Detections Consensus

Class: DetectionsConsensusBlockV1

Source: inference.core.workflows.core_steps.fusion.detections_consensus.v1.DetectionsConsensusBlockV1

Combine detections from multiple detection-based models based on a majority vote strategy.

This block is useful if you have multiple specialized models that you want to consult to determine whether a certain object is present in an image.

See the table below to explore the values you can use to configure the consensus block.

Type identifier

Use the following identifier in step "type" field: roboflow_core/detections_consensus@v1to add the block as as step in your workflow.

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
required_votes	int	Required number of votes for single detection from different models to accept detection as output detection.	✅
class_aware	bool	Flag to decide if merging detections is class-aware or only bounding boxes aware.	✅
iou_threshold	float	IoU threshold to consider detections from different models as matching (increasing votes for region).	✅
confidence	float	Confidence threshold for merged detections.	✅
classes_to_consider	List[str]	Optional list of classes to consider in consensus procedure..	✅
required_objects	Optional[Dict[str, int], int]	If given, it holds the number of objects that must be present in merged results, to assume that object presence is reached. Can be selector to InferenceParameter, integer value or dictionary with mapping of class name into minimal number of merged detections of given class to assume consensus..	✅
presence_confidence_aggregation	AggregationMode	Mode dictating aggregation of confidence scores and classes both in case of object presence deduction procedure..	❌
detections_merge_confidence_aggregation	AggregationMode	Mode dictating aggregation of confidence scores and classes both in case of boxes consensus procedure. One of average, max, min. Default: average. While using for merging overlapping boxes, against classes - average equals to majority vote, max - for the class of detection with max confidence, min - for the class of detection with min confidence..	❌
detections_merge_coordinates_aggregation	AggregationMode	Mode dictating aggregation of bounding boxes. One of average, max, min. Default: average. average means taking mean from all boxes coordinates, min - taking smallest box, max - taking largest box..	❌

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Available Connections

Compatible Blocks

Check what blocks you can connect to Detections Consensus in version v1.

• inputs: Detection Offset, Roboflow Custom Metadata, Size Measurement, Buffer, LMM, VLM as Detector, Distance Measurement, Detections Classes Replacement, Gaze Detection, Image Contours, Keypoint Detection Model, Instance Segmentation Model, SIFT Comparison, Object Detection Model, Path Deviation, Line Counter, Detections Stabilizer, Detections Transformation, Byte Tracker, Google Vision OCR, Roboflow Dataset Upload, Dynamic Zone, Llama 3.2 Vision, Clip Comparison, Perspective Correction, Object Detection Model, Webhook Sink, Identify Changes, Detections Filter, Email Notification, Instance Segmentation Model, Dimension Collapse, Slack Notification, Qwen2.5-VL, Detections Merge, Time in Zone, Time in Zone, Line Counter, Local File Sink, CogVLM, Keypoint Detection Model, Overlap Filter, JSON Parser, OpenAI, Moondream2, Path Deviation, Florence-2 Model, Twilio SMS Notification, Bounding Rectangle, Detections Stitch, Template Matching, Byte Tracker, SmolVLM2, OpenAI, Google Gemini, SIFT Comparison, Pixel Color Count, YOLO-World Model, VLM as Detector, Velocity, Roboflow Dataset Upload, Segment Anything 2 Model, VLM as Classifier, Model Monitoring Inference Aggregator, Clip Comparison, VLM as Classifier, Identify Outliers, Anthropic Claude, Florence-2 Model, Byte Tracker, Dynamic Crop, Detections Consensus
• outputs: Size Measurement, Detection Offset, Roboflow Custom Metadata, Multi-Label Classification Model, Distance Measurement, Relative Static Crop, Line Counter Visualization, Detections Classes Replacement, Gaze Detection, Background Color Visualization, Image Slicer, Keypoint Detection Model, Reference Path Visualization, Instance Segmentation Model, SIFT Comparison, Object Detection Model, Triangle Visualization, Path Deviation, Line Counter, Detections Stabilizer, Detections Transformation, Multi-Label Classification Model, Byte Tracker, Roboflow Dataset Upload, Perspective Correction, Object Detection Model, Crop Visualization, Webhook Sink, Identify Changes, Dot Visualization, Detections Filter, Model Comparison Visualization, Email Notification, Classification Label Visualization, Instance Segmentation Model, Slack Notification, Detections Merge, Stability AI Image Generation, Time in Zone, Line Counter, Time in Zone, Trace Visualization, Corner Visualization, Blur Visualization, Keypoint Detection Model, Overlap Filter, Circle Visualization, Path Deviation, Florence-2 Model, Twilio SMS Notification, Label Visualization, Stitch Images, Detections Stitch, Template Matching, Byte Tracker, Polygon Zone Visualization, Keypoint Visualization, Stitch OCR Detections, Bounding Box Visualization, Halo Visualization, Ellipse Visualization, Color Visualization, Pixelate Visualization, Single-Label Classification Model, YOLO-World Model, Velocity, Roboflow Dataset Upload, Segment Anything 2 Model, Polygon Visualization, Single-Label Classification Model, Model Monitoring Inference Aggregator, Image Slicer, Mask Visualization, Florence-2 Model, Identify Outliers, Byte Tracker, Dynamic Crop, Detections Consensus

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Detections Consensus in version v1 has.

Bindings

• input

  • predictions_batches (Union[instance_segmentation_prediction, object_detection_prediction, keypoint_detection_prediction]): Reference to detection-like model predictions made against single image to agree on model consensus.
  • required_votes (integer): Required number of votes for single detection from different models to accept detection as output detection.
  • class_aware (boolean): Flag to decide if merging detections is class-aware or only bounding boxes aware.
  • iou_threshold (float_zero_to_one): IoU threshold to consider detections from different models as matching (increasing votes for region).
  • confidence (float_zero_to_one): Confidence threshold for merged detections.
  • classes_to_consider (list_of_values): Optional list of classes to consider in consensus procedure..
  • required_objects (Union[dictionary, integer]): If given, it holds the number of objects that must be present in merged results, to assume that object presence is reached. Can be selector to InferenceParameter, integer value or dictionary with mapping of class name into minimal number of merged detections of given class to assume consensus..

• output

  • predictions (object_detection_prediction): Prediction with detected bounding boxes in form of sv.Detections(...) object.
  • object_present (Union[boolean, dictionary]): Boolean flag if boolean or Dictionary if dictionary.
  • presence_confidence (Union[float_zero_to_one, dictionary]): float value in range [0.0, 1.0] if float_zero_to_one or Dictionary if dictionary.

Example JSON definition of step Detections Consensus in version v1

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/detections_consensus@v1",
    "predictions_batches": [
        "$steps.a.predictions",
        "$steps.b.predictions"
    ],
    "required_votes": 2,
    "class_aware": true,
    "iou_threshold": 0.3,
    "confidence": 0.1,
    "classes_to_consider": [
        "a",
        "b"
    ],
    "required_objects": 3,
    "presence_confidence_aggregation": "max",
    "detections_merge_confidence_aggregation": "min",
    "detections_merge_coordinates_aggregation": "min"
}