WEEK 5- ACHIEVING OBJECT CONSTANCY ACROSS VIEWPOINT: DEPTH ROTATIONS, VISUAL AGNOSIA AND PROSOPAGNOSIA Flashcards
1
Q
what does it mean when an object is depth rotated?
A
depth rotations can profoundly alter the global shape of the 2D image of a 3D object (its occluding contour) its symmetry and which features and parts are visible
2
Q
what were there heated debates about for achieving object constancy for depth rotated objects?
A
- from the 1990’s on there were heated debates about whether the data on achieving object constancy for depth rotations best fitted the geon structural description or the multiple views plus transformations accounts
- resolving this issue was hard because Biederman and colleagues (GSDs) not only had a different theory and predictions to theorists like Tarr, Bulthoff and Edelman (multiple views), but they usually also used different experimental methods
3
Q
what was an example of an object used to test the GSD account?
A
a jug
4
Q
what objects were used to test the multiple views plus transformation account?
A
cubes (minor and novel objects)
5
Q
What were the findings of Biederman when he tested the GSD account?
A
- task: implicit priming of speeded basic level naming
- similarity: low between different items
- stimuli: line drawing of familiar every day objects
- results: view invariant
6
Q
what were the findings of Tarr et al testing the multiple views account?
A
- task: explicit subordinate level recognition or matching
- similarity: high between different items
- stimuli: bent cylinders like paperclips were popular as they do not self-occlude and have a fixed number of identical segments so only the angle between segments distinguishes them
- results: highly view sensitive
7
Q
What is the evidence that object recognition is invariant
A
- non- brain damaged people shown familiar, every day objects usually recognise foreshortened views and other unusual, depth rotated views slower than canonical views so object constancy across depth rotation is not perfect.. though the cost is usually not large- responses are usually only a fraction of a second slower and errors are rare
- non brain damaged people shown previously unseen views of novel unfamiliar objects often find them really difficult to recognise and they are also very poor at imagining what such objects would look like from a different angle
- this view-specificity of object recognition is consistent with multiple view theories has been taken as evidence against biederman’s GSD account
- Biederman argued his account was only suppose to explain very basic objects not objects at this subordinate level
8
Q
what did Biederman & Gerhardstein 1993 argue that these view-specific effects were due to?
A
- testing novel objects that:
- failed to meet at least one of the conditions for viewpoint invariance, either because the stimuli did not decompose into a geon structural description or because the set members did not activate distinctive geon structural descriptions or produced non-stable part structures
- they claimed such novel object were unlike most of the familiar objects that we recognise in our everyday lives and they argue that results from novel objects tell us little about normal object recognition
- he argued the objects used in experiments (ie blocks) are harder to identify because they are very similar to each other and they do not have different parts or real volume- they may test subordinate rather than basic level recognition
9
Q
What was the first experiment of B&G 1993?
A
- two blocks of naming
- block 2 stimuli showed the same/different view in depth as Block 1 objects from that category
- the two categories shown here are lamp and torch
- block 1 was a ‘priming’ or ‘study’ block
- block 2 was the test block
- found there was positive priming overall
- more priming for the same exemplar (solid lines) compared to when a different exemplar (dotted lines) was shown
- most importantly, same exemplar priming was similar in size for same view (0 degrees) and different view (67 degrees) conditions
10
Q
does the priming evidence from B&G’s 1993 task support Biederman or Tarr’s account?
A
- on same exemplar trials priming was shown in both blcoks compared to different views were shown
- unless prime and test views revealed different parts or different spatial relations between parts- in which case the views would produce different GSD s so biederman woiuldn’t predict view invariance
- the results of B&G93 support Biederman’s claim that everyday object recognition is largely invariant to depth rotation
- from other, similar, name priming studies, Biederman, Cooper and colleages claimed that human object recognition is invariant to mirror-image reflection, size and position changes of object
11
Q
what evidence did Lawson and Humphreys 1996 find against B&G’s accoutn?
A
- Lawson & Humphreys 1998: speeded sequential matching
- experiment 1 after naming a canonical prime view:
- canonical test views were named faster than foreshortened test views
- experiment 3 after naming a foreshortened prime view:
- foreshortened test views were named faster than canonical test view crtitically, here, priming was greater for the non-canonical (foreshortened) view so these view sepcific effects were not simply due to canonicality
- conclusion: people were faster to recognise test pictures showing a view similar to the initial prime view of that object that they had seen before- even when the second, test picture showed an unusual, foreshortened view
12
Q
what is the evidence for view-specific priming across depth rotation? - question 2
A
- the line drawings of familiar objects used by Lawson and Humphreys 1996 were similar to those used by Biederman and Gerhardstein
- these stimuli seemed to meet the three conditions set by B&G for obtaining view-invariance; they could easily be decomposed into sets of distinctive parts plus spatial relations between the parts so different objects would be assigned distinct GSDs and hence should be distinguishable
- L&H also used the same priming of object naming task as B&G
13
Q
Why might B&G93 have reported view-invariant priming whilst L&H98 and others found view-sensitive priming?
A
- B&G93- trend in many of their studies showed view-sensitivity and maybe they didnt try hard enough to detect view sensitivity (too few tested, too few trials ect?)
- Lawson et al 1998- at least a subset of their stimuli had different GSDs for different views- so view sensitivity not then predicted. also view effects only ever modest in size
14
Q
what have most researchers now agreed about view-sensitvity?
A
- that both view-invariant and view sensitive performance can be found depending on how you test
- we need to understand when and why view-sensitivity occurs
- recent studies hve shown the importance of:
- visual similarity
- level of categorisation
- most of the studies discussed today used different levels of these factors and stimuli were chosen without fully understanding the consequences of this on the results obtained
- variation in view-sensitivity in the results of different studies may largely reflect variation in these two factors
15
Q
what is the conclusion for whether stored object representations are view specific?
A
- Yes AND No
- Biederman argues weak view-sensitivity for familiar objects is due to testing weird accidental views whilst strong view sensitivity for novel objects occurs because these stimuli fail to meet his criteria for producing view-invariant recognition
- as for plane rotation there usually a cost to achieving object constancy for depth rotation (RT often slower and sometimes people make errors)- this is good evidence that we store multiple, view specific representations of the same object
- there is still a debate as to how many different view specific representations are stored, which views are stored, if view invariant representations are also stored and if so, when these are used.
- for example, using ERP data Eddy and Holcomb 2011 found both view specific and view invariant priming depending on prime duration and the size of depth rotation
