CVPR Daily - Tuesday

4 / 30

Vittorio, do we have in common an
Italian background?
Almost right - I am Swiss, from the
Italian part of Switzerland.
So we are neighbours! I am from
Milan, and there we say that we are
nearly Swiss.
Actually, in Lugano, I have been told by
the Swiss Germans that I am nearly
Italian. [We both have to laugh]
So you grew up in the Italian part of
Switzerland, then decided to become
a scientist and your career brought
you to different places.
Yes. Interestingly enough, when I was a
kid, there was no university in Lugano,
my hometown. So even just to study,
before being a scientist, you had to
move. That started my journey about
half a life ago.
Where has this journey led you, what
are you working on now?
I work on various problems in
computer vision, but my general life
mission since 5-6 years is to try to
learn computer vision models that are
able to localise objects in images with
high quality and least human
intervention possible at the same time.
This sometimes is described as weakly
supervised learning. Recently I have
been working a lot on human-in-the-
loop learning, where there is a bit of
human supervision and intervention
during training. For example, as a
typical outcome you would get a
model that is capable of labelling every
pixel in the image that is containing an
object, but at training time you
perhaps only need image-level labels.
And yet, you can get a localisation
model almost out of thin air.
Why are you so passionate about
these kind of problems?
There are two really good reasons. One
is an intrinsic, scientific reason and the
“I salute the
students that are
braving the new
world in these days.”
Vittorio Ferrari
4
Vittorio Ferrari is a professor at the
University of Edinburgh and a
research scientist at Google Zurich,
in both of which he runs a research
group.