PDF

Description

Deep neural networks for object classification have been used to build a map of low-dimensional object space in the primate IT cortex to describe images of general objects. Corresponding responses of IT cells to these objects show that single IT cells project them onto axes of this space. Through a similar process, it has been found that single cells are tuned to specific face axes that describe specific facial features. In this work, we use deep neural networks to build a common space for both faces and non-face objects. By generating images along axes in this space that would cause the maximal response of cells, we aim to answer the question of whether the encoding scheme of face cells could be used to encode non-face objects and produce meaningful code for general objects.

Details

Files

Statistics

from
to
Export
Download Full History