<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
landmark	B-Application
detection	I-Application
is	O
the	O
process	O
of	O
finding	O
significant	O
landmarks	B-Protocol
in	O
an	O
image	O
.	O
</s>
<s>
This	O
originally	O
referred	O
to	O
finding	O
landmarks	B-Protocol
for	O
navigational	O
purposes	O
–	O
for	O
instance	O
,	O
in	O
robot	O
vision	O
or	O
creating	O
maps	O
from	O
satellite	O
images	O
.	O
</s>
<s>
It	O
also	O
has	O
important	O
applications	O
in	O
medicine	O
,	O
identifying	O
anatomical	O
landmarks	B-Protocol
in	O
medical	B-Application
images	I-Application
.	O
</s>
<s>
Finding	O
facial	O
landmarks	B-Protocol
is	O
an	O
important	O
step	O
in	O
facial	O
identification	O
of	O
people	O
in	O
an	O
image	O
.	O
</s>
<s>
Facial	O
landmarks	B-Protocol
can	O
also	O
be	O
used	O
to	O
extract	O
information	O
about	O
mood	O
and	O
intention	O
of	O
the	O
person	O
.	O
</s>
<s>
Methods	O
used	O
fall	O
in	O
to	O
three	O
categories	O
:	O
holistic	O
methods	O
,	O
constrained	B-Application
local	O
model	O
methods	O
,	O
and	O
regression-based	O
methods	O
.	O
</s>
<s>
Holistic	O
methods	O
are	O
pre-progammed	O
with	O
statistical	O
information	O
on	O
face	O
shape	O
and	O
landmark	B-Protocol
location	O
coefficients	O
.	O
</s>
<s>
The	O
classic	O
holistic	O
method	O
is	O
the	O
active	B-General_Concept
appearance	I-General_Concept
model	I-General_Concept
(	O
AAM	O
)	O
introduced	O
in	O
1998	O
.	O
</s>
<s>
Analytical	O
methods	O
apply	O
nonlinear	B-Algorithm
optimization	I-Algorithm
methods	O
such	O
as	O
the	O
Gauss	B-Algorithm
–	I-Algorithm
Newton	I-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
This	O
algorithm	O
is	O
very	O
slow	O
but	O
better	O
ones	O
have	O
been	O
proposed	O
such	O
as	O
the	O
project	O
out	O
inverse	O
compositional	O
(	O
POIC	B-Application
)	O
algorithm	O
and	O
the	O
simultaneous	O
inverse	O
compositional	O
(	O
SIC	O
)	O
algorithm	O
.	O
</s>
<s>
These	O
can	O
use	O
linear	B-General_Concept
regression	I-General_Concept
,	O
nonlinear	O
regression	O
and	O
other	O
fitting	O
methods	O
.	O
</s>
<s>
Other	O
extensions	O
to	O
the	O
basic	O
AAM	O
method	O
analyse	O
wavelets	O
in	O
the	O
image	O
rather	O
than	O
pixel	B-Algorithm
intensity	O
.	O
</s>
<s>
The	O
purpose	O
of	O
landmark	B-Application
detection	I-Application
in	O
fashion	O
images	O
is	O
for	O
classification	O
purposes	O
.	O
</s>
<s>
An	O
example	O
of	O
a	O
fashion	O
landmark	B-Protocol
is	O
the	O
location	O
of	O
the	O
hemline	O
of	O
a	O
dress	O
.	O
</s>
<s>
Fashion	O
landmark	B-Application
detection	I-Application
is	O
particularly	O
difficult	O
due	O
to	O
the	O
extreme	O
deformation	O
that	O
can	O
occur	O
in	O
clothing	O
.	O
</s>
<s>
Some	O
classical	O
methods	O
of	O
feature	O
detection	O
such	O
as	O
scale-invariant	B-Algorithm
feature	I-Algorithm
transform	I-Algorithm
have	O
been	O
used	O
in	O
the	O
past	O
.	O
</s>
<s>
However	O
,	O
it	O
is	O
now	O
more	O
common	O
to	O
use	O
deep	B-Algorithm
learning	I-Algorithm
methods	O
.	O
</s>
<s>
These	O
methods	O
include	O
regression-based	O
models	O
,	O
constraint-based	O
models	O
,	O
and	O
attentive	B-General_Concept
models	O
.	O
</s>
<s>
The	O
particular	O
problems	O
of	O
fashion	O
landmark	B-Application
detection	I-Application
(	O
deformtion	O
)	O
have	O
led	O
to	O
pose	O
estimation	O
models	O
which	O
detect	O
and	O
take	O
into	O
account	O
the	O
pose	O
of	O
the	O
model	O
wearing	O
the	O
clothes	O
.	O
</s>
<s>
There	O
are	O
several	O
algorithms	O
for	O
locating	O
landmarks	B-Protocol
in	O
images	O
.	O
</s>
<s>
Nowadays	O
the	O
task	O
usually	O
is	O
solved	O
using	O
Artificial	B-Architecture
Neural	I-Architecture
Networks	I-Architecture
and	O
especially	O
Deep	B-Algorithm
Learning	I-Algorithm
algorithms	O
,	O
but	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
such	O
as	O
particle	B-Algorithm
swarm	I-Algorithm
optimization	I-Algorithm
are	O
also	O
can	O
be	O
useful	O
to	O
perform	O
this	O
task	O
.	O
</s>
<s>
Deep	B-Algorithm
learning	I-Algorithm
has	O
had	O
a	O
significant	O
impact	O
on	O
autonomous	O
facial	O
landmark	B-Application
detection	I-Application
by	O
enabling	O
more	O
accurate	O
and	O
efficient	O
detection	O
of	O
landmarks	B-Protocol
in	O
real-world	O
photos	O
.	O
</s>
<s>
With	O
traditional	O
computer	B-Application
vision	I-Application
techniques	O
,	O
detecting	O
facial	O
landmarks	B-Protocol
could	O
be	O
challenging	O
due	O
to	O
variations	O
in	O
lighting	O
,	O
head	O
position	O
,	O
and	O
occlusion	O
,	O
but	O
Convolutional	B-Architecture
Neural	I-Architecture
Networks	I-Architecture
(	O
CNNs	B-Architecture
)	O
,	O
have	O
revolutionized	O
landmark	B-Application
detection	I-Application
by	O
allowing	O
computers	O
to	O
learn	O
the	O
features	O
from	O
large	O
datasets	O
of	O
images	O
.	O
</s>
<s>
By	O
training	O
a	O
CNN	B-Architecture
on	O
a	O
dataset	O
of	O
images	O
with	O
labeled	O
facial	O
landmarks	B-Protocol
,	O
the	O
algorithm	O
can	O
learn	O
to	O
detect	O
these	O
landmarks	B-Protocol
in	O
new	O
images	O
with	O
high	O
accuracy	O
even	O
when	O
they	O
appear	O
in	O
different	O
lighting	O
conditions	O
,	O
at	O
different	O
angles	O
,	O
or	O
in	O
partially	O
occluded	O
views	O
.	O
</s>
<s>
In	O
particular	O
,	O
solutions	O
based	O
on	O
this	O
approach	O
have	O
achieved	O
real-time	O
efficiency	O
on	O
mobile	O
devices	O
 '	O
GPUs	B-Architecture
and	O
found	O
its	O
usage	O
within	O
Augmented	B-General_Concept
Reality	I-General_Concept
applications	O
.	O
</s>
<s>
Evolutionary	B-Algorithm
algorithms	I-Algorithm
at	O
the	O
training	O
stage	O
try	O
to	O
learn	O
the	O
method	O
of	O
correct	O
determination	O
of	O
landmarks	B-Protocol
.	O
</s>
<s>
As	O
a	O
result	O
of	O
the	O
completion	O
of	O
the	O
last	O
iteration	O
,	O
a	O
system	O
will	O
be	O
obtained	O
that	O
can	O
correctly	O
determine	O
the	O
landmark	B-Protocol
with	O
a	O
certain	O
accuracy	O
.	O
</s>
<s>
In	O
the	O
particle	B-Algorithm
swarm	I-Algorithm
optimization	I-Algorithm
method	O
,	O
there	O
are	O
particles	O
that	O
search	O
for	O
landmarks	B-Protocol
,	O
and	O
each	O
of	O
them	O
uses	O
a	O
certain	O
formula	O
in	O
each	O
iteration	O
to	O
optimize	O
landmark	B-Application
detection	I-Application
.	O
</s>
