surf seminar presentation
TRANSCRIPT
INFLUENCE OF PARIETAL CORTEX ON SALIENCY DRIVEN VISION IN NON HUMAN PRIMATESEshan GovilSURF Summer 2016Moore Labs at Stanford UniversityMentors: Xiaomo Chen, Mark Zirnsak, and Tirin Moore
Overview• 1 – Experimental Paradigm
• What was the experiment design• 2 – Basic Behavior
• Shift to the right• Main Sequence
• 3 – Optimization• IKN and GBVS model
• 4 – Saliency Driven fixations• Proportion as a function of Saliency Difference• Average saliency through the fixation sequence
• 5 – Future Direction
Key Terms• Saliency / Physical Salience
• The unique features of an image we subconsciously focus on when gauging dense sensory input to determine what we are looking at
• Saccade• Rapid, almost involuntary eye movements from one point to
another• Parietal Cortex
• Sensory input processing and integration• Frontal Eye Field
• Responsible for saccadic and voluntary eye movements, visual field perception, and awareness
• Ipsilateral/Contralateral• To the same side / to the other side of the body
Experimental Background and Design
EXPERIMENTAL GOALObserve and quantify the effects of reversible inactivation
of the parietal cortex on the processing of visual salience in the FEF.
IN GENERAL TERMSIf we switch off a region of our Parietal Cortex, how does that affect our vision? How does the Frontal Eye Field get
involved during inactivation?
Experimental Background and Design
http://www.vanderbilt.edu/exploration/text/index.php?action=view_section&id=1192&story_id=289&images=
Fixation
Time
5 s
Fixation
Reward
Free viewing task
Control Block (30 min)
Inactivation Block (30 min)
Fixation
Time
5 s
Fixation
Reward
Left visual hemifieldRight visual hemifield
Behavioral biases induced by parietal inactivation
Bias for eye movementsaway from the inactivatedhemifield
1. Free viewing task2. Two-target free-choice task
Two behavioral tests:
e.g.,Lynch & McLaren 1989Karnath et al. 1998
Intraparietal Sulcus (IPS)Inactivation
ControlNumber of Fixations
Left: 5 Right: 3Number of Fixations
Left: 1 Right: 10
Inactivation
Free viewing task: Example image
Scanpath of the eyes Scanpath of the eyes
Basic Behavior• Observed: IPS Inactivation in the right hemisphere of the
brain -> eye movements biased away from contralateral field• Ipsilateral shift in horizontal fixations• Both in free viewing and two target free choice task
2D Density Heatmap - Control
X axis (dva)
Y a
xis
(dva
)
Den
sity
X axis (dva)
Den
sity
Y a
xis
(dva
)
2D Density heatmap - Experimental
X axis (dva)
Den
sity
(Inc
or D
ec)
Y a
xis
(dva
)
2D Density Heatmap – Exp v Control Difference
Magnitude (degrees)
Dur
atio
n (m
s)
Magnitude (degrees)
Pea
k Ve
loci
ty (d
egre
es /
sec)
Optimization – GBVS toolboxIn order to compare fixation data to calculated visual saliency that, through optimization, will be accurate to how our vision typically identifies saliency
• Parameters optimized: • IKN - center and surround levels• GBVS – feature map resolution
• More parameters to choose from but did not optimize
Optimization• Default
• p.levels = [2 3 4];• p.ittiCenterLevels = [2 3 4];• p.ittiDeltaLevels = [3 4];
• Optimized• p.levels = [2 3 5 6 7 8 9];• p.ittiCenterLevels = [2 5 6];• p.ittiDeltaLevels = [1 3];
• Optimized across random set of images across all experiments
Pre-optimization
Post-optimization
Saliency Driven Vision• Previously observed: following inactivation, responses to
visual stimuli are reduced compared to control• Currently observed: responses to visual stimuli are about
the same in some cases
Fixations in order
Average Saliency (Left) per Fixation in order of Fixations across all ExperimentsA
vera
ge S
alie
ncy
Average Saliency (Right) per Fixation in order of Fixations across all Experiments
Fixations in order
Ave
rage
Sal
ienc
y
Saliency Difference (Right field – Left field)
Pro
porti
on o
f Fix
atio
ns in
Rig
ht fi
eld
GBVS - % Fixations to the Right as a function of Saliency
Future Direction• Run my code on future data sets
• Validate my results and look for greater significance• Improve and formalize the figures created and findings
from this data set• Separate images used in optimization from the data set• Run all tests for both GBVS and IKN
Acknowledgements• My mentor Dr. Xiaomo Chen• Dr. Tirin Moore, Dr. Marc Zirnsak, and all other members
of the Moore lab• Stanford University• the Caltech SFP program