So far, we have only dealt with image classification, but that is perhaps the most general task in image processing. In many cases, we would like to not only predict some global descriptors for the image, but also get some information about the contents of various subregions. Segmentation is a common task in biomedical and industrial applications where we would like to locate objects of interest at the level of individual pixels.

Goals:

• Gain experience with image segmentation tasks.
• Examine the U-Net and VGG16 architectures.
• Learn about transfer learning and using pre-trained feature extractors.