Panoptic Segmentation

Semantic segmentation tells you which pixels belong to "car" versus "road." It doesn't tell you which car. Panoptic segmentation combines that category-level labeling with instance-level detail, so every individual object gets its own boundary even when it shares a class with ten others in the frame. It's the harder, more complete version of segmentation that production computer vision systems increasingly need.

‍

‍