29
Vision (II) CSCI 5582, Fall 2007

AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Page 1: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

Vision (II)

CSCI 5582, Fall 2007

Page 2: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

Assignments

• For this week: Chapter 24• Poggio article will be posted at the website

today• Problem Set 3 is due Thursday• Problem Set 1 handed back today

Page 3: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

What Is It Like to Be a Bat?

Page 4: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

The Problems of Bathood

• How do you avoid hurting your own ears?• How can you estimate which objects are

closer to you and which are farther away?• How can you tell if an object is moving

toward or away from you?

Page 5: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 6: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

David Marr’s Levels ofAbstraction

• Computational Level: What is the goal of the computation, whyis it appropriate, and what is the logic of the strategy by which itcan be carried out?

• Algorithm Level: How can this computational theory beimplemented? In particular,what is the representation for theinput and output, and what is the algorithm for thetransformation?

• Mechanism Level: How can the representation and algorithm berealized physically?

Page 7: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 8: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 9: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 10: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 11: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 12: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

The Computational Approach toVision: A Sampler of Issues

• Low-level vision:Edge segmentation: which areas of the scene projected onto theretina correspond to the potentially interesting objects?Depth perception: how far away are the objects?

• Interpreting Edges as “Line Drawings”• Interpreting Shapes as (possibly familiar) objects

Page 13: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM

What Do We Mean by “Low-Level” Vision, Anyway?

• Based on a (not entirely accurate) portrait ofvision as proceeding “bottom-up” from raw (pixel-like) data to object recognition

• Tasks like: object boundary detection, surfaceorientation, depth perception, motion detection

• Fast, “rough”, domain-independent processing• Parallel processing• Unconscious levels of (human) vision processing

Page 14: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 15: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 16: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 17: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 18: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 19: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 20: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 21: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 22: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 23: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 24: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 25: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 26: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 27: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 28: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM
Page 29: AI 2007 L20 (Vision2)€¦ · Title: AI 2007 L20 (Vision2).ppt Author: Ann Eisenberg Created Date: 11/15/2007 9:31:26 AM