Instance segmentation

Instance segmentation is a computer vision task that identifies and delineates individual object instances in an image at the pixel level. Key aspects:

Assigns an object label to each pixel in the image.
Separates and segments each object instance.
Provides per-instance masks, not just bounding boxes.
More detailed than object detection or semantic segmentation.
Useful for precise analysis like counting, measurements.

Approaches include:

Two-stage methods like Mask R-CNN that detect then segment.
Single shot methods like YOLACT that jointly infer instances.
Panoptic segmentation unifying instance and semantic methods.

Applications involve robotics, medical imaging, surveillance, and augmented reality. Challenges include handling occlusion and slow inference speeds.

Deep learning driven by large annotated datasets has enabled major advances in instance segmentation accuracy and robustness.