Several Instance Segmentation

Instance Segmentation is the problem of identifying and outlying individual instances of one or several semantic classes in an image. The number of instances are usually not known in advance, and the visual appearance of the instances may be very similar. Instance segmentation is therefore often regarded as a harder counterpart to the semantic segmentation problem that attributes pixels to a known number of classes that usually have distinct visual appearance.

Two Stage

Mask RCNN 2017

PANet

One Stage

Associative Embedding

Deep Coloring

TernausNetV2