WEBVTT

00:00:00.350 --> 00:00:03.650
So what we see here, it starts with,
first of all a lot of product data,

00:00:03.683 --> 00:00:04.650
especially images.

00:00:04.650 --> 00:00:07.166
So this is a typical collection,
one collection.

00:00:07.166 --> 00:00:09.050
And there are lot during the year.

00:00:09.050 --> 00:00:13.200
So what we do is we take all these images,
run them through various AI models

00:00:13.466 --> 00:00:17.716
to extract simple things
like what is the color of all these items,

00:00:17.716 --> 00:00:20.716
but then also more in detail,
what are the different patterns?

00:00:20.916 --> 00:00:24.750
What are the shirts that do have prints
while other shirts without prints?

00:00:24.750 --> 00:00:27.016
But then even more importantly,

00:00:27.016 --> 00:00:29.966
also to get a rich semantic
understanding, like what

00:00:29.966 --> 00:00:34.016
items that I can wear for wedding, what items
for job interview for example.

00:00:34.016 --> 00:00:36.383
On the left, what are items
that I can wear at the beach?

00:00:37.133 --> 00:00:39.150
And yet to make this even harder,

00:00:39.150 --> 00:00:41.216
Fashion is not just about single items,

00:00:41.216 --> 00:00:43.800
but of course
the whole concept of an outfit.

00:00:43.800 --> 00:00:48.000
So I have to wear several things together
and I have to make this decision

00:00:48.000 --> 00:00:49.350
what fits together.

00:00:49.350 --> 00:00:53.933
So for this, we also use AI again
to also analyze the outfits themselves.

00:00:54.416 --> 00:00:59.266
So we basically take model images,
dissect the outfits to understand

00:00:59.266 --> 00:01:02.116
what are the different items,
how can they be combined,

00:01:02.116 --> 00:01:05.483
in terms of composition, but
also in terms of layering as we see here.

00:01:05.483 --> 00:01:07.133
So what can I wear on top of each other?

00:01:08.116 --> 00:01:11.100
So having all this information,
we can take that.

00:01:11.100 --> 00:01:14.416
And what we do is we essentially
built up a knowledge graph

00:01:14.450 --> 00:01:19.200
out of fashion understanding,
which we call the Fashion DNA.

00:01:20.116 --> 00:01:24.750
And this allows us to, well, essentially
create good recommendations that

00:01:24.750 --> 00:01:28.500
A, the items they work together,
but also then of course importantly

00:01:28.500 --> 00:01:31.950
that they also work with the preferences
of the person in the store.