Object Counter: Detect & Count Objects in Any Image

Object Counter: Detect & Count Objects in Any Image

Count people, cars, trucks, boats, bikes, animals and 80 everyday objects in photos, drone shots and high-resolution images. Normal detectors shrink a big image to about 640 px before looking at it, so anything small, like cars in a parking lot or people in a crowd, simply disappears. This Actor uses sliced inference: it scans the image in overlapping tiles at full resolution, then merges the results so each object is counted once.

Same public photos, same model, single pass vs. tiled:

Photo	Single pass	Tiled (this Actor)
Aerial marina, 3613×2400	5 boats	91 boats
Festival crowd, 3586×2981	18 people	105 people
Parking lot, 2304×2304	18 cars	31 cars

Try them in the examples below.

• 🔎 Small-object mode: overlapping tiles plus a full-image pass, with partial boxes at tile edges merged into whole objects
• 🧮 Counts per class plus a grand total for the run
• 📦 Every box: label, confidence and [x0, y0, x1, y1] in original image pixels
• 🖼️ Annotated preview JPEG per image, with a public link
• 🎯 Count only what you need: ["car", "truck"], ["people"], ["boat"]
• 🔒 Runs inside the Actor: images are not sent to any third-party AI API
• 💸 Pay per image (plus per tile in small-object mode); failed images are free

Use cases

• Parking and traffic: count cars, trucks and buses in lot or street photos
• Events and venues: estimate people in crowd photos
• Marinas and waterfronts: count boats
• Real estate and retail: count chairs, tables, bottles or other items in photos
• Data pipelines and AI agents: add "how many X are in this image?" to any workflow through the API or MCP

Ready-made examples

• Count cars in parking lot photos
• Count people in crowd and event photos
• Count boats in marina and aerial photos

How it compares

	Typical single-pass detector	Vision LLM API	Object Counter
Small objects in large images	Missed after downscaling to ~640 px	Approximate	Found with overlapping full-resolution tiles
Exact per-class counts and boxes	Yes	Rarely consistent	Yes, plus annotated preview
Images sent to a third party	Depends	Yes	No, runs inside the Actor
Pricing	Per image	Per token	Per image, plus per tile only for large images

How to use

1. Paste image links into Image URLs (or read them from a scraper's dataset).
2. Optionally list the Objects to count. Leave empty for all 80 classes.
3. Keep Small-object mode on auto and click Start.

Input example

{
  "urls": ["https://ultralytics.com/images/bus.jpg"],
  "classes": ["person", "bus"],
  "tiling": "auto",
  "minScore": "0.35",
  "saveAnnotatedImages": true
}

Output example

Real output for the image above:

{
  "source": "https://ultralytics.com/images/bus.jpg",
  "width": 810,
  "height": 1080,
  "total": 5,
  "counts": {"person": 4, "bus": 1},
  "tilesUsed": 4,
  "annotatedImageUrl": "https://api.apify.com/v2/key-value-stores/.../records/annotated-00001.jpg",
  "detections": [
    {"label": "bus", "score": 0.9421, "box": [16, 232, 791, 719]},
    {"label": "person", "score": 0.9094, "box": [668, 458, 810, 883]}
  ]
}

A SUMMARY record in the run's key-value store holds the totals across all images.

Pricing

Event	Price	When
Image analyzed	$0.002	Each image successfully processed
Tile analyzed	$0.0003	Each extra tile, only for images over ~960 px with tiling on
Actor start	$0.00005 per GB	Once per run

Example	Cost
1,000 phone-size photos (no tiling)	$2.00
One 2304×2304 photo (25 tiles)	$0.0095
One 12 MP photo (48 tiles)	$0.0164

Platform usage is included. Images that fail to download or decode are free. Set tiling to never for the cheapest single pass, cap Max tiles per image, or set the run's maximum cost; the Actor stops cleanly at the limit.

FAQ

Can it count objects that aren't in the list? Not out of the box. The model knows the 80 classes above. For custom objects (products, defects, equipment), see custom work below.

Do I pay for images that fail? No. Download or decode failures return an error row and are not charged.

Why are there extra classes in my counts? Leave Objects to count empty to count everything, or list only the classes you want.

Custom work

Need a custom-trained detector, video or a full vision pipeline? Contact Nookforge Studio.

Supported objects

person, bicycle, car, motorcycle, airplane, bus, train, truck, boat, traffic light, fire hydrant, stop sign, parking meter, bench, bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair drier, toothbrush

Use with AI agents (MCP)

$claude mcp add --transport http apify "https://mcp.apify.com?tools=nookforgestudio/object-counter"

Then ask: "How many cars are in this photo? https://..."

Use via API

curl -X POST "https://api.apify.com/v2/acts/nookforgestudio~object-counter/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"urls": ["https://ultralytics.com/images/bus.jpg"], "classes": ["person"]}'

Also works from Make, n8n and Zapier through Apify's integrations.

Accuracy notes

• Detection is probabilistic. Counts are estimates; very dense crowds, heavy occlusion and straight-down (nadir) aerial views of tiny objects reduce accuracy.
• Lower Minimum confidence to find more objects (and more false positives); raise it for precision.
• The model recognizes the 80 classes above only.

Model and license

Detection uses YOLOX-S (Apache-2.0) by Megvii, run with ONNX Runtime.

Support

Open an issue in the Issues tab with an image link and what you expected.