Understanding Vision with Language
Visual understanding consists of inferring scene properties from images. Some properties might refer to mid-level scene attributes such as motion or depth, while others may relate to high-level features like semantic segmentation.
In this part, we will focus on the semantics of visual processing; that is, the association between visual stimuli and meaning. The goal is to infer from visual input what the significance of what we see is. Therefore, there is a strong connection between semantic visual processing and natural language processing.
Outline
Chapter 50 Object Recognition describes how to learn to recognize and localize objects in images, assigning words to them.
Chapter 51.3 Learning Visual Representations from Language Supervision explores the role of language as a representation of the visual world and its connection with vision systems.