Purchase this article with an account.
Greet Kayaert, Rufin Vogels, Irving Biederman; Single inferior temporal neurons are tuned to metric shape dimensions as well as to nonaccidental differences. Journal of Vision 2002;2(7):687. https://doi.org/10.1167/2.7.687.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Inferior Temporal (IT) neurons contribute to object recognition and categorization by coding for complex object features. Still, it is not clear how their responses can be related to similarities between complex objects. Op de Beeck et al. (2001, Nature Neurosci.) varied the amplitude of 2 radial frequency components of complex 2D-shapes and found a faithful representation of these variations in IT-neurons. This illustrates the importance of systematically varying the underlying dimensions of high-dimensional stimuli in order to find a consistent presentation of a low-dimensional subspace in individual neurons.
We recorded the responses of 80 IT-neurons to 3D instead of 2D-shapes, composed of 1 or 2 single volumes. We parametrically manipulated broadness, height, and curvature or asymmetry of each volume. Most cells were sensitive to at least 2 of these dimensions. Their responses could be accurately predicted by a quadratic regression using these dimensions as factors. (Median r2 = .83, average n of stimuli = 21, average n of dimensions = 3) We could restrict the number of dimensions in the 2-part stimuli from 6 to 4, by averaging broadness and length over the two parts.
The influence of broadness and length on the neuronal responses couldn't be explained by mere sensitivity to area changes. The same neurons were even more sensitive to nonaccidental shape changes and view changes (Kayaert et al. (2001, Soc Neurosci Abstr)). These results indicate that the coding of metric and nonaccidental properties is accomplished by a single population of neurons rather than different subsets of neurons responding to different kinds of variations in object shape.
This PDF is available to Subscribers Only