mIoU

Gold definitionUpdated Apr 2, 2026

Definition

Mean Intersection over Union (mIoU) is a standard evaluation metric for semantic segmentation, quantifying the average overlap between predicted segmentation masks and ground truth masks across all classes. It provides a robust measure of pixel-level accuracy.

At a glance

Executive summary

Mean Intersection over Union (mIoU) is a key metric for judging how accurately AI models can outline objects in images, like detecting changes in forests. It works by measuring the overlap between the model's prediction and the true outline, averaging this score across all types of objects.

TL;DR

mIoU is a common score that tells you how well an AI model can draw accurate outlines around different objects in an image.

Key points

Calculates the average overlap between predicted and ground truth segmentation masks across all classes.
Solves the problem of robustly evaluating pixel-level accuracy in semantic segmentation, especially with class imbalance.
Used by researchers and engineers in computer vision, remote sensing, and medical imaging.
More robust than pixel accuracy as it averages per-class scores, preventing dominant classes from skewing results.
Remains a foundational and widely accepted metric for benchmarking new segmentation architectures and techniques.

Use cases

Evaluating autonomous driving systems for accurate lane and pedestrian segmentation.
Assessing medical imaging models for precise tumor or organ segmentation in diagnostics.
Benchmarking land cover mapping models in remote sensing for environmental monitoring.
Measuring the accuracy of robotic perception systems in segmenting objects for manipulation.
Quantifying the performance of models detecting forest change or deforestation from satellite imagery.

Also known as

Mean IoU, Jaccard Index (for a single class), Jaccard Similarity Coefficient