Overview

The Generic Detector node is designed to detect objects within a video frame based on the similarity of text prompts and the images of objects. This functionality is useful for applications requiring identification and categorization of objects in real-time.

Inputs & Outputs

Inputs: 1, Media Format: Raw Video
Outputs: 1, Media Format: Raw Video
Output Metadata: Objects

Properties

Property	Description	Type	Default	Required
`model_id`	Model Type. Options: Yolo-Small (`yolov8s-world`); Yolo-Medium (`yolov8m-world`); Yolo-Large (`yolov8l-world`); OWL-Medium (`owlvit-base-patch32`); OWL-Large (v2) (`owlv2-base`).	`enum`	`yolov8s-world`	Yes
`class_list`	Comma separated list of objects to detect, optionally with an alternate label.	`string`	`null`	Yes
`interval`	Infer on every nth frame. 1 means infer every frame. Range: minimum `1`. Unit: frames.	`number`	`1`	Yes
`confidence_threshold`	Enable to override the default minimum inference threshold for all classes. Range: 0-1. Recommended values usually are in the 0.001 to 0.2 range. Start with a low threshold value and progressively increase to reduce false positives. Range: minimum `0`, maximum `1.0`. Step: `0.01`. Precision: 4 decimal places.	`float`	`0.1`	No
`per_class_thresholds`	Comma separated list of per-class thresholds. Leave empty to use the default threshold for all classes.	`string`	`null`	No
`iou_threshold`	Increase the threshold to reduce potential duplicate detections of a single object. Range: minimum `0`, maximum `1.0`. Step: `0.1`.	`float`	`0.5`	No
`min_object_size`	Minimum object size	`string`	`null`	No
`source_objects`	Look for objects within these object types. Leave empty to detect objects within the entire frame.	model-labels	`null`	No
`enable_max_optimizations`	Enable advanced optimizations to improve performance. This currently increases deployment start time.	`bool`	`false`	No
`clear_cache`	Set to true to clear model cache. This will increase deployment start time.	`bool`	`false`	No

Output Metadata

The fields below are declared by this node's metadata schema; the JSON values are representative examples.

Path	Type	Description
`objects`	`array`	Value of objects.
`objects[].attributes`	`array`	Value of attributes.
`objects[].attributes[].class_id`	`integer`	Value of class id.
`objects[].attributes[].label`	`string`	Value of label.
`objects[].attributes[].probability`	`number`	Value of probability.
`objects[].class_id`	`integer`	Value of class id.
`objects[].id`	`integer`	Value of id.
`objects[].label`	`string`	Value of label.
`objects[].probability`	`number`	Value of probability.
`objects[].rect.height`	`integer`	Value of height.
`objects[].rect.left`	`integer`	Value of left.
`objects[].rect.top`	`integer`	Value of top.
`objects[].rect.width`	`integer`	Value of width.

JSON example

{
  "objects": [
    {
      "attributes": [
        {
          "class_id": 0,
          "label": "example",
          "probability": 0.9
        }
      ],
      "class_id": 0,
      "id": 1,
      "label": "person",
      "probability": 0.95,
      "rect": {
        "height": 320,
        "left": 120,
        "top": 80,
        "width": 160
      }
    }
  ]
}

Object labels and attributes

Object labels/classes added: Each detected object uses the configured output label and its configured prompt/class index.
Object attribute labels/classes added: Dot-suffixed values in a configured label (for example, vehicle.red) are added as attributes with class 28999.