Pcb - Vision Lecture 3

Front 1

What are examples of inference early in the visual system?

Back 1

V1 curve tracing & "object effect"

V2 border ownership explaining illusory contours

- Gestalt principles

- Figure-ground

- Shadows for depth perception

Front 2

What is neural mechanism of size constancy?

Back 2

1. V1 shows object-centered shifts in receptive field position (Ni et al., 2014)

2. Receptive fields are malleable

- If the 'Retinal ring image' is within a lot of receptive fields, considered as far

- If it is in between a bunch of receptive fields and the RFs only see the edges of the ring, then it is considered near

Front 3

How was binocular rivalry shown and who found it?

Back 3

Wheatstone, 1838

1. There is a large overlap between 2 eyes

2. Precept of 2 different directions since they are presented 1 to each eye (occluder between eyes)

3. One percept dominates the other (switches between left & right percepts)

4. More complex than simply eye competition since also works for monocular rivalry

Front 4

What is an explanation for binocular rivalry that has been disproven by electrophysiological recording?

Back 4

Mutual inhibition between monocularly driven orientation-selective neurons

Front 5

What did Logothetis find in his study on binocular rivalry?

Back 5

1. Weak activity changes in V4 & MT

2. Strong changes in IT

3. Activity depends on type of stimuli

Conclusion: Competition occurs at higher levels of the visual system

- Not just competition between different ocular columns!

Front 6

What distribution do lengths of percepts in time have?

How might this be explained?

Back 6

Characteristic gamma distribution

Could be explained via neuronal organisation (competition between neurons)

Front 7

What is the evidence for control of bistable switching? Where may this occur neuronally?

Back 7

1. Breaking the gamma distribution by introducing a transient in the image & "resetting" the whole thing

2. Control most likely in higher processing

- Stronger activity changes further along the visual pathway (strong changes in IT)

Front 8

What are observations of binocular rivalry?

Back 8

1. Same stimulus can be percieved in 2 different ways

2. Spontaneous 'perceptual reversals/switching'

3. Often voluntary switching is possible

Front 9

What is the physical analogy of multi-stable perception?

Back 9

Stable states as 'valleys' on a curve (e.g. W shaped)

- Percept becomes clear after it converges to a stable equilibrium state (local minima)

- Minima are probable, whereas higher states aren't

Unclear perceptual states are specified by the stimulus

Front 10

What happens in the physical analogy from fatigue?

Back 10

Neural fatigue causes percept to converge to the second 'minima / stable equilibrium state' since the first solution becomes less stable

Front 11

What can push specific percepts?

Back 11

Spontaneous competition between networks

- Slower dynamic of competition wherein the percept can switch

Front 12

What is an example of an image that we can see despite lower level visual processing not being able to extract any meaningful information? What does this suggest?

Back 12

The ambiguous image of old woman sitting at a bench

- Consists of amorphous black & white shapes

Suggests that there must be "feedback" based on "top-down" knowledge

Front 13

What is the model of top-down knowledge influencing perception?

Back 13

1. Stimulus enters as a 2D shape

2. 2D shape then enters memory and becomes perceived as a 3D shape

3. Bidirectional communication (feedback) between memory & 3D shape and 3D shape & 2D shape

Front 14

How might top-down control be implemented? And who thought of this?

Back 14

Triadic Architecture - Ron Rensik, 2000

System 1 = Low-level vision

- Pixels, edges & proto-objects

System 2 = Object (attentional)

- Coherent objects

System 3 = Setting (non-attentional)

- Scene schema (layout & gist)

1. Early processes segment proto-objects from the background rapidly and in parallel across the visual field

2. Focused attention can then access these structures forming an individuated object with both temporal & spatial coherence

3. Information about the context gets acquired outside of attention guides to various locations & sets scene priorities or salience

Low level visual system

Pixles -> Edges -> Proto-objects

-> Layout & Gist (Scene Schema) -> Setting (unattentional)

or

-> Coherent objects -> perception of the object via focused attention

Front 15

What are examples of attention affecting perception?

Back 15

1. Ambiguous figures

- Once seen, can never be unseen (powerful top-down signal)

2. Inferential vision

- Can cause the "wrong" inference

3. Biological motion

- Point-lights oscillate sinusoidally locally but still extract figure motion

Front 16

What is the spatial attention model & who created it?

Back 16

Cavanagh et al., 2010

1. Feedforward volley of information

2. Then reaches 'pointer-map' and spatial pointer which picks out relevant points

3. Pointer may be influenced by memory

4. Representation agnostic of visual features?

5. Functionally appears as though the receptive field has shrunk (due to interaction of neuron & pointer signal)

6. Pointer map directs attention to a specific point -> Limits responses to a point & suppresses other areas in the (large) receptive field

Front 17

Which hypothesis is the pointer map related to? What does the hypothesis solve?

Back 17

Re-entry hypothesis (Hamker)

1. Attentional pointer re-enters the large receptive field & constrains the response of the attended location

2. Information re-enters the brain

(feedforward from visual system to Frontal Eye Field [FEF], then feedback from FEF to visual system)

Solves the problem of large receptive fields

Front 18

How may the 'pointer map' work?

Back 18

1. Population coding

- Broadly tuned neurons to preferred features

- Maximal point of overall population response reflects the actual feature

- Each neuron codes for a broad region of the visual field

2. Winner takes all architecture

- Local excitation allows a "winning" location to sustain itself & suppress other competitors

- Fits with the "bottleneck" model of processing

- Fits with mechanism of eye movement (winner being the foveally fixated item since we only need to move eyes to one "winner")

(FEF, LIP & SC are all involved in eye movements + attention)

Front 19

What is the normalization model?

Back 19

1. Brings firing rates of neurons back to the level where variability can be seen

2. One way of filtering that allows one thing to win over another

Front 20

How are dynamics of perception controlled by attentional pointers?

Back 20

1. Recurrent signals

2. Shared neural circuitry with eye movements

Front 21

What are the Itti & Koch style models of visual saliency representation?

Back 21

1. Image is parallel processed

2. Feature analysis takes place

-> Center-surround inhibition

3. Feature map (feature-dependent)

Controlled by top-down info

-> Sum across all features

4. Saliency map (feature-independent)

-> Winner-takes-all

5. Priority map

6. Action (saccade)

Front 22

What are biological implementations of feature map vs feature agnostic?

Back 22

1. V1 can contribute to salience (e.g. the pop-out model)

2. However, V1 neurons are feature-sensitive (feature maps)

3. Superficial Superior colliculus layers demonstrate saliency representation in a feature-agnostic manner

Front 23

What are types of eye movements?

Back 23

1. Saccades

- Very fast eye movements

- Allow fast alignment of the fovea with objects of interest

- As fast/brief as possible to avoid disruption of the visual system

2. Smooth pursuit

- Allows stabilization of image of a moving object on the retina

- See the item without motion blur

3. Fixational eye movement

- After-image experiment: movement/jitter of the after image

- After-image is stable/fixed on the retina but jitter/movement of the eye causes movement of the after image

- Perceived movement is the movement of the eye whilst image is glued to the retina

4. Image stabilization reflexes

- e.g. VOR

5. Eye movements in depth

- e.g. Convergence

Front 24

What are the 2 components of fixation?

Back 24

1. Slow control

- Slow change in eye movements

2. Microsaccades

- Rapid, step-like changes in eye position

Front 25

What are challenges posed by eye movements?

Back 25

1. Brain receives jittered image despite perception being stable

2. Discrete snapshots with little spatial information between them

3. Each saccade causes massive changes to the retinal image

- Sudden changes to the retinal image

- Extremely fast retinal image motion

4. Smooth pursuit also leads to problems

- Is image motion due to self-motion or object-motion?

Front 26

What happens during looking around scenes?

Back 26

1. Foveating targets natural scenes

2. There is constant competition between different scene locations

3. Visual masking is a strong constraint

Front 27

How is visual masking tested? What was found?

Back 27

Method

1. Put a low spatial frequency Gabor in either uniform gray or in the natural image

2. Change contrast of stimulus & observe how often it is found & fixated

Results

1. Need higher contrast to see the natural image compared to uniform grey

2. At perceptual threshold (equally likely to see stimulus with or without natural background), saccade latency is 100ms for natural background

Conclusion

Statistics of natural scenes facilitate saccade generation

Front 28

How was the Superior Colliculus response to contrast studied?

Back 28

Method

Low frequency spatial gratings of different magnitudes presented to SC receptive field

Results

- If you change the spatial frequency, get a massive change in the properties of response

- Similar speed of visual response peak as V1 in SC for low spatial frequencies

- Much faster peak of response to higher frequencies

- Also facilitates faster saccades