<s>
Deep	B-Algorithm
image	I-Algorithm
compositing	I-Algorithm
is	O
a	O
way	O
of	O
compositing	O
and	O
rendering	O
digital	O
images	O
that	O
emerged	O
in	O
the	O
mid-2010s	O
.	O
</s>
<s>
Deep	O
data	O
is	O
encoded	O
by	O
advanced	O
3D	O
renderers	O
into	O
an	O
image	O
that	O
samples	O
information	O
about	O
the	O
path	O
each	O
rendered	O
pixel	B-Algorithm
takes	O
along	O
the	O
z	O
axis	O
extending	O
outward	O
from	O
the	O
virtual	O
camera	O
through	O
space	O
,	O
including	O
the	O
color	O
and	O
opacity	O
of	O
every	O
non-opaque	O
surface	O
or	O
volume	O
it	O
passes	O
through	O
along	O
the	O
way	O
,	O
as	O
well	O
as	O
neighboring	O
samples	O
.	O
</s>
<s>
It	O
might	O
be	O
considered	O
somewhat	O
analogous	O
to	O
the	O
way	O
ray	B-Algorithm
tracing	I-Algorithm
generates	O
simulated	O
photon	O
paths	O
through	O
such	O
mediums	O
;	O
however	O
,	O
ray	B-Algorithm
tracing	I-Algorithm
and	O
other	O
traditional	O
rendering	O
techniques	O
generally	O
produce	O
images	O
that	O
contain	O
only	O
three	O
or	O
four	O
channels	O
of	O
color	O
and	O
opacity	O
values	O
per	O
pixel	B-Algorithm
,	O
flattened	O
into	O
a	O
two	O
dimensional	O
frame	O
.	O
</s>
<s>
Even	O
the	O
popular	O
addition	O
of	O
cryptomattes	B-Operating_System
to	O
many	O
post-production	O
and	O
VFX	O
studios	O
 '	O
pipelines	O
,	O
while	O
providing	O
separate	O
color-coded	O
ID	O
shapes	O
for	O
individual	O
elements	O
in	O
a	O
rendered	O
scene	O
to	O
further	O
bridge	O
the	O
gap	O
between	O
CGI	O
and	O
compositing	O
,	O
do	O
n't	O
allow	O
for	O
the	O
nearly	O
automated	O
and	O
fully	O
non-linear	O
workflows	O
that	O
deep	O
data	O
does	O
.	O
</s>
<s>
One	O
downside	O
to	O
the	O
use	O
of	O
deep	O
images	O
is	O
their	O
substantial	O
file	O
size	O
,	O
since	O
they	O
encode	O
a	O
relatively	O
enormous	O
amount	O
of	O
data	O
per	O
frame	O
compared	O
to	O
even	O
multichannel	O
formats	O
such	O
as	O
OpenEXR	B-Application
.	O
</s>
<s>
Edge	O
artifacts	O
are	O
reduced	O
as	O
transparent	O
pixels	B-Algorithm
have	O
more	O
data	O
to	O
work	O
with	O
.	O
</s>
