CS 332 Exam 2 Coverage Exam 2: Friday, November 18

The second exam will be held in class on Friday, November 18. It will be an open book exam – you can refer to any of your handouts, readings, notes, assignments and solutions during the exam. The exam will focus on the recovery of observer motion and 3D scene layout from image motion, object recognition, and face detection and recognition. The recognition component will cover the modeling approaches that we talked about (e.g. feature- or parts-based approaches, the alignment method based on linear combinations of views, PCA-based face recognition, neural nets, and recognition methods based on deep networks (i.e. convolutional neural networks). You should also be familiar with the Viola & Jones face detection method. One problem will include a choice of questions related to the perceptual papers that you read for Assignment 5 (you will only need to answer question(s) related to the paper that you focused on). You do not need to remember specific facts about the neural processing underlying face recognition in the brain, but I may describe a neural behavior and ask you to relate it to a model. The exam will not include any MATLAB programming. The following material will be most useful for preparing for the exam:

Lecture handouts and your lecture notes related to observer motion and recognition — on the course schedule page, this includes classes #14 (not including the human motion slides) and #15, and classes #18-29 (not including color vision)

Assignments #4-6 and the three labs on observer motion, recognition, and eigenfaces (and solutions)

Readings:

1. On the topic of observer motion, pages 305-308 (pages 1-4 of the pdf file) of the following article provide a concise summary of the key computational ideas that we discussed (omit the general overview of early computational methods on page 306):

   Hildreth, E. (1992) Recovering heading for visually guided navigation in the presence of self-moving objects

2. Patrick Winston's Artificial Intelligence text provides concise introductions to the alignment method for recognition that we discussed, and to simple neural networks — see pages 443-453 and 531-539 of the text (pages 369-379 and 433-441 of the long pdf file)
3. For an overview of face processing, the following paper will be distributed in class on Nov. 8, with relevant sections marked:

   Tsao, D. Y. & Livingstone, M. S. (2008) Mechanisms of Face Perception

4. Readings on perceptual studies of face recognition from Assignment 5 — you only need to review the particular paper that you chose as a focus for the assignment, and you do not need to review the Sinha et al. paper
5. The Eigenface wikipedia page provides a short introduction to the Eigenfaces method for face recognition
6. The wikipedia page on the Viola-Jones object detection framework provides a short introduction to this method
7. The Analytics Vidhya page on deep learning for computer vision provides a short introduction to convolutional neural networks (see Sections 1-4, omitting the subsections on tanh and ReLU activation functions, and Xavier Initialization)
8. The first two pages of the following article summarize the HMAX model for rapid object categorization in the ventral visual pathway:

   Serre, T., Oliva, A. & Poggio, T. (2007) A feedforward architecture accounts for rapid categorization

We will hold a review for the exam on Wednesday, November 16th.