DiceIoU

DiceIoU computes foreground pixel overlap metrics for binary masks or label images. Any non-zero pixel is treated as foreground.

Inputs are predicted_label_image and reference_label_image. Outputs include true-positive, false-positive, and false-negative pixel counts plus foreground IoU and Dice.

Minimal Example

from bioimageflow_core import Arguments
from bioimageflow_measurement_tools import DiceIoU

metrics = DiceIoU().process_row(
    Arguments(predicted_label_image="predicted.tif", reference_label_image="truth.tif")
)

Inputs

• predicted_label_image: predicted mask or label image.

• reference_label_image: reference mask or label image with the same shape.

Outputs

• dice, iou, true_positive_pixels, false_positive_pixels, and false_negative_pixels.

Dependencies and Core Libraries

imageio and NumPy for foreground-mask overlap calculations.

Assumptions

All non-zero pixels are treated as foreground. Object identity is ignored.

Expected Results

Synthetic binary masks produce exact pixel counts and deterministic Dice/IoU values.

Failure Modes

Shape mismatches or unreadable images raise errors. Empty foreground in both images reports perfect agreement by convention.