Segmentation is often referred to as the clustering of pixels of a similar category. An example is as shown in the following screenshot. Here, we see that inputs are on the left and the segmentation results are on the right. The colors of an object are according to pre-defined object categories. These examples are taken from the Pascal VOC dataset:
In the top picture on the left, there are several small aeroplanes in the background and, therefore, we see small pixels colored accordingly in the corresponding image on the right. In the bottom-left picture, there are two pets laying together, therefore, their segmented image on the right has different colors for the pixels belonging to the cat and dog respectively. In this figure, the boundary is differently colored for convenience and does not imply a different category.
In traditional segmentation techniques, the...