<s>
The	O
Inception	B-General_Concept
Score	I-General_Concept
(	O
IS	O
)	O
is	O
an	O
algorithm	O
used	O
to	O
assess	O
the	O
quality	O
of	O
images	O
created	O
by	O
a	O
generative	O
image	O
model	O
such	O
as	O
a	O
generative	B-Algorithm
adversarial	I-Algorithm
network	I-Algorithm
(	O
GAN	O
)	O
.	O
</s>
<s>
The	O
score	O
is	O
calculated	O
based	O
on	O
the	O
output	O
of	O
a	O
separate	O
,	O
pretrained	O
Inceptionv3	B-General_Concept
image	O
classification	O
model	O
applied	O
to	O
a	O
sample	O
of	O
(	O
typically	O
around	O
30,000	O
)	O
images	O
generated	O
by	O
the	O
generative	O
model	O
.	O
</s>
<s>
The	O
Inception	B-General_Concept
Score	I-General_Concept
is	O
maximized	O
when	O
the	O
following	O
conditions	O
are	O
true	O
:	O
</s>
<s>
The	O
entropy	O
of	O
the	O
distribution	O
of	O
labels	O
predicted	O
by	O
the	O
Inceptionv3	B-General_Concept
model	O
for	O
the	O
generated	O
images	O
is	O
minimized	O
.	O
</s>
<s>
It	O
has	O
been	O
somewhat	O
superseded	O
by	O
the	O
related	O
Fréchet	B-Algorithm
inception	I-Algorithm
distance	I-Algorithm
.	O
</s>
<s>
While	O
the	O
Inception	B-General_Concept
Score	I-General_Concept
only	O
evaluates	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>
The	O
Inception	B-General_Concept
Score	I-General_Concept
of	O
relative	O
to	O
isEquivalent	O
rewrites	O
include	O
is	O
nonnegative	O
by	O
Jensen	O
's	O
inequality	O
.	O
</s>
<s>
A	O
higher	O
inception	B-General_Concept
score	I-General_Concept
is	O
interpreted	O
as	O
"	O
better	O
"	O
,	O
as	O
it	O
means	O
that	O
is	O
a	O
"	O
sharp	O
and	O
distinct	O
"	O
collection	O
of	O
pictures	O
.	O
</s>
<s>
The	O
highest	O
inception	B-General_Concept
score	I-General_Concept
is	O
achieved	O
if	O
and	O
only	O
if	O
the	O
two	O
conditions	O
are	O
both	O
true	O
:	O
</s>
