Human visual representations are not only translation-invariant (like CNNs), but also symmetry-invariant, & invariant for small rotations.
Show this thread
-
Essentially, kids learn orientation-invariant shapes & their orientations separately; when writing they pick the shape *then* an orientation
And the 2nd step is as likely to be mistaken as the first one. Very impressive when occurring with complex visual symbols (e.g. cursive)
