Learning vision is easy?

The visual cortex is a  major area in the human brain and the majority of input from the  environment is perceived by vision. Still, it was believed for decades  that computer vision is a simple problem that can be  addressed by student projects or clever system design based on concepts  from related fields. About a decade ago, the field observed a major  breakthrough with the advent of deep learning, to a large extent enabled  by means of novel computational resources in  form of GPUs.

Since then, the field  basically exploded and thousands of black-box approaches to solve vision  problems have been developed. However, it turned out that previously  known concepts such as uncertainty and geometry  are still relevant when we want to better understand deep learning  methods and achieve state of the art performance.