Identify Outliers

Class: IdentifyOutliersBlockV1

Source: inference.core.workflows.core_steps.sampling.identify_outliers.v1.IdentifyOutliersBlockV1

Identify outlier embeddings compared to prior data.

This block accepts an embedding and compares it to a sample of prior data. If the embedding is an outlier, the block will return a boolean flag and the percentile of the embedding.

Type identifier

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

Properties

Name	Type	Description	Refs
name	str	Unique name of step in workflows.	❌
threshold_percentile	float	The desired sensitivity. A higher value will result in more data points being classified as outliers..	✅
warmup	int	The number of data points to use for the initial average calculation. No outliers are identified during this period..	✅
window_size	int	The number of previous data points to consider in the sliding window algorithm..	✅

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 Identify Outliers in version v1.

• inputs: Identify Outliers, Perspective Correction, Pixel Color Count, Clip Comparison, Identify Changes, SIFT Comparison, Image Contours, Perception Encoder Embedding Model, Distance Measurement, SIFT Comparison, Line Counter, Template Matching, CLIP Embedding Model, Line Counter, Detections Consensus
• outputs: YOLO-World Model, Keypoint Detection Model, Roboflow Dataset Upload, Gaze Detection, Circle Visualization, PTZ Tracking (ONVIF).md), Roboflow Custom Metadata, Single-Label Classification Model, Template Matching, Dynamic Zone, Detections Stitch, Instance Segmentation Model, Color Visualization, Detections Consensus, Object Detection Model, Identify Outliers, Perspective Correction, Model Monitoring Inference Aggregator, Image Slicer, Keypoint Detection Model, Model Comparison Visualization, Stitch Images, Dynamic Crop, Webhook Sink, Pixelate Visualization, Byte Tracker, Reference Path Visualization, Time in Zone, Blur Visualization, Ellipse Visualization, Trace Visualization, Corner Visualization, Polygon Zone Visualization, Triangle Visualization, Stability AI Inpainting, Classification Label Visualization, Single-Label Classification Model, Bounding Box Visualization, Detections Stabilizer, Halo Visualization, Email Notification, Polygon Visualization, Slack Notification, Object Detection Model, Dot Visualization, Label Visualization, Byte Tracker, Stability AI Outpainting, Crop Visualization, Stability AI Image Generation, Identify Changes, SIFT Comparison, Mask Visualization, Time in Zone, Twilio SMS Notification, Segment Anything 2 Model, Roboflow Dataset Upload, Byte Tracker, Line Counter Visualization, Instance Segmentation Model, Background Color Visualization, Multi-Label Classification Model, Image Slicer, Keypoint Visualization, Multi-Label Classification Model, Relative Static Crop

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Identify Outliers in version v1 has.

Bindings

• input

  • embedding (embedding): Embedding of the current data..
  • threshold_percentile (float_zero_to_one): The desired sensitivity. A higher value will result in more data points being classified as outliers..
  • warmup (integer): The number of data points to use for the initial average calculation. No outliers are identified during this period..
  • window_size (integer): The number of previous data points to consider in the sliding window algorithm..

• output

  • is_outlier (boolean): Boolean flag.
  • percentile (float_zero_to_one): float value in range [0.0, 1.0].
  • warming_up (boolean): Boolean flag.

Example JSON definition of step Identify Outliers in version v1

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/identify_outliers@v1",
    "embedding": "$steps.clip.embedding",
    "threshold_percentile": "$inputs.sample_rate",
    "warmup": 100,
    "window_size": 5
}