This object counter counts things in photos without uploading them. For lab and workshop work — cells, colonies, seeds, beans, pills, screws or coins on a plain background — the built-in shape counter needs no download. For everyday scenes, download an AI detector once to count people, cars, animals, bottles and more. Drop many images and get annotated pictures and a CSV of counts.
| Situation | Tip |
|---|---|
| Shadows | Use even, diffuse light |
| Touching items | Spread them out |
| Light items | Choose “lighter than background” |
| Doubtful AI boxes | Raise the confidence |
Shape counting for separate items on a plain contrasting background; the AI detector for everyday objects in normal scenes.
The shape counter estimates touching items from their combined size; spread items out for the most accurate count.
The 80 COCO classes, including people, vehicles, animals, furniture and kitchen items.
No. Files and text are processed in your browser. Only the model itself is downloaded, once, when you ask for it.
Yes. The Downloaded models panel lists what this tool stored and deletes it after a warning; you'd need to download it again to use it.
The built-in shape counter converts the image to greyscale, separates objects from the background with Otsu's automatic threshold (or a manual one), labels 4-connected components, drops specks below a minimum area and anything implausibly large, and counts shapes much larger than the median size as several touching objects. The AI method runs YOLOS-tiny or DETR ResNet-50 object detection (80 COCO classes) through Transformers.js on your device, with a confidence threshold and optional class filter. Each image is annotated with boxes and numbers; counts export as CSV with annotated PNGs in a ZIP.