Dynamic Crop

Class: DynamicCropBlockV1

Source: inference.core.workflows.core_steps.transformations.dynamic_crop.v1.DynamicCropBlockV1

Create dynamic crops from an image based on detections from detections-based model.

This is useful when placed after an ObjectDetection block as part of a multi-stage workflow. For example, you could use an ObjectDetection block to detect objects, then the DynamicCropBlock block to crop objects, then an OCR block to run character recognition on each of the individual cropped regions.

In addition, for instance segmentation predictions (which provide segmentation mask for each bounding box) it is possible to remove background in the crops, outside of detected instances. To enable that functionality, set mask_opacity to positive value and optionally tune background_color.

Type identifier

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

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
mask_opacity	float	For instance segmentation, mask_opacity can be used to control background removal. Opacity 1.0 removes the background, while 0.0 leaves the crop unchanged..	✅
background_color	Union[Tuple[int, int, int], str]	For background removal based on segmentation mask, new background color can be selected. Can be a hex string (like '#431112') RGB string (like '(128, 32, 64)') or a RGB tuple (like (18, 17, 67))..	✅

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 Dynamic Crop in version v1.

• inputs: Byte Tracker, Polygon Zone Visualization, Time in Zone, Identify Outliers, LMM For Classification, VLM as Classifier, Dot Visualization, Morphological Transformation, Blur Visualization, Perspective Correction, CSV Formatter, Corner Visualization, LMM, PTZ Tracking (ONVIF).md), Florence-2 Model, Grid Visualization, Image Threshold, Florence-2 Model, Halo Visualization, OpenAI, Keypoint Detection Model, Multi-Label Classification Model, Detections Combine, Detection Offset, CogVLM, EasyOCR, Byte Tracker, Line Counter Visualization, Stitch OCR Detections, VLM as Detector, Stability AI Outpainting, Twilio SMS Notification, Keypoint Detection Model, Google Vision OCR, Identify Changes, Camera Focus, Roboflow Dataset Upload, SIFT, Instance Segmentation Model, Clip Comparison, Image Slicer, Image Convert Grayscale, Keypoint Visualization, Email Notification, Template Matching, OCR Model, Detections Stabilizer, Bounding Box Visualization, Instance Segmentation Model, Llama 3.2 Vision, Reference Path Visualization, Dynamic Crop, Roboflow Dataset Upload, Mask Visualization, Image Preprocessing, Background Color Visualization, Local File Sink, Webhook Sink, Camera Calibration, OpenAI, Depth Estimation, Image Slicer, QR Code Generator, Detections Stitch, Trace Visualization, Time in Zone, Velocity, Dominant Color, Object Detection Model, Detections Transformation, Contrast Equalization, Byte Tracker, Overlap Filter, Crop Visualization, Stability AI Image Generation, Moondream2, SIFT Comparison, Roboflow Custom Metadata, Object Detection Model, Model Comparison Visualization, Pixelate Visualization, Dynamic Zone, Model Monitoring Inference Aggregator, Gaze Detection, Line Counter, Anthropic Claude, Time in Zone, Relative Static Crop, Image Contours, Polygon Visualization, Path Deviation, OpenAI, Slack Notification, Triangle Visualization, Detections Filter, YOLO-World Model, Classification Label Visualization, Detections Classes Replacement, Bounding Rectangle, Circle Visualization, Image Blur, Label Visualization, VLM as Detector, Google Gemini, Absolute Static Crop, Stability AI Inpainting, Icon Visualization, Ellipse Visualization, Color Visualization, Detections Merge, Path Deviation, Single-Label Classification Model, Detections Consensus, Stitch Images, Segment Anything 2 Model
• outputs: Byte Tracker, VLM as Classifier, LMM For Classification, Polygon Zone Visualization, Distance Measurement, Time in Zone, Dot Visualization, Morphological Transformation, Size Measurement, Perspective Correction, Blur Visualization, Clip Comparison, Corner Visualization, LMM, Pixel Color Count, Florence-2 Model, PTZ Tracking (ONVIF).md), Image Threshold, Florence-2 Model, Halo Visualization, OpenAI, Keypoint Detection Model, Multi-Label Classification Model, CogVLM, Detections Combine, Detection Offset, EasyOCR, Byte Tracker, Line Counter Visualization, Perception Encoder Embedding Model, Stitch OCR Detections, VLM as Detector, Stability AI Outpainting, VLM as Classifier, Keypoint Detection Model, Google Vision OCR, CLIP Embedding Model, Camera Focus, Roboflow Dataset Upload, SIFT, Instance Segmentation Model, Clip Comparison, Image Slicer, Image Convert Grayscale, Keypoint Visualization, Template Matching, OCR Model, Llama 3.2 Vision, Bounding Box Visualization, Detections Stabilizer, Instance Segmentation Model, Line Counter, Reference Path Visualization, Dynamic Crop, Roboflow Dataset Upload, Mask Visualization, Image Preprocessing, Qwen2.5-VL, Background Color Visualization, OpenAI, Camera Calibration, Depth Estimation, Image Slicer, Detections Stitch, Dominant Color, Trace Visualization, Time in Zone, Velocity, Object Detection Model, Detections Transformation, Contrast Equalization, Buffer, Byte Tracker, Overlap Filter, Crop Visualization, Stability AI Image Generation, Stitch Images, SmolVLM2, SIFT Comparison, Roboflow Custom Metadata, Object Detection Model, Model Comparison Visualization, Pixelate Visualization, Dynamic Zone, Model Monitoring Inference Aggregator, Gaze Detection, Line Counter, Anthropic Claude, Time in Zone, QR Code Detection, Relative Static Crop, Polygon Visualization, Barcode Detection, Image Contours, OpenAI, Path Deviation, Triangle Visualization, Detections Filter, YOLO-World Model, Single-Label Classification Model, Classification Label Visualization, Detections Classes Replacement, Bounding Rectangle, Circle Visualization, Image Blur, Label Visualization, VLM as Detector, Google Gemini, Multi-Label Classification Model, Stability AI Inpainting, Absolute Static Crop, Icon Visualization, Ellipse Visualization, Color Visualization, Detections Merge, Path Deviation, Single-Label Classification Model, Detections Consensus, Moondream2, Segment Anything 2 Model

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Dynamic Crop in version v1 has.

Bindings

• input

  • images (image): The input image for this step..
  • predictions (Union[object_detection_prediction, instance_segmentation_prediction, keypoint_detection_prediction]): Detection model output containing bounding boxes for cropping..
  • mask_opacity (float_zero_to_one): For instance segmentation, mask_opacity can be used to control background removal. Opacity 1.0 removes the background, while 0.0 leaves the crop unchanged..
  • background_color (Union[rgb_color, string]): For background removal based on segmentation mask, new background color can be selected. Can be a hex string (like '#431112') RGB string (like '(128, 32, 64)') or a RGB tuple (like (18, 17, 67))..

• output

  • crops (image): Image in workflows.
  • predictions (Union[object_detection_prediction, instance_segmentation_prediction, keypoint_detection_prediction]): Prediction with detected bounding boxes in form of sv.Detections(...) object if object_detection_prediction or Prediction with detected bounding boxes and segmentation masks in form of sv.Detections(...) object if instance_segmentation_prediction or Prediction with detected bounding boxes and detected keypoints in form of sv.Detections(...) object if keypoint_detection_prediction.

Example JSON definition of step Dynamic Crop in version v1

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/dynamic_crop@v1",
    "images": "$inputs.image",
    "predictions": "$steps.my_object_detection_model.predictions",
    "mask_opacity": "<block_does_not_provide_example>",
    "background_color": "#431112"
}