Image segmentation can provide lots of information about the contents of images, but it is sometimes detrimental to simply assign class labels to every pixel. In some cases, we would also like to separate object instances, i.e. retrieve the pixels belonging to one specific object as opposed to all objects in a given class. There are many ways to approach this problem, but most start with detecting objects within the image, then performing segmentation at a finer level.

Goals:

• Learn how to use and train Single-Shot Detector (SSD) architectures.
• Learn about the importance of detection anchors, as well as designing, generating and storing them.
• (Optional) Become familiar with the Google Mediapipe framework.