒 no upload💸 no cost🎚 adjustable
📷 Click or drop an image
jpg · png · webp
First use downloads a small vision model (~45 MB), then it's instant and works offline.

⚡ Batch detect & dataset export Pro

Pre-labelling a folder of images for a computer-vision project? Pro runs detection across many images at once and exports one ready-to-train dataset — still 100% in your browser, nothing uploaded.

  • Drop a whole folder of images and detect them in sequence
  • Adjust the confidence slider to set what counts across the batch
  • Export COCO JSON, YOLO, Pascal VOC or CSV — the standard training formats
  • Single-image export too: grab annotations for one photo right under the picture
Get Pro / All-Access →
📁 Click or drop images to detect
select many at once

Object Spotter Pro one-time key

The detector — adjustable confidence, class toggles and annotated-image download — stays free and unlimited forever. Pro unlocks dataset export (COCO / YOLO / Pascal VOC / CSV) and batch processing for people building computer-vision datasets, still entirely on your device.

Try it now with demo code AV-OBJECT-SPOTTER-DEMO (or AV-ALL-DEMO for every AppVitamins app). A real key unlocks locally; your images still never leave this device.

Get the free Computer-Vision Starter checklist

A short guide to building a clean object-detection dataset — labelling consistency, train/val splits, class balance and which export format to use for YOLO, Detectron2 or your own pipeline.

Send me the checklist

On-device object detection and auto-labeling, for free

Object Spotter runs a real object-detection neural network — DETR (DEtection TRansformer) trained on the COCO dataset — inside your browser to find and label common objects in any photo. Drop an image and it draws a coloured bounding box around each thing it recognises, from people, cats and dogs to cars, bikes, chairs, laptops, bottles and food, with a confidence score on every box. Your image never leaves your device, there is no account, and there is no per-image cost. The model downloads just once (about 45 MB) and then works instantly, even offline.

Tune the confidence and toggle classes

Real photos are messy, so a single fixed threshold rarely fits every picture. The confidence slider lets you raise the bar to keep only the detections the model is sure about, or lower it to surface faint or partly hidden objects — and it re-renders instantly because it filters detections the model already produced rather than running it again. Click any class chip to toggle that category on or off on the canvas, which is handy when you only care about, say, people and cars and want to hide the clutter. When the picture looks right, download it as an annotated PNG or copy the object list as text.

Export training-ready annotations

Because Object Spotter already produces bounding boxes, it doubles as an AI pre-labeling tool for computer-vision projects. Pro exports your detections in the formats training pipelines actually expect: COCO JSON (images, annotations and categories in one file), YOLO label files with a classes.txt, Pascal VOC XML per image, and a flat CSV for spreadsheets. Batch mode runs a whole folder of images in one pass and bundles them into a single dataset, so you can seed a model with auto-labels and then refine them in your editor of choice — all without uploading a single private image.

Part of a privacy-first image toolkit

Object Spotter is one of a family of on-device image tools: write alt-text and captions with Image Describer, cut out backgrounds with BG Remover, and shrink or convert pictures with SqueezeIMG. One AppVitamins pass unlocks the Pro extras across all of them.

Frequently asked questions

Are my images uploaded to a server?
No. The detection model runs entirely inside your browser, so your images never leave your device. There is no account and no per-image cost.
What objects can it detect?
It uses DETR trained on COCO, which recognises 80 everyday categories — people, animals, vehicles, furniture, food, sports gear, kitchenware and more — with a labelled box and confidence score on each.
Can I change how strict the detection is?
Yes. The confidence slider raises or lowers the threshold live without re-running the model, and you can toggle individual classes on or off.
Can I export the bounding boxes to train a model?
Yes — Pro exports COCO JSON, YOLO, Pascal VOC and CSV, and batch mode turns a folder of images into one dataset, all on your device.
Is the AI always right?
No. AI detection is a strong first pass but can miss objects, mislabel them or draw loose boxes, so review and correct the output rather than trusting it as ground truth.