Delta Filter

Class: DeltaFilterBlockV1

Source: inference.core.workflows.core_steps.flow_control.delta_filter.v1.DeltaFilterBlockV1

Trigger workflow execution only when an input value changes from its previous state, enabling change detection, avoiding redundant processing when values remain constant, and optimizing system efficiency by executing downstream steps only on state transitions.

How This Block Works

This block monitors a value and only continues workflow execution when that value changes compared to its previous state. The block:

1. Takes an image (for video metadata context) and a value to monitor as input
2. Extracts video metadata from the image to identify the video stream (video_identifier)
3. Retrieves the previously cached value for this video identifier from an internal cache
4. Compares the current input value against the cached previous value
5. If the value has changed (current value ≠ previous value):
6. Updates the cache with the new value for this video identifier
7. Continues execution to the specified next_steps blocks, allowing downstream processing
8. If the value has not changed (current value == previous value):
9. Terminates the current workflow branch, preventing redundant downstream execution
10. Returns flow control directives that either continue to next steps or terminate the branch

The block maintains separate cached values for each video stream (identified by video_identifier), allowing it to track value changes independently across multiple video sources. This per-video tracking ensures that the filter resets appropriately when switching between different video streams. The block supports monitoring any value type (numbers, strings, detection counts, etc.), making it versatile for detecting changes in counters, metrics, detection results, or any other workflow data. By only triggering downstream blocks when values actually change, the Delta Filter prevents unnecessary processing when values remain constant, which is especially useful in video workflows where many frames may have the same detection count or metric value.

Common Use Cases

• Change Detection for Counters: Trigger actions only when counter values change (e.g., execute data logging when line counter count_in changes from 5 to 6, skip processing when count remains at 6), avoiding redundant writes or updates when values are stable
• State Transition Monitoring: Detect transitions in system states or detection results and trigger workflows only on state changes (e.g., execute notification when detection class changes from "empty" to "occupied", skip when state remains "occupied"), preventing repeated actions for the same state
• Conditional Data Logging: Write to databases, CSV files, or external systems only when values change (e.g., log count changes to OPC or PLC systems, skip logging when counts are unchanged), reducing storage and network overhead
• Event-Based Notifications: Send alerts or notifications only when values transition (e.g., trigger email notification when zone count changes, avoid spam when count remains constant), ensuring notifications represent meaningful changes rather than repeated states
• Optimized Processing Pipelines: Reduce computational load in video workflows by skipping downstream processing when monitored values haven't changed (e.g., skip expensive analysis when detection count is unchanged across frames), improving overall workflow efficiency
• Multi-Stream Change Tracking: Monitor value changes independently across multiple video streams (e.g., track zone counts separately for different camera feeds), with automatic per-video caching ensuring correct change detection for each stream

Connecting to Other Blocks

This block monitors values and controls workflow execution flow, and can be connected:

• After counting or metric blocks (e.g., Line Counter, Time in Zone, Velocity, Detection Filter) to detect when counts, metrics, or aggregated values change and conditionally trigger downstream processing based on value transitions
• After detection blocks (e.g., Object Detection, Classification, Keypoint Detection) to monitor detection results, class changes, or confidence metrics and execute actions only when detection outcomes change from previous frames
• After data processing blocks (e.g., Property Definition, Expression, Delta Filter) to track computed values or processed metrics and trigger workflows only when these computed values transition, avoiding redundant processing
• Before data storage blocks (e.g., Local File Sink, CSV Formatter, Roboflow Dataset Upload, Webhook Sink) to conditionally log or store data only when monitored values change, preventing duplicate entries or unnecessary writes when values remain constant
• Before notification blocks (e.g., Email Notification, Slack Notification, Twilio SMS Notification) to trigger alerts only when meaningful changes occur (e.g., count changes, state transitions), avoiding notification spam when values are stable
• In video processing workflows where per-frame values may remain constant for many frames, using the block to efficiently detect changes and trigger expensive downstream operations only when necessary, optimizing resource usage

Type identifier

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

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌

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 Delta Filter in version v1.

• inputs: Icon Visualization, Image Preprocessing, LMM, Blur Visualization, Detections Classes Replacement, Detections Merge, Color Visualization, Contrast Equalization, Cache Set, Llama 3.2 Vision, Velocity, Reference Path Visualization, SIFT, Buffer, OpenAI, SAM 3, Halo Visualization, Roboflow Dataset Upload, Dimension Collapse, Trace Visualization, Twilio SMS/MMS Notification, Detections Transformation, Single-Label Classification Model, VLM as Detector, Image Convert Grayscale, Path Deviation, Background Color Visualization, Multi-Label Classification Model, Qwen2.5-VL, Single-Label Classification Model, Camera Calibration, VLM as Detector, Triangle Visualization, Dynamic Zone, Ellipse Visualization, Seg Preview, Slack Notification, Absolute Static Crop, Google Gemini, Time in Zone, Webhook Sink, Line Counter Visualization, Florence-2 Model, Dominant Color, Property Definition, Detections Consensus, QR Code Generator, Anthropic Claude, Object Detection Model, Keypoint Detection Model, Byte Tracker, Image Slicer, Pixel Color Count, Image Contours, Stability AI Inpainting, VLM as Classifier, Line Counter, Google Gemini, Motion Detection, SIFT Comparison, Line Counter, Byte Tracker, Dot Visualization, Camera Focus, Keypoint Detection Model, YOLO-World Model, Bounding Box Visualization, OCR Model, Background Subtraction, OpenAI, SAM 3, Dynamic Crop, Qwen3-VL, Distance Measurement, Keypoint Visualization, Email Notification, Image Threshold, Anthropic Claude, Corner Visualization, Rate Limiter, Pixelate Visualization, SmolVLM2, SIFT Comparison, Twilio SMS Notification, Moondream2, Morphological Transformation, Roboflow Custom Metadata, Stitch Images, Detections Filter, Circle Visualization, Stability AI Image Generation, Image Blur, Template Matching, Detections List Roll-Up, Bounding Rectangle, Email Notification, Continue If, Perception Encoder Embedding Model, Detection Event Log, EasyOCR, Google Gemini, Environment Secrets Store, Instance Segmentation Model, Delta Filter, Classification Label Visualization, Detection Offset, Clip Comparison, CogVLM, Google Vision OCR, Data Aggregator, Stitch OCR Detections, QR Code Detection, Segment Anything 2 Model, LMM For Classification, Text Display, CLIP Embedding Model, SAM 3, Multi-Label Classification Model, Barcode Detection, Mask Visualization, OpenAI, Local File Sink, Anthropic Claude, Polygon Visualization, Polygon Zone Visualization, Model Comparison Visualization, JSON Parser, Label Visualization, Perspective Correction, Overlap Filter, Image Slicer, Identify Changes, Cosine Similarity, Instance Segmentation Model, Stability AI Outpainting, VLM as Classifier, Grid Visualization, Relative Static Crop, CSV Formatter, Size Measurement, Path Deviation, Camera Focus, Time in Zone, Florence-2 Model, Identify Outliers, Object Detection Model, Detections Stitch, Cache Get, First Non Empty Or Default, Crop Visualization, Clip Comparison, Expression, Byte Tracker, Detections Stabilizer, Roboflow Dataset Upload, PTZ Tracking (ONVIF).md), Model Monitoring Inference Aggregator, Time in Zone, Detections Combine, Depth Estimation, OpenAI, Gaze Detection
• outputs: None

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Delta Filter in version v1 has.

Bindings

• input

  • image (image): not available.
  • value (*): Value to monitor for changes. Can be any data type (numbers, strings, detection counts, metrics, etc.) from workflow inputs or step outputs. The workflow branch continues to next_steps only when this value differs from the previously cached value for the current video stream. If the value remains the same, the branch terminates to avoid redundant processing. Example: Monitor a line counter count ($steps.line_counter.count_in) and trigger actions only when the count changes..
  • next_steps (step): List of workflow steps to execute when the monitored value changes from its previous state. These steps receive control flow only when a change is detected, allowing conditional downstream processing. If the value hasn't changed, these steps will not execute as the branch terminates. Each step selector references a block in the workflow that should execute on value transitions..

• output

Example JSON definition of step Delta Filter in version v1

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/delta_filter@v1",
    "image": "<block_does_not_provide_example>",
    "value": "$steps.line_counter.count_in",
    "next_steps": "$steps.write_to_csv"
}