Brain MRI *Conditional atlases* synthesized by our model as a function of age.
Left: single atlas for entire population
Right: atlases sampled for ages 15 - 90
Experimental faces from CelebAMask-HQ dataset, which are affinely aligned (eye features will match across population). Attributes: Female 0/1, Smiling 0/1, Young 0/1.
Young smiling female
aligned average (left) vs our conditional atlas (right)
Young smiling male
aligned average (left) vs our conditional atlas (right)
More experimentally, here is a visualization of a templates for MNIST digits, where only digits were given during training, but at inference time we also visualize connecting atlases (e.g. "2.7").
