Object Recognition

Question 1

How do we generate mental representations of 3D objects?

Answer 1

There are two theories
- Marr’s model
- Biedermans theory

Question 2

Marr’s Model

Answer 2

Three stages wher visual information is represented in different ways
1. Primal sketch - 2D description, viewpoin dependent and basic features
2. 2,5D sketch - Depth cues added, still viewpoint dependent
3. 3D representation
Not sure how stage 2 to 3 happens

Question 3

Biederman’s recognition-by-components Theory

Answer 3

• Made out of besic shapes
  Geons
• Recognition involves recognizing geons and their configuration
  Breaking them down
  Mug, telephone, flashlight etc

Question 4

RBC - Identifying geons

Answer 4

• Edge detection
  Different combinations allow us to identify geons, even when things are partily covered (if we see the impoartna shapes)
• View-point independent

Question 5

RBC - Shortcomings

Answer 5

• Its incomplete, tied to a bottom-up processing
  Object recognition is influenced by context
• We see objects as a “whole”, not components
• Several processes cooperate to create a 3D representation

Question 6

What are some top-down influences on object recognition?

Answer 6

• Conceptual priming effects with ambigious figures
  What are we more used to seeing?
  Differences between similar objects
• What you see first is often influenced from previous information
• Context can help reognizing and detecting objects in the real world, where we cant rely purely on local features (blurry picture)
  Location, size, pose and semantics

Question 7

Apperceptive agnosia

Answer 7

Disurption in perceiving patterns
- Cant discriminate objects from each other

Question 8

The role of context in object recognition - Intro

Answer 8

• Objects never occur in isolations
• Surprising ability to recognize thousands of objects
• Visual search
  Problem solving with help of context

Question 9

Contextual influences on object recognition - Article

Answer 9

• Efficiency of search and recognition
  Familiar settings - easier to find some objects
• Real-world configurations governed by “rules”
  Finding common patterns - 100s of pictures
• Important when local features are insufficient
  Blurry pictures
• The object relationship to other objects and the context its in

Question 10

Multiple levels of context - Article

Answer 10

Semantic - what makes sense
Spatial configuration - expectations on where objects should be at
Pose

First two are good for quick recognitions

Question 11

Implicit learning of contextual cues - Article

Answer 11

Contextual cueing
- Learning certain contingencies by observing
- Certain relationships between objects
Affects on attention
- Biases
Prior knowledge / memory
Context help with simplifying object discriminations

Question 12

Cognitive Research and Context - Article

Answer 12

• Statistical summaries
  Effective source of contextual inference
  Elements that do or don’t correspond with each other
• Scene representations
  Global scene
  Holistic view, adds more understanding
• Fixing eye movements
  Direct attention where it makes sense