Kimberly Kirkpatrick , Tannis Bilton , Bruce Hansen

advertisement
Kimberly Kirkpatrick1, Tannis Bilton1,
Bruce Hansen2, and Les Loschky1
1Kansas State University
2Colgate University

The gist of a scene refers to the general meaning of
a scene, which can be operationally defined as the
basic level category that the scene belongs to
 Oliva, 2005; Tversky & Hemenway, 1983

Scene gist is:
 Very rapid (<= 24 ms SOA)
▪ Bacon-Mace et al., 2005; Loschky et al., 2007; Loschky, Hansen et al., 2010
 Superior when discriminating between
superordinate categories (natural vs. man-made)
than when discriminating within a superordinate
category (natural vs. natural)
▪ Joubert et al., 2007; Loschky & Larson, 2010
 Superior with terrestrial views compared to aerial
views
▪ Loschky et al., 2011

We know that pigeons can discriminate categories of
items
 Most research has focused on basic object categories (people
vs. cars)
 Not much on scenes

What we don’t know
 Can pigeons categorize scene gist with short stimulus exposure
times?
▪ How much time do they need?
 Do pigeons differ in their scene gist depending on whether the
discrimination is between or within superordinate categories?
 Do pigeons differ in their scene gist categorization depending
on viewpoint (aerial vs. terrestrial)?

We know that pigeons can discriminate categories of
items
 Most research has focused on basic object categories (people
vs. cars)
 Not much on scenes

What we don’t know
Experiment 1
 Can pigeons categorize scene gist with short stimulus exposure
times?
▪ How much time do they need?
 Do pigeons differ in their scene gist depending on whether the
discrimination is between or within superordinate categories?
 Do pigeons differ in their scene gist categorization depending
on viewpoint (aerial vs. terrestrial)?

We know that pigeons can discriminate categories of
items
 Most research has focused on basic object categories (people
vs. cars)
 Not much on scenes

What we don’t know
Experiment 1
 Can pigeons categorize scene gist with short stimulus exposure
times?
▪ How much time do they need?
 Do pigeons differ in their scene gist depending on whether the
discrimination is between or within superordinate categories?
 Do pigeons differ in their scene gist categorization depending
on viewpoint (aerial vs. terrestrial)? Experiment 2

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to S-

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to SFixation – peck 1 x

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to SGray square 300-900 ms

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to STarget Image – 5 s

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to SGo/no go response – 5 s

Two groups of birds
 Natural-natural (beach vs.
mountain); n=4
 Natural-manmade (beach vs.
street); n=3

Train with go/no-go
procedure
 20 S+ vs. 20 S- images,
delivered in 5 blocks, 200
trials/session
 Food reinforcement if
respond to S+
 5-s blackout if respond to SITI – 10 s




Test with 40 novel S+ and 40 Novel S- images
Test trials randomly intermixed with normal
training trials
5-s image presentation for all test trials
Neutral contingency (no food, no black-out)
for all test trials


Both groups displayed
high-level performance
in training (> .85)
Only a small
generalization
decrement was observed
during testing
No difference between
natural-natural (beach
vs. mountain) and
natural-manmade
(beach vs. street)
1
0.8
DISCRIMINATION RATIO .

0.6
0.4
0.2
0
TRAINING
TEST
GROUP N-N
TRAINING
TEST
GROUP N-M

Rapid categorization training
 Same as original training, but reduced image
presentation time: 4 s, 2 s, 1 s, .75 s, .50 s, .35 s, .20
s, .10 s
 Image duration reduced over successive phases
 Train in each phase to at least 80% correct for two
consecutive days
Group N-N
Stimulus
Duration
Y3
Y4
4
2 (.91) 9 (.80)
2
2 (.92) 3 (.83)
1
8 (.65) 2 (.86)
0.75
2 (.85) 5 (.80)
0.5
18 (.71) 3 (.81)
0.35
... (.65) 24 (.76)
0.2
X
… (.71)
0.1
X
X
Y9
2 (.88)
2 (.95)
2 (.85)
2 (.87)
4 (.79)
2 (.83)
… (.70)
X
Group N-M
Y10
2 (.94)
2 (.95)
2 (.86)
2 (.85)
20 (.80)
5 (.80)
… (.81)
X
Y2
2 (.90)
2 (.85)
2 (.87)
2 (.88)
3 (.84)
2 (.87)
18 (.76)
… (.63)
Y6
2 (.93)
2 (.93)
6 (.81)
2 (.89)
10 (.76)
… (.67)
X
X
Y8
2 (.86)
2 (.93)
2 (.99)
2 (.89)
2 (.88)
2 (.94)
3 (.87)
… (.62)
Sessions
to Criterion
Group N-N
Stimulus
Duration
Y3
Y4
4
2 (.91) 9 (.80)
2
2 (.92) 3 (.83)
1
8 (.65) 2 (.86)
0.75
2 (.85) 5 (.80)
0.5
18 (.71) 3 (.81)
0.35
... (.65) 24 (.76)
0.2
X
… (.71)
0.1
X
X
Y9
2 (.88)
2 (.95)
2 (.85)
2 (.87)
4 (.79)
2 (.83)
… (.70)
X
Group N-M
Y10
2 (.94)
2 (.95)
2 (.86)
2 (.85)
20 (.80)
5 (.80)
… (.81)
X
Criterion = 2 consecutive days at > .80
… = Trained, but did not make criterion
X = Did not receive training
Y2
2 (.90)
2 (.85)
2 (.87)
2 (.88)
3 (.84)
2 (.87)
18 (.76)
… (.63)
Y6
2 (.93)
2 (.93)
6 (.81)
2 (.89)
10 (.76)
… (.67)
X
X
Y8
2 (.86)
2 (.93)
2 (.99)
2 (.89)
2 (.88)
2 (.94)
3 (.87)
… (.62)
Sessions
to Criterion
Mean DR
over all Sessions
Group N-N
Stimulus
Duration
Y3
Y4
4
2 (.91) 9 (.80)
2
2 (.92) 3 (.83)
1
8 (.65) 2 (.86)
0.75
2 (.85) 5 (.80)
0.5
18 (.71) 3 (.81)
0.35
... (.65) 24 (.76)
0.2
X
… (.71)
0.1
X
X
Y9
2 (.88)
2 (.95)
2 (.85)
2 (.87)
4 (.79)
2 (.83)
… (.70)
X
Group N-M
Y10
2 (.94)
2 (.95)
2 (.86)
2 (.85)
20 (.80)
5 (.80)
… (.81)
X
Criterion = 2 consecutive days at > .80
… = Trained, but did not make criterion
X = Did not receive training
Y2
2 (.90)
2 (.85)
2 (.87)
2 (.88)
3 (.84)
2 (.87)
18 (.76)
… (.63)
Y6
2 (.93)
2 (.93)
6 (.81)
2 (.89)
10 (.76)
… (.67)
X
X
Y8
2 (.86)
2 (.93)
2 (.99)
2 (.89)
2 (.88)
2 (.94)
3 (.87)
… (.62)
Sessions
to Criterion
Mean DR
over all Sessions
Group N-N
Stimulus
Duration
Y3
Y4
4
2 (.91) 9 (.80)
2
2 (.92) 3 (.83)
1
8 (.65) 2 (.86)
0.75
2 (.85) 5 (.80)
0.5
18 (.71) 3 (.81)
0.35
... (.65) 24 (.76)
0.2
X
… (.71)
0.1
X
X
Y9
2 (.88)
2 (.95)
2 (.85)
2 (.87)
4 (.79)
2 (.83)
… (.70)
X
Group N-M
Y10
2 (.94)
2 (.95)
2 (.86)
2 (.85)
20 (.80)
5 (.80)
… (.81)
X
Criterion = 2 consecutive days at > .80
… = Trained, but did not make criterion
X = Did not receive training
Y2
2 (.90)
2 (.85)
2 (.87)
2 (.88)
3 (.84)
2 (.87)
18 (.76)
… (.63)
Y6
2 (.93)
2 (.93)
6 (.81)
2 (.89)
10 (.76)
… (.67)
X
X
Y8
2 (.86)
2 (.93)
2 (.99)
2 (.89)
2 (.88)
2 (.94)
3 (.87)
… (.62)
Sessions
to Criterion
Mean DR
over all Sessions
Group N-N
Stimulus
Duration
Y3
Y4
4
2 (.91) 9 (.80)
2
2 (.92) 3 (.83)
1
8 (.65) 2 (.86)
0.75
2 (.85) 5 (.80)
0.5
18 (.71) 3 (.81)
0.35
... (.65) 24 (.76)
0.2
X
… (.71)
0.1
X
X
Y9
2 (.88)
2 (.95)
2 (.85)
2 (.87)
4 (.79)
2 (.83)
… (.70)
X
Group N-M
Y10
2 (.94)
2 (.95)
2 (.86)
2 (.85)
20 (.80)
5 (.80)
… (.81)
X
Criterion = 2 consecutive days at > .80
… = Trained, but did not make criterion
X = Did not receive training
Y2
2 (.90)
2 (.85)
2 (.87)
2 (.88)
3 (.84)
2 (.87)
18 (.76)
… (.63)
Y6
2 (.93)
2 (.93)
6 (.81)
2 (.89)
10 (.76)
… (.67)
X
X
Y8
2 (.86)
2 (.93)
2 (.99)
2 (.89)
2 (.88)
2 (.94)
3 (.87)
… (.62)
As duration decreased
below 1 s, performance
began to deteriorate
 No overall difference
between groups
 Group N-N < Group NM at .75 s
 All durations were
responded to abovechance (.5)

1
*
Discrimination Ratio
0.9
0.8
Group N-N
Group N-M
0.7
0.6
0.5
4
3
2
1
Stimulus Duration (s)
0

The birds learned the scene gist
categorization task and generalized to new
images
 No difference in natural-natural vs. natural-manmade with
longer processing times
 Trend towards better performance with natural-manmade
under processing time challenges
▪ Same general pattern as in humans
 The birds appear to have learned a category
discrimination instead of memorizing individual images
 Generalization decrement <10%

All seven birds were able to categorize the images
with less than 1 s of exposure time, but as delays fell
below .35 s, performance deteriorated
 Two birds did perform above chance at .10 s exposure
 Pigeons appear to require more processing time than
humans
 May be due to greater degree of binocularity in humans or
perhaps due to pigeons’ eyes being on the side
 Kramer (2010) examined rapid discrimination of simple
geometric stimuli and naturalistic slides (containing humans
vs. no humans) with 0.1-10 s exposure time, and also found
that discrimination accuracy was reduced with display times
of 1 s or less.


Do pigeons differ in their scene gist
categorization depending on viewpoint
(aerial vs. terrestrial)?
Comparative approach
 Humans are better with terrestrial compared to aerial
viewpoints
 Experience vs. evolutionary history?
 Pigeons have evolved for flight, so may be better with
aerial?
 Can control experience

Eight pigeons were trained on go/no-go task
(beach vs. mountain)
 Four were from Experiment 1 and four were
experimentally-naïve
 None of the pigeons had any outdoor flight
experience



Three viewpoints: 0, 45 or 90 degrees
Each viewpoint trained individually in a
counterbalanced order
Then, gave intermixed training for 20 days with
all three viewpoints
During the first half of
training, the pigeons
performed better on
the aerial (45 and 90
degree) views than the
terrestrial (0 degree)
 With additional
training, performance
on the terrestrial views
improved to a similar
level as the aerial views

1.0
Discrimination Ratio
First Half
Second Half
0.9
0.8
0.7
0.6
*
0.5
0
45
Viewing Angle
90
During the first half of
training, the pigeons
performed better on
the aerial (45 and 90
degree) views than the
terrestrial (0 degree)
 With additional
training, performance
on the terrestrial views
improved to a similar
level as the aerial views

1.0
Discrimination Ratio
First Half
Second Half
0.9
0.8
0.7
0.6
*
0.5
0
45
Viewing Angle
90
During the first half of
training, the pigeons
performed better on
the aerial (45 and 90
degree) views than the
terrestrial (0 degree)
 With additional
training, performance
on the terrestrial views
improved to a similar
level as the aerial views

1.0
Discrimination Ratio
First Half
Second Half
0.9
0.8
0.7
0.6
*
0.5
0
45
Viewing Angle
90

Pigeons displayed superior categorization of the
aerial views compared to the terrestrial
viewpoint
 This is the opposite pattern to what humans show
 May reflect evolution of the pigeon visual system to capitalize
on scene information from the air
 This can be overcome with experience
Download