Categorization: Scenes & Objects (P) Lavanya Sharan March 16th, 2011.

Slides:



Advertisements
Similar presentations
Posner and Keele; Rosch et al.. Posner and Keele: Two Main Points Greatest generalization is to prototype. –Given noisy examples of prototype, prototype.
Advertisements

Chapter 44 Visual Perception of Objects Copyright © 2014 Elsevier Inc. All rights reserved.
Fast Readout of Object Identity from Macaque Inferior Tempora Cortex Chou P. Hung, Gabriel Kreiman, Tomaso Poggio, James J.DiCarlo McGovern Institute for.
Are faces special?. Brain damage can produce problems in face recognition - even own reflection (Bodamer, 1947) Prosopagnosia usually results from localized.
FMRI Activation of the Fusiform Gyrus and Amygdala to Cartoon Characters but not to Faces in a boy with Autism  Looking specifically at the fusiform gyrus.
The neural basis of face recognition? Tim Andrews.
FMRI Reveals a Dissociation Between Object Grasping and Object Recognition Culham et al. (submitted)
Comparing Thompson’s Thatcher effect with faces and non-face objects Elyssa Twedt 1, David Sheinberg 2 & Isabel Gauthier 1 Vanderbilt University 1, Brown.
FMRI - What Is It? Then: Example of fMRI in Face Processing Psychology 355: Cognitive Psychology Instructor: John Miyamoto 04/06 /2015: Lecture 02-1 This.
I. Face Perception II. Visual Imagery. Is Face Recognition Special? Arguments have been made for both functional and neuroanatomical specialization for.
Visual Cognition II Object Perception. Theories of Object Recognition Template matching models Feature matching Models Recognition-by-components Configural.
Lecture 30: Light, color, and reflectance CS4670: Computer Vision Noah Snavely.
Psychophysics of the structure of object memory Cristina Savin.
Object Recognition: History and Overview Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and Jean Ponce.
Attention II Selective Attention & Visual Search.
Visual Cognition II Object Perception. Theories of Object Recognition Template matching models Feature matching Models Recognition-by-components Configural.
Object Recognition: Conceptual Issues Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and K. Grauman.
Visual Expertise Is a General Skill Maki Sugimoto University of California, San Diego November 20, 2000.
Object Recognition: Conceptual Issues Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and K. Grauman.
Materials II Lavanya Sharan March 2nd, Computational thinking about materials Physics-basedPseudo physics-based.
What should be done at the Low Level?
Extreme, Non-parametric Object Recognition 80 million tiny images (Torralba et al)
Michael Arbib & Laurent Itti: CS664 – USC, spring Lecture 6: Object Recognition 1 CS664, USC, Spring 2002 Lecture 6. Object Recognition Reading Assignments:
Laurent Itti: CS564 – Brain theory and artificial intelligence. Scene Perception 1 Brain theory and artificial intelligence Lecture 23. Scene Perception.
Cognitive Processes PSY 334 Chapter 2 – Perception.
An aside: peripheral drift illusion illusion of motion is strongest when reading text (such as this) while viewing the image in your periphery. Blinking.
Laurent Itti: CS599 – Computational Architectures in Biological Vision, USC Lecture 13: Scene Perception 1 Computational Architectures in Biological.
Biases: An Example Non-accidental properties: Properties that appear in an image that are very unlikely to have been produced by chance, and therefore.
WORD SEMANTICS 2 DAY 27 – OCT 30, 2013 Brain & Language LING NSCI Harry Howard Tulane University.
Pattern recognition = perception Template theory  has problems Prototype theory  better Distinctive features theory  better.
1 Perception and VR MONT 104S, Fall 2008 Session 13 Visual Attention.
Perceptual and Sensory Augmented Computing Visual Object Recognition Tutorial Visual Object Recognition Bastian Leibe & Computer Vision Laboratory ETH.
PERCEPTION AND PATTERN RECOGNITION Making sense of sensation –Local vs. Global scope –Data-driven (sensory, bottom-up) vs. Concept-driven (knowledge, “top-down”)
Lecture 2b Readings: Kandell Schwartz et al Ch 27 Wolfe et al Chs 3 and 4.
Chapter 4: Object Recognition What do various disorders of shape recognition tell us about object recognition? What do various disorders of shape recognition.
Visual Recognition: The Big Picture Jitendra Malik University of California at Berkeley.
Functional Magnetic Resonance Imaging ; What is it and what can it do? Heather Rupp Common Themes in Reproductive Diversity Kinsey Institute Indiana University.
Representations for object class recognition David Lowe Department of Computer Science University of British Columbia Vancouver, Canada Sept. 21, 2006.
Chapter 5: Perceiving Objects and Scenes. The Puzzle of Object and Scene Perception The stimulus on the receptors is ambiguous. –Inverse projection problem:
Elements of Art. Line Line is the basic element of art. Line can be 2-D or 3-D. Line has many variations.
Vision. 2 Brodmann Original Calcarine 17 Collateral Sulcus Fusiform Gyrus 18.
Computer Vision Group University of California Berkeley On Visual Recognition Jitendra Malik UC Berkeley.
Human vision Jitendra Malik U.C. Berkeley. Visual Areas.
Expertise, Millisecond by Millisecond Tim Curran, University of Colorado Boulder 1.
EMPATH: A Neural Network that Categorizes Facial Expressions Matthew N. Dailey and Garrison W. Cottrell University of California, San Diego Curtis Padgett.
3:01 PM Three points for today Sensory memory (SM) contains highly transient information about the dynamic sensory array. Stabilizing the contents of SM.
Visual Agnosias Specification: Theories of perceptual organisation
High level vision.
Face Recognition in the Brain Arash Afraz.
Fundamentals of Sensation and Perception
Object and face recognition
Image features and properties. Image content representation The simplest representation of an image pattern is to list image pixels, one after the other.
Signal Detection Theory.
Chapter 15. Cognitive Adequacy in Brain- Like Intelligence in Brain-Like Intelligence, Sendhoff et al. Course: Robots Learning from Humans Cinarel, Ceyda.
Max-Margin Training of Upstream Scene Understanding Models Jun Zhu Carnegie Mellon University Joint work with Li-Jia Li *, Li Fei-Fei *, and Eric P. Xing.
Event-related gamma-band activity in visual object representation: the coding of object features Jasna Martinovic Thomas Gruber * Matthias M. Müller *
Chapter 2 Cognitive Neuroscience. Some Questions to Consider What is cognitive neuroscience, and why is it necessary? How is information transmitted from.
Li Fei-Fei, Stanford Rob Fergus, NYU Antonio Torralba, MIT Recognizing and Learning Object Categories: Year 2009 ICCV 2009 Kyoto, Short Course, September.
Chapter 2 Cognitive Neuroscience. Some Questions to Consider What is cognitive neuroscience, and why is it necessary? How is information transmitted from.
Perception & Pattern Recognition 1 Perception Pattern Recognition Theories of Pattern Recognition Bottom-up vs. Top-Down Processing & Pattern Recognition.
9.012 Presentation by Alex Rakhlin March 16, 2001
Representational Similarity Analysis
Representational Similarity Analysis
© 2016 by W. W. Norton & Company Recognizing Objects Chapter 4 Lecture Outline.
fMRI: What Does It Measure?
Lecture 25: Introduction to Recognition
Neuropsychology of Vision Anthony Cate April 19, 2001
ICCV 2009 Kyoto, Short Course, September 24
Brief Review of Recognition + Context
SIFT keypoint detection
Presentation transcript:

Categorization: Scenes & Objects (P) Lavanya Sharan March 16th, 2011

Last time What is category? Functional vs. communicational Basic-level categories (Rosch) Entry-level categories and prototypicality

Visual perception of categories Objects (e.g., animal vs. non-animal, cars vs. houses, German shepherds vs. other dogs etc.) Kirchner & Thorpe, 2005 Grill-Spector & Kanwisher, 2005

Visual perception of categories Scenes (e.g., desert vs. canyon, low openness vs. high openness etc.) Oliva & Schyns, 2000 Greene & Oliva, 2009

Why does the choice of category matter? Object Non-object Detection task: Should be easiest and fastest.

Why does the choice of category matter? House Dog Categorization task: Should be harder and slower? Image sources: dogbreedinfo.com, cambridge2000.com

Why does the choice of category matter? Danish Farm Dog Old Danish Chicken Dog Categorization task: This one should be hardest and slowest? Image source: dogbreedinfo.com,

Classic results in object recognition Mostly from psychophysical experiments. We can tell object from non-object in a brief glance. We can tell scene categories in a brief glance. Beyond categories, details about scenes and objects can be inferred in a brief glance. Short reaction times and ERP data suggests we can do these tasks quickly. Objects: Thorpe et al. 1996, Grill-Spector & Kanwisher 2005, Kanwisher et al. 1997, Potter 1975, Rosch 1978, Nakayama et al. 1995, Biederman 1987, Intraub 1981, Peterson & Gibson 1993,… Scenes: Biederman 1972, Potter 1975, 1976, Intraub 1981, Oliva and Schyns 2000, Oliva & Torralba 2001, Rousselet et al. 2005, Evans & Treisman 2005, Fei-fei et al. 2007, Greene & Oliva 2009,...

Today: Object recognition & fMRI Slide source: Jody Culham

Today: Object recognition & fMRI Slide source: Jody Culham

Why care about what fMRI has to say? Slide source: Jody Culham

Category-specific regions Image source: Grill-Spector 2008 Blue = Object > Scrambled objects Lateral Occipital Complex (LOC) Red = Faces > Non-face objects Fusiform Face Area (FFA) + few others Green = Places > Objects Parahippocampal place area (PPA) + few others Magenta = Faces + Objects Dark Green = Places + Objects Also regions for body parts, letter vs. textures, tools vs. animals etc.

LOC: Basics Two regions: LO + pFus/OTS Shape, surfaces, contours. Not low-level features (e.g., colors, textures). Global shape, not local contours. Image source: Grill-Spector 2008

LOC: Basics Image source: Grill-Spector 2008

Objection recognition implies invariances Size Position Rotation Illumination... Can fMRI activations tell us how these invariances are achieved? Unlikely. But, let’s study how invariant LOC responses are. Is it truly correlated with successful object recognition?

Measuring position invariance in LO Image source: Grill-Spector 2008

Position effects larger than category effects in LO! Image source: Grill-Spector 2008 For pFus/OTS greater position invariance than LO (Schwarzlose et al. 2008)

Rotation sensitivity in LOC Image source: Grill-Spector 2008 Andreson et al used fMRI adaptation. If a region is sensitive to rotation, then repeating the same (or similar view) will cause a change in response. Mixed story. Rotation sensitivity depends on categories and brain regions.

Implications for object recognition theories/models View dependence. What is the representation for object recognition? View-sensitive neurons (low-level representations) and population coding? Or view-invariant neurons outside regions measured? Do we really need view-invariant representations for object recognition?

Implications for object recognition theories/models Domain specialization. Given that some object categories seem special, what does that imply for object recognition theories? General-purpose computations vs. specialized features and computation? Alternative interpretation of fMRI data: These regions are really about expertise with a visual category rather than the category itself (e.g., faces).

Implications for object recognition theories/models Distributed processing. It is possible to predict (using machine learning techniques on fMRI data) which object category a person is looking at (even when FFA etc. are not considered). Advantage of distributed code, recognize more objects?

Implications for object recognition theories/models Effect of experience. Size of category-specific regions changes as children mature to become adults. These changes are not simply geometric scaling. Learned representations vs. innate modules?