Detect Objects

Find the objects in an image, with where each one is and how sure the model is, using a model that runs in your browser.

What it does

Detect Objects runs the Image through the detection model you pick, such as COCO-SSD (MobileNet v2) (TensorFlow.js) or DETR ResNet-50 (transformers.js). It returns every object found with its label, a score from 0 to 1 and a box around it. The built-in models know about 80 everyday kinds of objects, such as person, car, dog, cup and laptop.

Objects below Confidence threshold are dropped, the rest are sorted best first and cut to Max objects. There is no cost per run and no API key.

When to use it

• Count things in an image: people in a photo, cars in a screenshot.
• Check whether an image contains a given object before you save or share it.
• Get positions to crop or highlight an object later.

For one overall label, use Classify Image. For a sentence describing the scene, use Caption Image.

Inputs and settings

Setting	Type	What it does
Model	Model picker	Any installed object-detection model, across engines.
Image	Media	Image URL, data URL, or $binary from a previous node.
Confidence threshold	Number	Default: 0.5
Max objects	Number	Keep at most this many objects (highest score first). Default: 20

Image has four tabs. From input takes an expression that points to an image from an earlier node, such as {{ $input.dataUrl }} after Screenshot Page or {{ $binary }} for an attached file. URL takes a web address. Upload and Base64 store a fixed image with the node, handy for trying a model.

Outputs

• Output

Each output item has these fields:

Field	Type	What it holds
objects	List of entries
count	Number
top	Text	The best label, or null when none passed

Shape of an output item

{
  "objects": [],
  "count": 0,
  "top": "…"
}

• objects: each entry is { label, score, box }, best first. box holds xmin, ymin, xmax and ymax, in pixels from the top-left corner of the image.
• count: how many objects are in objects, after the threshold and Max objects. It counts all kinds together.
• top: the label of the most confident object, or null when nothing was found.

Dependencies and credentials

None. This node needs no connection and has no slots for other nodes.

The Model picker lists your installed models first, then models that download on first use. Manage them on the Local AI page.

Example workflow

Check whether the visible page shows a person:

1. Screenshot Page with Mode viewport (the visible part of the page).
2. Detect Objects with Model COCO-SSD (MobileNet v2), Image (From input) {{ $input.dataUrl }} and Confidence threshold 0.6.
3. IF with the condition {{ $input.top }} equals person, then continue from the true branch.

Common issues

• “Choose a detection model.” Model is empty. Pick one.
• “No image provided. Pass an image URL, a data URL, or a $binary image from a previous node.” Image is empty, or its expression returned nothing. Check the field you point to.
• count is 0. Nothing passed Confidence threshold. Lower it, or check that the object is one the model knows.
• Fewer objects than expected in a busy image. Raise Max objects.
• The first run is slow. The model downloads and loads on first use; later runs are fast.

Related nodes

• Classify Image: one label for the whole image.
• Caption Image: describe the image in a sentence.
• Edit Image: crop or resize an image.
• Run Local Model: segmentation and depth maps.