Image Preprocessing¶

Class: ImagePreprocessingBlockV1

Source: inference.core.workflows.core_steps.classical_cv.image_preprocessing.v1.ImagePreprocessingBlockV1

Apply geometric transformations to images including resizing to specified dimensions (with aspect ratio preservation), rotating by specified degrees (clockwise or counterclockwise), or flipping vertically, horizontally, or both, providing flexible image preprocessing for model input preparation, image orientation correction, and geometric image manipulation workflows.

How This Block Works¶

This block applies one geometric transformation operation (resize, rotate, or flip) to an input image based on the selected task_type. The block:

Receives an input image and selects one transformation task (resize, rotate, or flip)
Validates task-specific parameters (width/height for resize, rotation_degrees for rotate, flip_type for flip)
Applies the selected transformation:

For resize task: - Validates width and height are positive integers (greater than 0) - Supports aspect ratio preservation: if only width or only height is provided, calculates the missing dimension to maintain the original aspect ratio - If both width and height are provided, resizes to exact dimensions (may distort aspect ratio) - Uses OpenCV's INTER_AREA interpolation for high-quality downsampling - Returns resized image with specified dimensions

For rotate task: - Validates rotation_degrees is between -360 and 360 degrees - Positive values rotate clockwise, negative values rotate counterclockwise - Calculates rotation matrix around image center - Automatically adjusts canvas size to contain the rotated image (no cropping) - Uses OpenCV's warpAffine for smooth rotation with bilinear interpolation - Returns rotated image with canvas sized to fit the full rotated image

For flip task: - Validates flip_type is "vertical", "horizontal", or "both" - Vertical flip: flips image upside down (mirrors along horizontal axis) - Horizontal flip: flips image left-right (mirrors along vertical axis) - Both: applies both vertical and horizontal flips simultaneously (180-degree rotation equivalent) - Uses OpenCV's flip function for efficient mirroring - Returns flipped image with same dimensions as input

Preserves image metadata from the original image (parent metadata, image properties)
Returns the transformed image maintaining original image metadata structure

The block performs one transformation at a time - select resize, rotate, or flip via task_type. Each transformation is applied independently and produces a clean output. Resize supports flexible aspect ratio handling, rotation automatically adjusts canvas size to prevent cropping, and flip operations provide efficient mirroring along different axes. The transformations use OpenCV for efficient, high-quality geometric image manipulation.

Common Use Cases¶

Model Input Preparation: Resize images to match model input requirements (e.g., resize images to specific dimensions for object detection models, adjust image sizes for classification model inputs, normalize image dimensions for consistent model processing), enabling proper model input formatting
Image Orientation Correction: Rotate images to correct orientation issues (e.g., rotate images captured in wrong orientation, correct camera rotation, adjust image orientation for proper display), enabling image orientation workflows
Data Augmentation: Apply geometric transformations for data augmentation (e.g., flip images horizontally for augmentation, rotate images for training data variety, apply transformations to increase dataset diversity), enabling data augmentation workflows
Image Display Preparation: Transform images for display or presentation purposes (e.g., flip images for mirror effects, resize images for display dimensions, rotate images for correct viewing orientation), enabling image presentation workflows
Workflow Image Standardization: Standardize image dimensions or orientation across workflow inputs (e.g., resize all images to consistent dimensions, normalize image orientations, prepare images for uniform processing), enabling image standardization workflows
Image Formatting for Downstream Blocks: Prepare images for blocks that require specific dimensions or orientations (e.g., resize before detection models, rotate for proper processing, flip for compatibility with other blocks), enabling image preparation workflows

Connecting to Other Blocks¶

This block receives an image and produces a transformed image:

After image input blocks to preprocess images before further processing (e.g., resize input images, correct image orientation, prepare images for workflow processing), enabling image preprocessing workflows
Before detection or classification models to format images for model requirements (e.g., resize to model input dimensions, adjust orientation for proper detection, prepare images for model processing), enabling model-compatible image preparation
Before crop blocks to prepare images before cropping (e.g., resize before cropping, rotate before region extraction, adjust orientation before cropping), enabling pre-crop image preparation
Before visualization blocks to prepare images for display (e.g., resize for display, rotate for proper viewing, flip for presentation), enabling image display preparation workflows
In image processing pipelines where geometric transformations are needed (e.g., resize in multi-stage pipelines, rotate in processing workflows, flip in transformation chains), enabling geometric transformation pipelines
After other transformation blocks to apply additional geometric operations (e.g., resize after cropping, rotate after other transformations, flip after processing), enabling multi-stage geometric transformation workflows

Type identifier¶

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

Properties¶

Name	Type	Description	Refs
`name`	`str`	Enter a unique identifier for this step..	❌
`task_type`	`str`	Type of geometric transformation to apply to the image: 'resize' to change image dimensions (requires width/height), 'rotate' to rotate the image by specified degrees (requires rotation_degrees), or 'flip' to mirror the image along axes (requires flip_type). Only one transformation is applied per block execution. Select the appropriate task type based on your preprocessing needs..	❌
`width`	`int`	Target width in pixels for resizing. Required when task_type is 'resize'. Must be a positive integer (greater than 0). If only width is provided (height is None), the height is automatically calculated to preserve aspect ratio. If both width and height are provided, the image is resized to exact dimensions (may distort aspect ratio). Default is 640 pixels. Use this to resize images to specific dimensions for model inputs or display requirements..	✅
`height`	`int`	Target height in pixels for resizing. Required when task_type is 'resize'. Must be a positive integer (greater than 0). If only height is provided (width is None), the width is automatically calculated to preserve aspect ratio. If both width and height are provided, the image is resized to exact dimensions (may distort aspect ratio). Default is 640 pixels. Use this to resize images to specific dimensions for model inputs or display requirements..	✅
`rotation_degrees`	`int`	Rotation angle in degrees. Required when task_type is 'rotate'. Must be between -360 and 360 degrees. Positive values rotate the image clockwise, negative values rotate counterclockwise. The rotation is performed around the image center, and the canvas size is automatically adjusted to contain the full rotated image (no cropping occurs). For example, 90 rotates 90 degrees clockwise, -90 rotates 90 degrees counterclockwise, 180 rotates 180 degrees. Default is 90 degrees..	✅
`flip_type`	`str`	Type of flip operation to apply. Required when task_type is 'flip'. Options: 'vertical' flips the image upside down (mirrors along horizontal axis, top becomes bottom), 'horizontal' flips left-right (mirrors along vertical axis, left becomes right), 'both' applies both vertical and horizontal flips simultaneously (equivalent to 180-degree rotation). The image dimensions remain unchanged after flipping. Default is 'vertical'. Use this for mirroring images or data augmentation..	✅

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 Image Preprocessing in version v1.

Input and Output Bindings¶

The available connections depend on its binding kinds. Check what binding kinds Image Preprocessing in version v1 has.

Bindings

input
- image (image): Input image to transform. The image will have one geometric transformation applied (resize, rotate, or flip) based on the selected task_type. Supports images from inputs, previous workflow steps, or crop outputs. The output image maintains the original image's metadata structure..
- width (integer): Target width in pixels for resizing. Required when task_type is 'resize'. Must be a positive integer (greater than 0). If only width is provided (height is None), the height is automatically calculated to preserve aspect ratio. If both width and height are provided, the image is resized to exact dimensions (may distort aspect ratio). Default is 640 pixels. Use this to resize images to specific dimensions for model inputs or display requirements..
- height (integer): Target height in pixels for resizing. Required when task_type is 'resize'. Must be a positive integer (greater than 0). If only height is provided (width is None), the width is automatically calculated to preserve aspect ratio. If both width and height are provided, the image is resized to exact dimensions (may distort aspect ratio). Default is 640 pixels. Use this to resize images to specific dimensions for model inputs or display requirements..
- rotation_degrees (integer): Rotation angle in degrees. Required when task_type is 'rotate'. Must be between -360 and 360 degrees. Positive values rotate the image clockwise, negative values rotate counterclockwise. The rotation is performed around the image center, and the canvas size is automatically adjusted to contain the full rotated image (no cropping occurs). For example, 90 rotates 90 degrees clockwise, -90 rotates 90 degrees counterclockwise, 180 rotates 180 degrees. Default is 90 degrees..
- flip_type (string): Type of flip operation to apply. Required when task_type is 'flip'. Options: 'vertical' flips the image upside down (mirrors along horizontal axis, top becomes bottom), 'horizontal' flips left-right (mirrors along vertical axis, left becomes right), 'both' applies both vertical and horizontal flips simultaneously (equivalent to 180-degree rotation). The image dimensions remain unchanged after flipping. Default is 'vertical'. Use this for mirroring images or data augmentation..
output
- image (image): Image in workflows.

Example JSON definition of step Image Preprocessing in version v1

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/image_preprocessing@v1",
    "image": "$inputs.image",
    "task_type": "<block_does_not_provide_example>",
    "width": 640,
    "height": 640,
    "rotation_degrees": 90,
    "flip_type": "vertical"
}