Semantic Segmentation

Semantic segmentation assigns a predefined class label to every pixel in an image without separating individual instances within the same class.

For example, every pixel belonging to a car is labeled “car,” regardless of how many cars appear in the scene.

Common architectures include U-Net, DeepLab, and the Fully Convolutional Network (FCN).

The method focuses on scene composition and region understanding, making it suitable for road and drivable-area detection in autonomous driving, organ and lesion segmentation in medical imaging, and land-cover classification.