Overview

Image segmentation provides a more granular understanding of an image than object detection. Instead of bounding boxes, it assigns a label to every single pixel in the image.

Types of Segmentation

  • Semantic Segmentation: All pixels belonging to the same class (e.g., all 'trees') are given the same label, without distinguishing between individual instances.
  • Instance Segmentation: Distinguishes between different objects of the same class (e.g., labeling each individual 'car' separately).
  • Panoptic Segmentation: Combines semantic and instance segmentation to provide a complete pixel-level understanding of the entire scene.

Common Architectures

  • U-Net: Widely used in medical imaging.
  • Mask R-CNN: An extension of Faster R-CNN for instance segmentation.
  • DeepLab: A popular framework for semantic segmentation.

Applications

  • Medical image analysis (segmenting organs or tumors).
  • Autonomous driving (identifying road boundaries, sidewalks).
  • Image editing (background removal).

Related Terms