<s>
StyleGAN	B-Application
is	O
a	O
generative	B-Algorithm
adversarial	I-Algorithm
network	I-Algorithm
(	O
GAN	O
)	O
introduced	O
by	O
Nvidia	O
researchers	O
in	O
December	O
2018	O
,	O
and	O
made	O
source	B-License
available	I-License
in	O
February	O
2019	O
.	O
</s>
<s>
StyleGAN	B-Application
depends	O
on	O
Nvidia	O
's	O
CUDA	B-Architecture
software	O
,	O
GPUs	O
,	O
and	O
Google	O
's	O
TensorFlow	B-Language
,	O
or	O
Meta	O
AI	O
's	O
PyTorch	B-Algorithm
,	O
which	O
supersedes	O
TensorFlow	B-Language
as	O
the	O
official	O
implementation	O
library	O
in	O
later	O
StyleGAN	B-Application
versions	O
.	O
</s>
<s>
The	O
second	O
version	O
of	O
StyleGAN	B-Application
,	O
called	O
StyleGAN2	O
,	O
was	O
published	O
on	O
February	O
5	O
,	O
2020	O
.	O
</s>
<s>
Nvidia	O
introduced	O
StyleGAN3	O
,	O
described	O
as	O
an	O
"	O
alias-free	O
"	O
version	O
,	O
on	O
June	O
23	O
,	O
2021	O
,	O
and	O
made	O
source	B-License
available	I-License
on	O
October	O
12	O
,	O
2021	O
.	O
</s>
<s>
A	O
direct	O
predecessor	O
of	O
the	O
StyleGAN	B-Application
series	O
is	O
the	O
Progressive	O
GAN	O
,	O
published	O
in	O
2017	O
.	O
</s>
<s>
In	O
December	O
2018	O
,	O
Nvidia	O
researchers	O
distributed	O
a	O
preprint	O
with	O
accompanying	O
software	O
introducing	O
StyleGAN	B-Application
,	O
a	O
GAN	O
for	O
producing	O
an	O
unlimited	O
number	O
of	O
(	O
often	O
convincing	O
)	O
portraits	O
of	O
fake	O
human	O
faces	O
.	O
</s>
<s>
StyleGAN	B-Application
was	O
able	O
to	O
run	O
on	O
Nvidia	O
's	O
commodity	O
GPU	O
processors	O
.	O
</s>
<s>
In	O
February	O
2019	O
,	O
Uber	B-Application
engineer	O
Phillip	O
Wang	O
used	O
the	O
software	O
to	O
create	O
This	B-Application
Person	I-Application
Does	I-Application
Not	I-Application
Exist	I-Application
,	O
which	O
displayed	O
a	O
new	O
face	O
on	O
each	O
web	O
page	O
reload	O
.	O
</s>
<s>
Wang	O
himself	O
has	O
expressed	O
amazement	O
,	O
given	O
that	O
humans	O
are	O
evolved	O
to	O
specifically	O
understand	O
human	O
faces	O
,	O
that	O
nevertheless	O
StyleGAN	B-Application
can	O
competitively	O
"	O
pick	O
apart	O
all	O
the	O
relevant	O
features	O
(	O
of	O
human	O
faces	O
)	O
and	O
recompose	O
them	O
in	O
a	O
way	O
that	O
's	O
coherent.	O
"	O
</s>
<s>
Similarly	O
,	O
two	O
faculty	O
at	O
the	O
University	O
of	O
Washington	O
's	O
Information	O
School	O
used	O
StyleGAN	B-Application
to	O
create	O
Which	O
Face	O
is	O
Real	O
?,	O
which	O
challenged	O
visitors	O
to	O
differentiate	O
between	O
a	O
fake	O
and	O
a	O
real	O
face	O
side	O
by	O
side	O
.	O
</s>
<s>
The	O
second	O
version	O
of	O
StyleGAN	B-Application
,	O
called	O
StyleGAN2	O
,	O
was	O
published	O
on	O
February	O
5	O
,	O
2020	O
.	O
</s>
<s>
Dubbed	O
"	O
alias-free	O
"	O
,	O
this	O
version	O
was	O
implemented	O
with	O
pytorch	B-Algorithm
.	O
</s>
<s>
In	O
December	O
2019	O
,	O
Facebook	B-Application
took	O
down	O
a	O
network	O
of	O
accounts	O
with	O
false	O
identities	O
,	O
and	O
mentioned	O
that	O
some	O
of	O
them	O
had	O
used	O
profile	O
pictures	O
created	O
with	O
artificial	O
intelligence	O
.	O
</s>
<s>
Here	O
,	O
the	O
functions	O
are	O
image	O
up	O
-	O
and	O
down-sampling	O
functions	O
,	O
and	O
is	O
a	O
blend-in	O
factor	O
(	O
much	O
like	O
an	O
alpha	B-Algorithm
in	O
image	O
composing	O
)	O
that	O
smoothly	O
glides	O
from	O
0	O
to	O
1	O
.	O
</s>
<s>
StyleGAN-1	O
is	O
designed	O
as	O
a	O
combination	O
of	O
Progressive	O
GAN	O
with	O
neural	B-Algorithm
style	I-Algorithm
transfer	I-Algorithm
.	O
</s>
<s>
The	O
key	O
architectural	O
choice	O
of	O
StyleGAN-1	O
is	O
a	O
progressive	O
growth	O
mechanism	O
,	O
similar	O
to	O
Progressive	O
GAN	O
.	O
</s>
<s>
Each	O
style	O
block	O
applies	O
a	O
"	O
style	O
latent	O
vector	O
"	O
via	O
affine	O
transform	O
(	O
"	O
adaptive	O
instance	O
normalization	O
"	O
)	O
,	O
similar	O
to	O
how	O
neural	B-Algorithm
style	I-Algorithm
transfer	I-Algorithm
uses	O
Gramian	B-Algorithm
matrix	I-Algorithm
.	O
</s>
<s>
StyleGAN-2	O
improves	O
upon	O
StyleGAN-1	O
in	O
two	O
ways	O
.	O
</s>
<s>
This	O
was	O
updated	O
by	O
the	O
StyleGAN-2-ADA	O
(	O
"	O
ADA	O
"	O
stands	O
for	O
"	O
adaptive	O
"	O
)	O
,	O
which	O
uses	O
invertible	O
data	O
augmentation	O
.	O
</s>
<s>
StyleGAN-3	O
improves	O
upon	O
StyleGAN-2	O
by	O
solving	O
the	O
"	O
texture	O
sticking	O
"	O
problem	O
,	O
which	O
can	O
be	O
seen	O
in	O
the	O
official	O
videos	O
.	O
</s>
<s>
To	O
solve	O
this	O
,	O
they	O
proposed	O
imposing	O
strict	O
lowpass	B-Algorithm
filters	I-Algorithm
between	O
each	O
generator	O
's	O
layers	O
,	O
so	O
that	O
the	O
generator	O
is	O
forced	O
to	O
operate	O
on	O
the	O
pixels	O
in	O
a	O
way	O
faithful	O
to	O
the	O
continuous	O
signals	O
they	O
represent	O
,	O
rather	O
than	O
operate	O
on	O
them	O
as	O
merely	O
discrete	O
signals	O
.	O
</s>
<s>
The	O
resulting	O
StyleGAN-3	O
is	O
able	O
to	O
generate	O
images	O
that	O
rotate	O
and	O
translate	O
smoothly	O
,	O
and	O
without	O
texture	O
sticking	O
.	O
</s>
