Scene perception after those first
few hundred milliseconds
Jeremy Wolfe
Brigham and Women's Hospital and Harvard Medical
School
What is the evidence for this claim?
The repeated search task
A
a
D
G
A
B
F
Y
f
D
G
A
B
F
Y
h
D
G
A
B
F
Y
B
y
D
G
F
Y
• The capital letters are the search array.
• The lower case letter at the center tell you
what to look for on this trial.
• In repeated search, the search array does not
change from trial to trial.
Repeated Search
K
e
T
E
V
R
Repeated Search
K
c
T
E
V
R
Repeated Search
K
t
E
R
T
V
Repeated Search
K
r
T
E
V
R
Repeated Search
K
E
R
a
T
V
Repeated Search
1
T
b
Q
T
A
Standard Search
1
T
b
Q
S
G
B
J
F
A
B
G
f
Z
B
1
350
M
V
L
N
t
T
S
E
350
2
p
A
G
B
3
p
RV
T
Q
H
f
Memory Search
A
2
t
Q
G
B
350
T
A
p
Q
G
B
3
T
A
f
Q
G
B
2
t
Repeated search never becomes efficient.
Slope (msec/item)
80
60
40
20
0
-20
0
100
200
300
Trial number
REPEATED SEARCH
Slope of the RT x set size function is the measure of search efficiency
Repeated search is up at the “ceiling”
80
60
S
L
O
P
E
40
20
0
-20
0
100
200
300
Trial number
REPEATED SEARCH
CEILING
Memory search is actually more efficient
80
60
S
L
O
P
E
40
20
0
-20
0
100
200
300
Trial number
REPEATED SEARCH
CEILING
MEMORY SEARCH
Actually, only 2AFC memory search is efficient
Localization responses
100
80
Slope 60
msec/item
40
20
100
200
300
Trials
400
500
Memory
Repeated
Unrepeated
Melina Kunar’s mouseclick experiments. (Kunar, Flusberg, & Wolfe)
But this is a meeting about scenes
So Aude and I did a version with scenes like this
Same basic result
1100
Reaction Time (ms)
Unrepeated
y = 34.6x + 704
y = 34x + 687
900
y = 33.1x + 489
y = 30.7x + 507
700
Repeated
500
2
4
6
Set Size
8
10
Epoch 1 - Repeated
Epoch 1 - Unrepeated
Epoch 2 - Repeated
Epoch 2 - Unrepeated
Figure by MIT OCW.
Note: Mean RT is faster, but search efficiency is unchanged.
We also did “panoramic” search
Panorama Experiments
Panorama Experiments
Panorama Experiments
Panorama Experiments
Panorama Experiments
Panorama Experiments
Panorama Experiments
You can search for a visible target
You still get a slope
If the same target is hidden
The slope can vanish
Subjects make a pragmatic choice
Search if the target is visible
The “IsWas” Paradigm
Minimal Change Blindness
Behold, I bring you a mystery.
Does the cued dot change color?
No
No
Yes
No
So…to summarize
Early Vision
Visual experience
A selective pathway
Early Vision
and
subsequent
awareness
with an
attentional
bottleneck
feeding
object
recognition
Binding
Recognition
The selective path
Visual experience
Access to the bottleneck is controlled by
guiding representation.
Early Vision
Color
Orientation
Size
Motion
Depth
etc.
The Guiding
Representation
Binding
Recognition
The selective path
Visual experience
A non-selective pathway can
fill in the rest of the experience
Early Vision
Color
Orientation
Size
Motion
Depth
etc.
The Guiding
Representation
Binding
Recognition
The selective path