Example Workflows - Workflows with classical Computer Vision methods¶
Below you can find example workflows you can use as inspiration to build your apps.
Workflow generating camera focus measure¶
In this example, we demonstrate how to evaluate camera focus using a specific block.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
}
],
"steps": [
{
"type": "roboflow_core/camera_focus@v1",
"name": "camera_focus",
"image": "$inputs.image"
}
],
"outputs": [
{
"type": "JsonField",
"name": "camera_focus_image",
"coordinates_system": "own",
"selector": "$steps.camera_focus.image"
},
{
"type": "JsonField",
"name": "camera_focus_measure",
"selector": "$steps.camera_focus.focus_measure"
}
]
}
Workflow detecting contours¶
In this example we show how classical contour detection works in cooperation with blocks performing its pre-processing (conversion to gray and blur).
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
}
],
"steps": [
{
"type": "roboflow_core/convert_grayscale@v1",
"name": "image_convert_grayscale",
"image": "$inputs.image"
},
{
"type": "roboflow_core/image_blur@v1",
"name": "image_blur",
"image": "$steps.image_convert_grayscale.image"
},
{
"type": "roboflow_core/threshold@v1",
"name": "image_threshold",
"image": "$steps.image_blur.image",
"thresh_value": 200,
"threshold_type": "binary_inv"
},
{
"type": "roboflow_core/contours_detection@v1",
"name": "image_contours",
"image": "$steps.image_threshold.image",
"raw_image": "$inputs.image",
"line_thickness": 5
}
],
"outputs": [
{
"type": "JsonField",
"name": "number_contours",
"coordinates_system": "own",
"selector": "$steps.image_contours.number_contours"
},
{
"type": "JsonField",
"name": "contour_image",
"coordinates_system": "own",
"selector": "$steps.image_contours.image"
},
{
"type": "JsonField",
"name": "contours",
"coordinates_system": "own",
"selector": "$steps.image_contours.contours"
},
{
"type": "JsonField",
"name": "grayscale_image",
"coordinates_system": "own",
"selector": "$steps.image_convert_grayscale.image"
},
{
"type": "JsonField",
"name": "blurred_image",
"coordinates_system": "own",
"selector": "$steps.image_blur.image"
},
{
"type": "JsonField",
"name": "thresholded_image",
"coordinates_system": "own",
"selector": "$steps.image_threshold.image"
}
]
}
Workflow calculating pixels with dominant color¶
This example shows how Dominant Color block and Pixel Color Count block can be used together.
First, dominant color gets detected and then number of pixels with that color is calculated.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
}
],
"steps": [
{
"type": "roboflow_core/dominant_color@v1",
"name": "dominant_color",
"image": "$inputs.image"
},
{
"type": "roboflow_core/pixel_color_count@v1",
"name": "pixelation",
"image": "$inputs.image",
"target_color": "$steps.dominant_color.rgb_color"
}
],
"outputs": [
{
"type": "JsonField",
"name": "matching_pixels_count",
"coordinates_system": "own",
"selector": "$steps.pixelation.matching_pixels_count"
}
]
}
Workflow calculating dominant color¶
This example shows how Dominant Color block can be used against input image.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
}
],
"steps": [
{
"type": "roboflow_core/dominant_color@v1",
"name": "dominant_color",
"image": "$inputs.image"
}
],
"outputs": [
{
"type": "JsonField",
"name": "color",
"coordinates_system": "own",
"selector": "$steps.dominant_color.rgb_color"
}
]
}
Workflow resizing the input image¶
This example shows how the Image Preprocessing block can be used to resize an input image.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
}
],
"steps": [
{
"type": "roboflow_core/image_preprocessing@v1",
"name": "resize_image",
"image": "$inputs.image",
"task_type": "resize",
"width": 1000,
"height": 800
}
],
"outputs": [
{
"type": "JsonField",
"name": "resized_image",
"coordinates_system": "own",
"selector": "$steps.resize_image.image"
}
]
}
SIFT in Workflows¶
In this example we check how SIFT-based pattern matching works in cooperation with expression block.
The Workflow first calculates SIFT features for input image and reference template, then image features are compared to template features. At the end - switch-case statement is built with Expression block to produce output.
Important detail: If there is empty output from SIFT descriptors calculation
for (which is a valid output if no feature gets recognised) the sift comparison won't
execute - hence First Non Empty Or Default block is used to provide default outcome
for images_match output of SIFT comparison block.
Please note that a single image can be passed as template, and batch of images are passed as images to look for template. This workflow does also validate Execution Engine capabilities to broadcast batch-oriented inputs properly.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image"
},
{
"type": "InferenceImage",
"name": "template"
}
],
"steps": [
{
"type": "roboflow_core/sift@v1",
"name": "image_sift",
"image": "$inputs.image"
},
{
"type": "roboflow_core/sift@v1",
"name": "template_sift",
"image": "$inputs.template"
},
{
"type": "roboflow_core/sift_comparison@v1",
"name": "sift_comparison",
"descriptor_1": "$steps.image_sift.descriptors",
"descriptor_2": "$steps.template_sift.descriptors",
"good_matches_threshold": 50
},
{
"type": "roboflow_core/first_non_empty_or_default@v1",
"name": "empty_values_replacement",
"data": [
"$steps.sift_comparison.images_match"
],
"default": false
},
{
"type": "roboflow_core/expression@v1",
"name": "is_match_expression",
"data": {
"images_match": "$steps.empty_values_replacement.output"
},
"switch": {
"type": "CasesDefinition",
"cases": [
{
"type": "CaseDefinition",
"condition": {
"type": "StatementGroup",
"statements": [
{
"type": "UnaryStatement",
"operand": {
"type": "DynamicOperand",
"operand_name": "images_match"
},
"operator": {
"type": "(Boolean) is True"
}
}
]
},
"result": {
"type": "StaticCaseResult",
"value": "MATCH"
}
}
],
"default": {
"type": "StaticCaseResult",
"value": "NO MATCH"
}
}
}
],
"outputs": [
{
"type": "JsonField",
"name": "result",
"coordinates_system": "own",
"selector": "$steps.is_match_expression.output"
}
]
}
Workflow stitching images¶
In this example two images of the same scene are stitched together. Given enough shared details order of the images does not influence final result.
Please note that images need to have enough common details for the algorithm to stitch them properly.
Workflow definition
{
"version": "1.0",
"inputs": [
{
"type": "InferenceImage",
"name": "image1"
},
{
"type": "InferenceImage",
"name": "image2"
},
{
"type": "InferenceParameter",
"name": "count_of_best_matches_per_query_descriptor"
},
{
"type": "InferenceParameter",
"name": "max_allowed_reprojection_error"
}
],
"steps": [
{
"type": "roboflow_core/stitch_images@v1",
"name": "stitch_images",
"image1": "$inputs.image1",
"image2": "$inputs.image2",
"count_of_best_matches_per_query_descriptor": "$inputs.count_of_best_matches_per_query_descriptor",
"max_allowed_reprojection_error": "$inputs.max_allowed_reprojection_error"
}
],
"outputs": [
{
"type": "JsonField",
"name": "stitched_image",
"selector": "$steps.stitch_images.stitched_image"
}
]
}