Skip to main content

Binary Classification Detectors

Binary classification detectors are used to answer yes/no questions about images. Most of Groundlight's documentation examples are for binary classification detectors, as they are the simplest type of detector.

In order to create a binary classification detector, you need to provide a query that asks a yes/no question. For example, "Is there an eagle visible?" or "Is the door fully closed?".

from groundlight import Groundlight
gl = Groundlight()

detector = gl.create_detector(
    name="eagle-detector",
    query="Is there an eagle visible?",
    confidence_threshold=0.9,
)

Submit an Image Query to a Binary Classification Detector​

Now that you have created a binary classification detector, you can submit an image query to it.

from groundlight import Groundlight
gl = Groundlight()

detector = gl.get_detector_by_name("eagle-detector")

# Check if an eagle is visible in an image
image_query = gl.submit_image_query(detector, "path/to/image.jpg")

print(f"Result: {image_query.result.label}")
print(f"Confidence: {image_query.result.confidence}")

Binary classification detectors return a label attribute in the result object, which will be either "YES" or "NO". If a query is ambiguous, it is also possible for the detector to return an "UNCLEAR" label.

The confidence attribute represents the confidence level in the predicted label, which (for a binary classification detector) is a value between 0.5 and 1. A higher confidence score indicates that the model is more certain about its prediction.

Add a label to a Binary Classification Detector​

To provide ground truth labels for binary classification detectors, you can specify the label as either "YES", "NO", or "UNCLEAR". This helps improve the accuracy of your detector over time.

from groundlight import Groundlight
gl = Groundlight()

# Add a binary label to the image query from the previous example
gl.add_label(image_query, label="YES")