Analyzing the view dependence of population codes in inferior temporal cortex

The recent discovery of cells in inferior temporal (IT) cortex that show responses selective for specific views of learned wire-like objects [4] supports a computational model of object recognition built on mio-dimensional view-based representations [7] over models based on three-dimensional structural descriptions [1], in line with many psychophysical demonstrations of view-dependent recognition performance [2]. However, many more IT cells show significantly view-sensitive responses without strict view-tuning, which suggests that an analysis of population encodings may be necessary to fully understand the nature of object representation in IT. By analyzing how the similarity of IT population responses to different views of two objects rotated in depth depends on the orientation difference between the views, we provide further evidence that neural representations of object shape depend on abstract two-dimensional views (possibly built from collections of image features), and demonstrate a clear correspondence between the view dependence of IT population responses and the view dependence of human observers' similarity judgments.