Similar to deep semantic segmentation, deep instance segmentation also assigns a label to each pixel in an image. The labels collectively produce pixel-based masks for each object in an input image. The difference between these two techniques is that even if multiple objects have the same class label (for example, two cats and a dog in the input image shown in the following figure), the instance segmentation should report each object instance as a unique one (for example, a total of three unique objects: two cats and a dog), as opposed to the semantic segmentation that reports the total number of unique class labels found (for example, two unique classes, namely a cat and dog), as shown in the following screenshot:
In this recipe, you will learn how to use a pretrained mask R-CNN deep learning model to perform instance segmentation.
The region-based...
