Andrew Yip <ayip@MIT.EDU>
Open research question/project idea:
Biederman's experiment with jumbled scenes was criticized because 
jumbling the images changed their visual complexity. If test images 
could be created with reordered perceptual structure but with the 
same visual complexity as the original, it could provide a better 
test for the influence of context on object recognition.


Short answer question:
What are the general types of models proposed for the effects of 
scene context on object recognition?


Multiple choice question:
How is information thought to be stored across a saccade?
a) as an abstract conceptual representation
b) as a detailed perception composited from multiple saccades
c) no information is maintained across a saccade

Open research:
1) open research question/project idea:
Consider environments like kitchens and bathrooms. They probably have
similar scene statistics, yet are home to very different collections of
objects. It would be interesting to see whether objects that belong in a
kitchen are less inconsistent in bathroom scenes than objects that belong
in locations with very different statistical properties. Or to phrase the
question another way, one could perform experiments in which the
behavioral
reaction to inconsistent stimuli was the dependent variable and the
similarity of the scene statistics (eg bathroom) of the given scene and
the
scene statistics of the inconsistent object (eg kitchen) is an independent
variable.


1. Open research question: Show scenes to subjects for a given amount of
time and then ask what objects were in the scene. For example, in a
kitchen scene, one would have a loaf of bread, cups, some pots and pans as
well as a teddy bear, and a lawn hose (consistent vs. inconsistent
objects). What would the subject remember best, the consistent objects or
the inconsistent objects due to the contextual scene.


1. If it is accepted that context aids in recognition of objects, what
aspect of the context is it that provides this function?



Further Research:
What happens in experiments where novel, unnatural objects are
placed in natural scenes?


Can there be such a thing as "context agnosia," where patients are not
able
to use context to help recognize objects?


1. The Biederman experiment used line drawings as stimuli. However,
this
may not be the best (or most "ecological") choice when testing a concept
such
as interposition. Are these results repeatable when using photographs
(color
and/or monochrome), in which interposition might be more readily
detectable
than in line drawings?


Research Project:
Replicate Hollingsworth and Henderson's experiment using a larger number
of objects and scenes (instead of having the subjects seeing the same
objects and scenes multiple times).


Does consistent scene context facilitate object recognition or
post-recognition processes?


Are inconsistent objects just more interesting, or is there a delay due to
integration?


2) Short answer:
Briefly describe three models of ways in which context might influence
object recognition.


2. What evidence suggests that scenes are recognized holistically?


Short Answer:
Does context influence the recognition of incongruent objects


2. What is the difference between the perceptual schema idea, the priming
model and functional isolation, which is most plausible?


2. What explanation is given for Palmer^Rs results showing that subjects
are
more accurate in naming an object when it is presented after a scene
inconsistent with the object but visually dissimilar from a consistent
object than when the object is presented after a scene inconsistent with
the object but visually similar to a consistent object?


2. What are the five classes of relations used in the Biedreman
experiment?


What evidence supports the notion that context facilitates object
recognition?




2. Short answer question


How can a priming model of the effects of scene context on
object recognition
account for the inhibitory effect reported by Palmer where a
contextually
inappropriate object is the hardest to recognize if it is
perceptually
similar to a contextually appropriate object?


According to Biederman's analysis of contextual relations of objects, what
is the difference between the following two classes of relations: (1)
probability, position, and familiar size vs. (2) interposition and
support?



3) Multiple choice:


Multiple Choice:
What is the term for when context does not aid in object
recognition?
a. cortical blindness
b. functional isolation
c. perceptual schema
d. reverse priming



Biederman's study of the physical and semantic relations of objects led
him
to all of the following results except:
a) people were less accurate at identifying objects with violations
b) people were slower at identifying object with violations
b) as violations increased, detection decreased
c) violations of physical replations were more disruptive than semantic
violations


One problem with the theory that context aids object recognition is that
it
seems that one must recognize the other objects in the scene to understand
what the context is, which seems a bit circular. How does the Torralba and
Sinha paper get through that objection?
a. The context is parsed in a very rough fashion. Low-resolution
heuristics
are used to make educated guesses about the general class of scene
(street,
skyline, inside).
b. The background or context of the scene can generally be parsed more
quickly than the objects within it, because it is made up of simpler
units.
c. The context and objects are parsed in two different streams of
processing, which then merge.
d. None of the above.


Hydrants are likely to be found
A) on top of mailboxes
B) in the air
C) in the kitchen (rarely)
D) interpenetrating cars


Which of the following experiments found no benefit of context in object
recognition?
a) Palmer's study using inappropriate vs. appropriate objects primed for
scenes.
b) Biederman's study testing 5 type of scene relations.
c) Friedman's study demonstrating people looked longer at inconsistent
objects.
d) Hollingsworth & Henderson's study finding no advantage of consistent
objects in a recognition task.


3. According to Biederman's trial types, which of the following is a
violation of support?
a. car driving through a tree
b. car balanced on top of a sapling
c. car driving under a person's legs
d. car parked in a kitchen



3. What data would not support the view that context enhances object
recognition?
a. Subjects shown only a background image with the desired object missing
are able to predict what type of object belongs in that location.
b. Subjects inaccurately identify an object that is inconsistent with a
scene, though the object is visually similar to an object that would be
consistent with that scene
c. Subjects inaccurately identify an object that is consistent with a
scene, and that object is visually similar to an object that would be
inconsistent with that scene
d. Subjects^R detection of objects decreases as the number of violations
of
physical relationships in a scene increase.


3.
Masked priming effects of scene context on object recognition
are predicted by


(i) the perceptual schema model.
(ii) the priming model.
(iii) the functional isolation model.


3. Which of these is NOT a semantic relation used in the Biederman
experiment?
a) probability
b) support (correct)
c) position
d) familiar size