Instance Segmentation
Use Artificial intelligence to detect and delineate objects in an image.
Similar to semantic segmentation the goal of instance segmentation is to be able to predict what class of every pixel in an image, but it does this by determining the contours at a pixel level. This enables it to differentiate between two overlapping objects of the same class.
Instance Segmentation is applied to picture of traffic, source
What is Instance Segmentation used for?
So instance segmentation provides the best of both worlds in terms of object detection and semantic segmentation where one can identify individual objects on a pixel level. Some potential use cases of this are
- Self-driving cars: the fine-grained results of instance segmentation provides better decision-making foundations in comparison to both object detection and semantic segmentation, with so many different objects all moving independently,
- Medical scans: Many scans want to identify individual objects of the same type, and since medical images commonly involve 3D structures (organs) represented in 2D, it is prone to overlapping objects.