Background

The human visual system has a sophisticated recognition mechanism, one that is extremely effective, but is yet to be replicated by automated computer systems. Imagine recognizing a person you met ten years ago from his/her sihouette approaching you from a hundred feet away? How does the human visual system do it? How do you make a computer do the same task?

This course will cover topics related to recognition of objects, activities and their interactions from images and videos, which is an area of great interest within computer vision in recent years. Considering that recent innovations in the area are based on machine learning concepts, the course will highlight the growing intersection between the two areas. The course is intended for anyone interested in the area of recognition and computer vision, or a domain for applying machine learning methods and advancements.

Specific topics that will be covered in the course are listed below.

Topics

  • Introduction, Brief history, Problem definitions, Representation, Learning, Detection, Recognition, Challenges
  • Human visual recognition system; Why is vision difficult?; Retina, LGN and primary visual cortex; Computational models of human visual object recognition
  • Recognition methods: Low-level modeling (e.g. features), Mid-level abstraction (e.g. segmentation), High-level reasoning (e.g. scene understanding); Bag-of-words models, Part-based models, Discriminative models
  • Detection/Segmentation methods: Sliding window approaches, Branch and bound, Structure prediction, Hough voting and neural network approaches, Hierarchical models, CRFs and structured-SVMs for learning
  • Context and scenes, Importance and saliency, Large-scale search and recognition, Egocentric vision systems, Human-in-the-loop interactive systems, 3D scene understanding

Pre-requisites/Eligibility

This course is intended for PhD, MTech and BTech 4th year students. Pre-requisites are below.

  • (Mandatory) Prior coursework in machine learning (CS3140 Machine Learning or equivalent)
  • (Mandatory) Programming experience (for course projects)
  • (Optional, but recommended) Ability and interest to understand and analyze research papers in this area

Resources/References