Hi,
I will be running an experiment that requires participants to distinguish between several novel objects. Ideally, each novel object will be a configuration of 3D geometric shapes (e.g., pyramids, pentagonal-prisms, spirals, discs, cuboids) but objects *cannot* be distinguished from one another based on one particular local feature: the only defining aspect of an object should be its overall configuration.
For example, if we have object A (a configuration of a cuboid, cylinder, and a pyramid) for each of its features there will be at least one other novel object that contains the identical feature (e.g., object B might have the identical cuboid, object C might have the identical pyramid, and so on…) - and thus the objects cannot be differentiated based on local features, and must be differentiated by overall configuration instead. So I’m looking for a stimuli set where features have been manipulated systematically such that objects can be distinguished only by their configuration of features (something corresponding to the linked table would be ideal):
https://docs.google.com/document/d/1DBjYjnbzKbRJABlsyk93KeNIKnEW030OQRibVgcug1U/edit?usp=sharing
Has such a stimuli set has been used in the past, and if so has it been made available? Any suggestions welcome.
Ryan
This is a very good question.
One thing that sometimes goes unnoticed is polyhedra edges and surfaces that are curved instead of straight. For example, an approach to replacing planar polytopes with curved edges is given in
Article The Virtual Element Method with curved edges
A basic approach to moving from straight edges on a cube, for example, is to project the cube onto the surface of a sphere. See, e.g.,
Article Trading spaces: Building three-dimensional nets from two-dim...
I mention curved rather than straight edges because it opens up the possibility of classify shapes based on the curvature of their edges and surfaces. This also paves the way to pigeonholing different curvilinear shapes in shape classes based on the average curvature of either the edges of surfaces for the members of a shape class. So, for example, a fearture to consider in classifying shapes is uniform curvature.
An approach to measuring uniform curvature is given in
Article Geometry and curvature of spin networks
The Q is very interesting & sharing may not be easy for an old Biology
person. But I'll try my level best to be a participant. And quit if you recommend that I'm not up to the mark.
Katherine Twomey has used objects with similar properties and used 3D-printing to create these shapes. As far as I know, a recent paper uses animals of various configurations, so they don't use basic geometric configurations, but I think she has been using simpler stuff as well.
One of the papers that's using the animal configurations is Twomey, K. E., & Westermann, G. (2017). Curiosity-based learning in infants: a neurocomputational approach. Developmental Science, e12629. doi: 10.1111/desc.12629
Hi Ryan
Take a look at Gauthier T, Tarr MJ Becoming a Greeble Expert etc..., Vision Research 1997, 37, pp 1673-1681.
It may help.
Regards
Michael
Dear Dr M Bross:
Could not get to see the Ref U mention.
Where do we find it?
May be U can help me offering its copy.
Sure that will be great & kind of U.
Dear Dr. P. Khurana:
Here is the article, you' find many much more her work on object/face recognition on her page on research gate.
Regards
Michael
Hi, Ryan
The question is very interesting and practical. You can represent geometric shapes object by its shape number. First you find segment the object then find out the shape number. It is explained in Digital Image Processing book by Gonzalez.
Many thanks for your insights everyone - I am currently considering all of them. A colleague of mine has also recommended the following database for novel objects, which might be useful for other people coming to this question: http://wiki.cnbc.cmu.edu/Novel_Objects
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Computer Graphics and Animation
- 3D