Semantic Mask

A semantic mask is a pixel-level annotation in which each pixel’s value corresponds to the semantic class of the matching location in the source image.

It is usually stored as a single-channel image, where integer values or color codes map to specific classes such as “road,” “tree,” or “building.”

Semantic masks serve as ground truth for training semantic segmentation models and can be produced through manual labeling, semi-automatic tools, or model-assisted annotation.

Modern platforms such as BasicAI integrate model-assisted segmentation tools to improve efficiency.

Quality is often measured with IoU (Intersection over Union) and mIoU (mean IoU).