<s>
The	O
Fréchet	B-Algorithm
inception	I-Algorithm
distance	I-Algorithm
(	O
FID	O
)	O
is	O
a	O
metric	O
used	O
to	O
assess	O
the	O
quality	O
of	O
images	O
created	O
by	O
a	O
generative	O
model	O
,	O
like	O
a	O
generative	B-Algorithm
adversarial	I-Algorithm
network	I-Algorithm
(	O
GAN	O
)	O
.	O
</s>
<s>
Unlike	O
the	O
earlier	O
inception	B-General_Concept
score	I-General_Concept
(	O
IS	O
)	O
,	O
which	O
evaluates	O
only	O
the	O
distribution	O
of	O
generated	O
images	O
,	O
the	O
FID	O
compares	O
the	O
distribution	O
of	O
generated	O
images	O
with	O
the	O
distribution	O
of	O
a	O
set	O
of	O
real	O
images	O
(	O
"	O
ground	O
truth	O
"	O
)	O
.	O
</s>
<s>
For	O
two	O
multidimensional	O
Gaussian	O
distributions	O
and	O
,	O
it	O
is	O
explicitly	O
solvable	O
asThis	O
allows	O
us	O
to	O
define	O
the	O
FID	O
in	O
pseudocode	B-Language
form:INPUT	O
a	O
function	O
.	O
</s>
<s>
RETURN	O
.In	O
most	O
practical	O
uses	O
of	O
the	O
FID	O
,	O
is	O
the	O
space	O
of	O
images	O
,	O
and	O
is	O
an	O
Inception	B-General_Concept
v3	I-General_Concept
model	O
trained	O
on	O
the	O
ImageNet	B-General_Concept
,	O
but	O
without	O
its	O
final	O
classification	O
layer	O
.	O
</s>
<s>
Rather	O
than	O
directly	O
comparing	O
images	O
pixel	O
by	O
pixel	O
(	O
for	O
example	O
,	O
as	O
done	O
by	O
the	O
L2	O
norm	O
)	O
,	O
the	O
FID	O
compares	O
the	O
mean	O
and	O
standard	O
deviation	O
of	O
the	O
deepest	O
layer	O
in	O
Inception	B-General_Concept
v3	I-General_Concept
.	O
</s>
<s>
Finally	O
,	O
while	O
FID	O
is	O
more	O
consistent	O
with	O
human	O
judgement	O
than	O
previously	O
used	O
inception	B-General_Concept
score	I-General_Concept
,	O
</s>
