I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI)
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
,
p
p
.
3
8
5
8
~
3
8
6
8
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
1
3
.
i
4
.
p
p
3
8
5
8
-
3
8
6
8
3858
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Deep learni
ng
appro
a
ch f
o
r f
o
re
nsic facia
l recons
tru
ction
depends o
n
unide
ntif
ied
skull
Do
a
a
M
.
M
o
ha
m
m
ed
,
M
o
s
t
a
f
a
E
lg
endy
,
M
o
ha
m
ed
T
a
ha
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
e
r
s
a
n
d
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
B
e
n
h
a
U
n
i
v
e
r
si
t
y
,
B
e
n
h
a
,
E
g
y
p
t
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
11
,
2
0
2
3
R
ev
is
ed
Feb
24
,
2
0
2
4
Acc
ep
ted
Mar
21
,
2
0
2
4
F
a
c
ial
re
c
o
n
stru
c
ti
o
n
,
o
r
fa
c
ial
a
p
p
r
o
x
ima
ti
o
n
,
is
a
n
e
ss
e
n
ti
a
l
p
ro
b
lem
in
a
c
rimin
a
l
in
v
e
stig
a
ti
o
n
i
n
v
o
l
v
in
g
r
e
c
o
n
stru
c
ti
n
g
a
v
ictim'
s
fa
c
e
fro
m
h
is
sk
u
ll
to
d
e
term
in
e
t
h
e
v
ict
im'
s
id
e
n
ti
fic
a
ti
o
n
a
t
a
c
rime
sc
e
n
e
.
F
a
c
ial
a
p
p
r
o
x
ima
ti
o
n
p
lay
s
a
c
ru
c
ial
p
a
rt
wh
e
n
t
h
e
re
is
a
lac
k
o
f
c
lu
e
s
wit
h
in
v
e
stig
a
to
rs.
In
v
e
stig
a
t
o
rs
u
ti
li
z
e
fa
c
ial
a
p
p
r
o
x
ima
ti
o
n
t
o
g
u
e
ss
th
e
v
ictims
'
id
e
n
t
it
ies
.
Th
i
s
re
se
a
rc
h
a
tt
e
m
p
ted
to
u
se
c
o
m
p
u
ter
-
a
id
e
d
fa
c
e
re
c
o
n
stru
c
ti
o
n
ra
th
e
r
t
h
a
n
trad
it
io
n
a
l
a
p
p
ro
a
c
h
e
s.
Trad
i
ti
o
n
a
l
m
e
th
o
d
s
o
f
fa
c
e
re
c
o
n
stru
c
ti
o
n
i
n
c
lu
d
e
t
h
e
u
se
o
f
c
lay
o
r
g
y
p
su
m
.
Tra
d
it
io
n
a
l
p
r
o
c
e
d
u
re
s
n
e
c
e
ss
it
a
te
fo
re
n
sic
p
ro
fe
ss
io
n
a
ls
to
re
b
u
il
d
t
h
e
v
icti
m
'
s
fa
c
e
.
Th
is
re
se
a
rc
h
u
se
s
th
e
c
o
n
v
o
l
u
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
sk
u
ll
p
a
rt
wit
h
sift
(CNN
S
P
S
)
m
o
d
e
l
is
e
m
p
lo
y
e
d
t
o
re
c
o
n
stru
c
t
fa
c
ial
fe
a
tu
re
s
fro
m
a
sk
u
ll
ima
g
e
u
ti
li
z
in
g
p
u
b
l
ic
d
a
tas
e
ts
Ce
leb
AMas
k
-
HQ
a
n
d
M
UG
5
0
0
+
.
Th
e
p
ro
p
o
se
d
a
l
g
o
rit
h
m
wa
s
tes
ted
o
n
u
n
i
d
e
n
ti
fie
d
sk
u
ll
d
a
tab
a
se
s,
a
n
d
c
e
leb
rit
y
fa
c
e
s
we
re
u
se
d
.
T
h
e
g
e
n
u
i
n
e
d
a
tas
e
ts are
n
o
t
a
v
a
il
a
b
le,
wh
ich
i
s th
e
k
e
y
issu
e
i
n
th
is r
e
se
a
rc
h
.
K
ey
w
o
r
d
s
:
Facial
ap
p
r
o
x
im
atio
n
Facial
r
ec
o
n
s
tr
u
ctio
n
Fo
r
en
s
ic
d
ee
p
lear
n
i
n
g
Un
id
en
tifie
d
s
k
u
lls
Victim
's f
ac
e
Victim
's id
en
tific
atio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Do
aa
M
.
Mo
h
am
m
e
d
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
C
o
lleg
e
o
f
C
o
m
p
u
ter
s
an
d
Ar
tific
ial
I
n
tellig
en
ce
,
B
en
h
a
Un
iv
er
s
ity
Far
ee
d
Nad
a
Stre
et,
B
an
h
a,
Al
-
Qaly
u
b
ia
Go
v
e
r
n
o
r
ate
1
3
5
1
1
,
E
g
y
p
t
E
m
ail: d
o
aa
.
ib
r
a
h
em
@
f
ci.
b
u
.
e
d
u
.
eg
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
tr
ad
itio
n
al
ap
p
r
o
ac
h
es
f
o
r
h
u
m
a
n
id
en
tific
atio
n
a
r
e
in
e
f
f
ec
tiv
e
in
s
o
m
e
f
o
r
en
s
ic
in
v
e
s
tig
atio
n
s
.
T
h
e
p
o
lice
m
ay
h
av
e
ju
s
t
a
f
e
w
clu
es
to
id
e
n
tify
t
h
e
in
d
i
v
id
u
al,
DNA,
f
in
g
er
p
r
i
n
ts
,
an
d
d
e
n
tal
r
ec
o
r
d
s
ar
e
t
h
e
m
o
s
t
co
m
m
o
n
id
en
tific
atio
n
t
ec
h
n
iq
u
es
[
1
]
,
[
2
]
.
T
h
e
id
e
n
tify
in
g
tec
h
n
iq
u
es
n
ee
d
a
k
n
o
wn
in
d
iv
id
u
al
with
wh
o
m
to
co
m
p
ar
e
ev
id
en
ce
.
I
t
is
ch
allen
g
in
g
to
m
atch
d
ata
with
r
ec
o
r
d
s
f
r
o
m
an
en
tire
c
o
m
m
u
n
ity
if
th
e
r
e
ar
e
n
o
s
u
s
p
ec
ts
f
o
r
i
d
en
tific
atio
n
.
I
n
o
t
h
er
cir
cu
m
s
tan
ce
s
,
a
u
th
o
r
ities
m
ay
u
s
e
a
d
if
f
er
en
t
t
ec
h
n
i
q
u
e.
Fo
r
en
s
ic
f
ac
ial
r
ec
o
n
s
tr
u
ctio
n
tec
h
n
o
lo
g
y
is
u
s
ed
to
aid
in
id
e
n
tify
in
g
h
u
m
a
n
r
em
ain
[
3
]
–
[
5
]
.
Fo
r
en
s
ic
f
ac
e
r
ec
o
n
s
tr
u
ctio
n
o
r
f
o
r
e
n
s
ic
f
ac
e
ap
p
r
o
x
im
ati
o
n
is
a
tech
n
iq
u
e
u
s
ed
to
r
e
co
n
s
tr
u
ct
a
p
er
s
o
n
'
s
f
ac
ial
f
ea
tu
r
es
f
r
o
m
b
o
n
e
r
e
m
ain
s
.
Ph
y
s
ical
d
ata
is
u
s
ed
to
r
ec
o
n
s
tr
u
ct
f
ac
es
f
r
o
m
g
iv
en
im
ag
es
o
r
v
id
eo
s
.
Facial
r
ec
o
n
s
tr
u
ctio
n
is
s
ig
n
if
ican
t
in
a
v
ar
iety
o
f
f
ield
s
.
Facial
r
ec
o
n
s
tr
u
ctio
n
is
u
s
ed
in
f
o
r
en
s
ics,
ar
ch
ae
o
lo
g
y
,
an
d
an
th
r
o
p
o
lo
g
y
to
cr
ea
te
a
s
cu
lp
ted
m
o
d
el
o
f
an
u
n
k
n
o
wn
f
ac
e
[
6
]
.
Un
i
d
en
ti
f
ied
h
u
m
a
n
r
em
ain
s
ca
n
b
e
id
en
tifie
d
th
r
o
u
g
h
f
ac
i
al
r
ec
o
n
s
tr
u
ctio
n
d
u
e
to
th
e
la
ck
o
f
clu
es
an
d
th
e
g
lo
b
al
s
p
r
ea
d
o
f
m
u
r
d
er
ca
s
es.
T
h
e
p
o
lice
u
s
e
f
ac
e
r
ec
o
n
s
tr
u
c
tio
n
to
id
en
tif
y
h
u
m
an
r
e
m
ain
s
at
cr
im
e
s
ce
n
es
[
7
]
.
Acc
o
r
d
in
g
t
o
wo
r
ld
p
o
p
u
latio
n
r
ev
iew,
th
e
co
m
p
r
eh
en
s
iv
e
r
ate
o
f
cr
im
e
is
ca
lcu
lated
b
y
t
h
e
o
v
er
all
cr
im
es
r
ep
o
r
ted
ac
r
o
s
s
v
ar
i
o
u
s
ca
teg
o
r
ies
d
iv
id
ed
b
y
th
e
o
v
er
all
p
o
p
u
latio
n
,
th
en
m
u
ltip
lied
b
y
1
0
0
,
0
0
0
.
T
h
er
e
is
a
d
is
tin
ct
a
s
s
o
ciatio
n
b
etwe
en
cr
im
e
an
d
ag
e,
as
th
e
m
aj
o
r
ity
o
f
o
f
f
en
s
es,
s
ev
er
e
o
n
es,
ar
e
p
er
p
etr
ated
b
y
in
d
iv
id
u
als
in
th
e
ag
e
r
an
g
e
o
f
2
0
to
3
0
y
ea
r
s
o
ld
.
Ven
ez
u
ela
h
o
ld
s
th
e
t
o
p
s
p
o
t f
o
r
th
e
h
ig
h
est
in
d
ex
o
f
cr
im
e
g
lo
b
ally
at
8
3
.
7
6
,
th
e
n
Pap
u
a
New
Gu
in
ea
at
8
0
.
7
9
o
n
th
e
cr
im
e
in
d
ex
,
an
d
So
u
t
h
Af
r
ica
r
an
k
s
as
th
e
th
ir
d
-
h
ig
h
est cr
im
e
r
ate
wo
r
l
d
wid
e
[
8
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2
2
5
2
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
fo
r
fo
r
en
s
ic
fa
cia
l reco
n
s
tr
u
ctio
n
.
.
.
(
Do
a
a
M.
Mo
h
a
mme
d
)
3859
Acc
o
r
d
in
g
t
o
th
e
r
ep
o
r
t,
h
o
m
icid
e
r
ates
ca
n
b
e
i
n
f
lu
en
ce
d
b
y
v
ar
io
u
s
s
o
cio
ec
o
n
o
m
ic
f
ac
t
o
r
s
.
T
h
ese
r
ea
s
o
n
s
in
clu
d
e
s
o
cio
ec
o
n
o
m
ic
in
eq
u
ality
,
u
n
em
p
lo
y
m
e
n
t,
p
o
liti
ca
l
in
s
tab
ilit
y
,
g
u
n
p
o
s
s
ess
io
n
,
an
d
i
n
p
ar
ticu
lar
,
g
an
g
s
,
o
r
g
an
ized
cr
im
e,
an
d
th
e
d
r
u
g
tr
ad
e.
I
n
d
ee
d
,
"Ap
p
r
o
x
im
ately
6
5
,
0
0
0
k
illi
n
g
s
p
er
y
ea
r
wer
e
ass
o
ciate
d
with
o
r
g
a
n
ized
g
an
g
s
,
an
d
cr
im
e
b
etwe
en
2
0
0
0
an
d
2
0
1
7
,
c
o
n
s
titu
tin
g
ap
p
r
o
x
im
ately
1
9
%
o
f
all
r
ec
o
r
d
e
d
h
o
m
icid
es wo
r
ld
wid
e
in
2
0
1
7
.
"
As s
h
o
wn
in
Fig
u
r
es 1
an
d
2
[
8
]
,
[
9
]
.
Fig
u
r
e
1
.
Statis
tics
o
f
cr
im
e
r
a
te
in
2
0
2
4
[
8
]
Fig
u
r
e
2
.
Statis
tics
o
n
th
e
n
u
m
b
er
o
f
v
ictim
s
in
2
0
1
9
[
9
]
Acc
o
r
d
in
g
t
o
wo
r
ld
c
r
im
e
r
at
e
an
d
s
tatis
tics
2
0
0
0
-
2
0
2
4
,
th
e
in
cr
ea
s
e
in
2
0
2
0
was
5
.
6
1
p
er
1
0
0
,
0
0
0
p
o
p
u
latio
n
,
eq
u
iv
alen
t
t
o
a
r
is
e
o
f
0
.
7
4
%
in
a
n
n
u
al
ch
an
g
e
f
r
o
m
2
0
1
9
.
T
h
e
s
tatis
tics
in
2
0
1
9
we
r
e
5
.
5
6
p
er
1
0
0
,
0
0
0
p
o
p
u
latio
n
,
eq
u
al
to
a
3
.
6
5
%
d
ec
lin
e
in
y
ea
r
ly
ch
a
n
g
e
f
r
o
m
2
0
1
8
.
T
h
e
s
tatis
tics
th
at
ex
is
ted
in
2
0
1
8
wer
e
5
.
7
7
p
er
1
0
0
,
0
0
0
p
o
p
u
lat
io
n
,
eq
u
iv
alen
t
to
a
2
.
2
4
%
d
ec
lin
e
f
r
o
m
2
0
1
7
.
T
h
e
s
tatis
tics
t
h
at
ex
is
ted
in
2
0
1
7
wer
e
5
.
9
1
p
er
1
0
0
,
0
0
0
p
o
p
u
lat
io
n
,
eq
u
al
to
a
0
.
6
9
% d
ec
lin
e
f
r
o
m
2
0
1
6
,
as sh
o
wn
in
Fig
u
r
e
3
[
1
0
]
.
Pre
v
io
u
s
r
esear
ch
u
s
ed
d
ee
p
lear
n
in
g
tech
n
iq
u
es
an
d
co
m
p
u
ter
v
is
io
n
to
r
ec
o
n
s
tr
u
ct
t
h
e
f
ac
e
o
f
a
h
u
m
an
f
r
o
m
a
s
k
u
ll,
s
u
ch
as
s
u
ch
as
o
b
ject
d
etec
tio
n
,
r
ec
o
g
n
itio
n
,
im
ag
e
s
eg
m
e
n
tatio
n
,
m
u
lti
-
m
o
d
al
im
ag
e
tr
an
s
latio
n
,
an
d
3
D
r
ec
o
n
s
tr
u
c
tio
n
[
1
1
]
–
[
1
8
]
.
T
h
is
r
esear
c
h
p
r
o
p
o
s
es
a
co
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
s
k
u
ll
p
a
r
t
with
s
if
t
(
C
NNSP
S)
m
o
d
el
to
r
ec
o
n
s
tr
u
ct
h
u
m
an
f
ac
es
f
r
o
m
h
u
m
a
n
r
em
ai
n
s
o
f
u
n
id
e
n
tifie
d
s
k
u
lls
.
T
h
e
s
am
e
p
er
s
o
n
'
s
s
k
u
ll
an
d
s
k
in
f
ac
e
d
atasets
h
av
e
an
o
b
s
tacle
d
u
e
to
d
atasets
b
ein
g
u
n
av
ailab
l
e.
T
o
o
v
er
co
m
e
th
is
is
s
u
e,
we
u
s
ed
two
p
u
b
lic
d
atasets
MU
G5
0
0
+
[
1
9
]
an
d
C
eleb
AM
ask
HQ
[
2
0
]
.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
ap
p
lies
s
o
m
e
p
r
ep
r
o
ce
s
s
in
g
o
n
th
e
s
k
u
ll
d
ataset
to
en
h
an
ce
th
e
s
k
u
ll
im
ag
e.
T
h
en
,
th
e
f
ac
e
d
ataset
an
d
s
k
u
ll
d
ataset
ar
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell,
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
3
8
5
8
-
3
8
6
8
3860
s
eg
m
en
ted
in
to
p
ar
ts
to
b
e
r
ea
d
y
f
o
r
th
e
C
NNSPS
m
o
d
el.
T
h
is
m
o
d
el
is
u
s
ed
to
m
ap
ea
ch
p
ar
t
o
f
th
e
s
k
u
ll
to
b
e
eq
u
iv
ale
n
t
to
th
e
f
ac
e
p
ar
t
.
T
h
e
r
em
ai
n
in
g
s
ec
tio
n
s
o
f
t
h
e
r
esear
ch
p
ap
er
a
r
e
o
r
g
an
iz
ed
in
th
e
f
o
llo
win
g
m
an
n
er
:
s
ec
tio
n
2
r
elate
d
wo
r
k
,
s
ec
tio
n
3
m
eth
o
d
o
lo
g
y
,
s
ec
tio
n
4
ex
p
e
r
im
en
tal
a
n
d
r
e
s
u
lt,
an
d
s
ec
tio
n
5
co
n
clu
s
io
n
.
Fig
u
r
e
3
.
W
o
r
ld
cr
im
e
r
ate
a
n
d
s
tatis
tic
s
2
0
0
0
-
2
0
2
3
[
1
0
]
2.
RE
L
AT
E
D
WO
RK
Sh
o
r
t o
v
er
v
iew
o
f
s
o
m
e
p
r
ev
i
o
u
s
s
tu
d
ies o
f
f
ac
ial
ap
p
r
o
x
im
atio
n
u
s
in
g
s
o
m
e
tech
n
iq
u
es in
co
m
p
u
ter
v
is
io
n
an
d
d
ee
p
lea
r
n
in
g
to
r
ec
o
n
s
tr
u
ct
th
e
f
ac
e.
Kn
y
az
et
a
l.
[
1
2
]
p
r
esen
ted
d
ee
p
lear
n
in
g
tech
n
iq
u
es
s
h
o
wca
s
e
cu
ttin
g
-
ed
g
e
o
u
tc
o
m
es
f
r
o
m
c
o
n
tem
p
o
r
ar
y
d
ee
p
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
m
o
d
els
ac
r
o
s
s
d
iv
er
s
e
co
m
p
u
ter
v
is
io
n
task
s
in
clu
d
i
n
g
o
b
ject
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
,
im
a
g
e
s
eg
m
e
n
tatio
n
,
m
u
lti
-
m
o
d
al
im
a
g
e
tr
an
s
latio
n
,
an
d
3
D
r
ec
o
n
s
tr
u
c
tio
n
.
T
h
e
g
e
n
er
ativ
e
ad
v
e
r
s
ar
ial
ap
p
r
o
ac
h
in
d
ee
p
lear
n
in
g
e
n
ab
les
ex
ce
p
tio
n
al
p
er
f
o
r
m
an
ce
in
task
s
s
u
ch
as
im
ag
e
-
to
-
im
a
g
e
tr
an
s
latio
n
,
im
ag
e
s
eg
m
en
tatio
n
,
an
d
t
h
e
3
D
r
ec
o
n
s
tr
u
cti
o
n
s
tr
u
ctu
r
es
f
r
o
m
a
s
in
g
le
p
h
o
to
.
T
h
is
p
ap
er
p
r
esen
ted
th
e
s
k
u
ll2
f
ac
e
m
o
d
el
u
s
in
g
g
e
n
er
ato
r
G
an
d
d
is
cr
im
in
ato
r
D
o
f
th
e
g
an
m
o
d
el.
T
h
e
ac
c
u
r
a
cy
o
f
th
e
f
ac
ial
a
p
p
r
o
x
im
atio
n
was
ass
ess
ed
b
y
m
ea
s
u
r
in
g
t
h
e
s
u
r
f
ac
e
-
to
-
s
u
r
f
ac
e
d
ev
iatio
n
b
etwe
en
th
e
r
ec
o
n
s
t
r
u
cted
an
d
t
h
e
f
ac
es
r
ea
l.
T
h
e
r
esu
lts
in
d
icate
an
ac
cu
r
ac
y
o
f
u
n
d
er
2
.
5
m
m
f
o
r
ap
p
r
o
x
im
ately
6
0
%
o
f
th
e
r
ec
o
n
s
tr
u
cted
f
ac
ial
s
u
r
f
ac
e
[
1
2
]
.
Kn
iaz
et
a
l.
[
1
3
]
p
r
o
p
o
s
ed
a
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
is
u
tili
ze
d
to
cr
ea
te
an
d
s
h
o
wca
s
e
r
ea
lis
tic
tex
tu
r
es
f
o
r
th
e
3
D
r
ec
o
n
s
tr
u
c
tio
n
o
f
f
ac
es
b
elo
n
g
in
g
to
p
r
e
h
is
to
r
ic
h
u
m
an
s
.
Gen
er
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
(
GAN)
is
a
n
ew
n
eu
r
al
n
etwo
r
k
ty
p
e.
GANs
ar
e
co
m
p
r
is
ed
o
f
two
d
ee
p
C
NN
:
d
is
cr
im
in
ato
r
an
d
g
en
e
r
ato
r
.
A
C
y
cleG
AN
is
u
s
ed
to
o
v
er
c
o
m
e
th
e
d
if
f
icu
lty
o
f
Ga
n
'
s
im
ag
e
-
to
-
im
ag
e
tr
an
s
latio
n
task
s
.
A
n
ew
m
o
d
el
o
f
Sty
leGAN
is
im
p
lem
en
ted
to
e
n
h
an
ce
th
e
p
er
ce
p
tu
al
r
ea
lis
m
an
d
q
u
ality
o
f
th
e
r
ec
o
n
s
tr
u
cte
d
im
ag
e
to
im
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
.
T
h
e
is
s
u
es
p
r
esen
t
in
th
e
in
itial
Sty
leGAN
m
o
d
el
wer
e
p
a
r
tially
r
eso
lv
e
d
in
t
h
e
s
u
b
s
eq
u
e
n
t
Sty
leGAN
v
2
m
o
d
el.
Fo
r
tr
ai
n
in
g
th
e
d
e
v
elo
p
ed
s
k
u
ll2
p
h
o
to
f
r
am
ewo
r
k
,
a
d
is
tin
ct
C
eleb
AM
ask
-
HQ
an
d
cr
an
ia
-
to
-
f
ac
ial
d
ataset
(
C
2
F)
wer
e
s
p
ec
if
ically
g
en
e
r
ated
.
T
h
e
o
u
tc
o
m
es
o
f
th
e
ex
p
e
r
im
en
ts
r
ev
ea
l
th
at
th
e
en
h
an
ce
d
g
en
er
at
o
r
G1
c
o
n
tr
i
b
u
tes to
an
1
1
% im
p
r
o
v
em
en
t
in
th
e
q
u
ality
o
f
3
D
r
ec
o
n
s
tr
u
c
tio
n
[
1
3
]
.
B
u
s
h
r
a
an
d
Ma
h
eswar
i
[
1
4
]
p
r
esen
ted
an
in
n
o
v
ativ
e
ar
tific
ial
in
tellig
en
ce
m
eth
o
d
to
d
ig
itall
y
id
en
tify
a
cr
im
in
al
t
h
r
o
u
g
h
f
ac
ial
r
ec
o
g
n
itio
n
d
ig
itally
,
t
h
r
o
u
g
h
f
o
r
e
n
s
ic
s
k
etch
is
tr
an
s
f
o
r
m
ed
in
to
a
r
ea
l
p
h
o
to
u
s
in
g
a
d
ee
p
co
n
v
o
lu
tio
n
al
g
e
n
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
(
DC
G
AN)
.
Ad
v
an
c
ed
GAN
in
v
o
l
v
es
im
ag
e
p
r
o
ce
s
s
in
g
th
r
o
u
g
h
a
C
NN
k
n
o
wn
as
th
e
DC
GAN.
E
x
ten
d
ed
d
if
f
er
en
ce
-
of
-
g
au
s
s
ian
s
(
XDo
G)
alg
o
r
ith
m
s
eg
r
e
g
ates
s
k
etch
es
f
r
o
m
th
e
clip
p
ed
r
ea
l
p
ictu
r
es.
T
h
e
d
atasets
em
p
lo
y
ed
in
clu
d
e
th
e
ce
leb
r
ity
f
ac
e
a
ttrib
u
tes
(
C
eleb
A)
.
T
h
e
cy
cle
-
g
en
e
r
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
(
C
y
cleG
AN)
is
tr
ain
ed
o
n
a
r
ea
l
f
ac
ial
r
ec
o
g
n
itio
n
tech
n
o
lo
g
y
(
FERET
)
d
ataset,
co
n
s
is
tin
g
o
f
1
,
1
9
4
s
k
etch
-
p
h
o
t
o
p
air
s
o
f
h
a
n
d
-
d
r
aw
n
s
k
etch
im
ag
es.
C
o
n
d
itio
n
al
C
y
cleG
A
N
is
ass
o
ciate
d
with
C
y
cleG
A
N,
ac
h
iev
in
g
a
n
ac
cu
r
a
cy
o
f
6
1
.
3
4
% f
o
r
C
y
cleG
AN
an
d
6
5
.
5
3
% f
o
r
cc
GAN
[
1
4
]
.
Sh
u
i
et
a
l.
[
1
5
]
p
r
esen
te
d
m
a
n
u
al
2
D
an
d
3
D
ap
p
r
o
ac
h
es f
o
r
f
o
r
e
n
s
ic
p
u
r
p
o
s
es
an
d
a
r
ch
ae
o
lo
g
y
.
T
h
e
r
esu
lt
o
f
th
e
co
m
p
u
ter
ized
co
d
e
o
f
f
ed
er
al
r
eg
u
latio
n
s
(
C
F
R
)
is
co
n
tin
g
en
t
o
n
th
e
ex
ce
l
len
ce
o
f
th
e
s
k
u
ll
d
ig
italized
,
with
th
e
ac
q
u
is
itio
n
o
f
a
d
ig
itized
r
ep
r
esen
tati
o
n
ac
h
ie
v
ed
t
h
r
o
u
g
h
th
e
u
tili
za
tio
n
o
f
co
m
p
u
ted
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2
2
5
2
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
fo
r
fo
r
en
s
ic
fa
cia
l reco
n
s
tr
u
ctio
n
.
.
.
(
Do
a
a
M.
Mo
h
a
mme
d
)
3861
to
m
o
g
r
a
p
h
y
(
C
T
)
s
ca
n
n
in
g
a
n
d
laser
s
ca
n
n
in
g
tech
n
iq
u
es.
I
n
a
p
r
e
v
io
u
s
s
tu
d
y
th
e
g
eo
m
etr
ic
d
ev
iatio
n
o
f
m
o
d
els
o
b
tain
ed
th
r
o
u
g
h
laser
s
ca
n
n
in
g
an
d
C
T
s
ca
n
n
in
g
tech
n
iq
u
es
was
co
m
p
ar
ed
,
in
d
icatin
g
th
at
b
o
t
h
ap
p
r
o
ac
h
es
ca
n
g
en
er
ate
a
d
ig
ital
s
k
u
ll
o
f
h
i
g
h
q
u
al
ity
.
C
lass
if
ier
s
o
r
d
is
cr
im
in
an
t
e
q
u
atio
n
s
ar
e
p
r
o
v
id
e
d
f
o
r
s
ex
p
r
ed
ictio
n
,
in
ex
is
tin
g
r
esear
ch
h
as
ad
o
p
ted
th
is
to
r
ea
lize
s
ex
class
if
icatio
n
.
Su
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVM)
ar
e
em
p
lo
y
e
d
to
b
u
ild
a
class
if
ier
f
o
r
p
r
ed
ictin
g
s
ex
b
ased
o
n
PC
s
co
r
es
an
d
ce
n
t
r
o
i
d
s
ize.
T
h
e
r
esear
ch
en
lis
ted
1
4
0
liv
in
g
in
d
i
v
id
u
al
s
,
co
m
p
r
is
in
g
7
0
f
em
ales
an
d
7
0
m
ales
o
f
Ha
n
C
h
in
ese
n
ati
o
n
ality
to
estab
lis
h
th
e
d
ataset
o
f
th
e
f
ac
e
an
d
s
k
u
ll d
ataset
[
1
5
]
.
T
an
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
th
e
r
e
co
g
n
itio
n
o
f
h
u
m
an
r
em
ain
s
th
r
o
u
g
h
cr
a
n
io
f
ac
ial
s
u
p
er
im
p
o
s
itio
n
(
C
S)
as
a
n
o
tab
le
f
o
ca
l
p
o
in
t
w
ith
in
th
e
f
ield
o
f
f
o
r
en
s
ic
s
cien
ce
s
.
T
h
e
r
esear
ch
in
tr
o
d
u
ce
d
a
(
3
D
-
3
D)
s
u
p
er
im
p
o
s
itio
n
tech
n
i
q
u
e
f
o
r
th
e
m
o
d
el
o
f
a
3
D
s
k
u
ll
an
d
3
D
f
ac
e
to
o
v
e
r
c
o
m
e
th
e
c
h
allen
g
es
m
en
tio
n
ed
ea
r
lier
.
T
h
e
a
n
aly
tical
cu
r
v
atu
r
e
B
-
s
p
lin
e
(
AC
B
-
s
p
lin
e)
b
eg
in
s
b
y
r
ec
o
n
s
tr
u
ctin
g
a
3
D
f
ac
e
m
o
d
el
f
r
o
m
a
p
r
o
v
id
e
d
2
D
f
ac
e
im
ag
e
,
b
y
u
s
in
g
th
e
m
ea
n
s
im
p
lifie
d
g
en
e
r
i
c
elastic m
o
d
el.
T
h
e
f
ac
e
m
o
d
el
is
alig
n
ed
with
a
3
D
s
k
u
ll
o
n
th
e
jawlin
e
th
r
o
u
g
h
th
e
ap
p
licatio
n
o
f
th
e
AC
B
-
s
p
lin
e.
T
h
e
z
-
co
o
r
d
i
n
ates
o
f
th
e
2
D
f
ac
e
im
ag
e
ar
e
o
b
tain
ed
b
y
r
ec
o
n
s
tr
u
ctin
g
it
in
to
a
3
D
f
ac
e
m
o
d
el
u
s
in
g
a
m
ea
n
-
s
im
p
lifie
d
g
e
n
er
ic
elastic
m
o
d
el
(
MS
-
GE
M)
.
MS
-
GE
M
is
an
en
h
an
ce
d
v
er
s
io
n
o
f
S
-
GE
M,
am
alg
am
atin
g
th
e
f
ac
e
m
o
d
els
cr
ea
t
ed
th
r
o
u
g
h
S
-
GE
M
to
au
g
m
en
t
f
ac
ial
f
ea
t
u
r
es
m
o
r
e
ef
f
ec
tiv
el
y
.
t
h
e
AC
B
-
s
p
lin
e
ac
h
iev
es
a
n
ac
c
u
r
ac
y
is
0
.
8
0
3
an
d
th
e
PC
A
-
T
R
m
eth
o
d
e
x
h
ib
its
an
ac
cu
r
ac
y
o
f
8
3
.
5
%
[
1
6
]
.
Af
za
l
et
a
l.
[
1
7
]
p
r
o
p
o
s
ed
a
m
eth
o
d
f
o
r
r
ec
o
n
s
tr
u
ctin
g
3
D
f
a
ce
s
.
I
t
tak
es
an
in
p
u
t
o
f
a
s
in
g
l
e
2
D
im
a
g
e
an
d
y
ield
s
3
D
im
ag
es
r
ec
o
n
s
tr
u
cted
as
th
e
o
u
tp
u
t.
T
h
e
b
r
u
te
f
o
r
ce
m
atch
in
g
(
B
FM)
in
v
o
lv
es
th
r
ee
p
r
im
ar
y
s
tag
es:
ex
tr
ac
tin
g
f
ea
tu
r
es,
d
eter
m
in
in
g
d
e
p
th
,
an
d
g
en
er
atin
g
a
3
D
im
ag
e
f
r
o
m
th
e
p
r
o
ce
s
s
ed
im
ag
e
th
r
o
u
g
h
th
e
B
FM.
Su
b
s
eq
u
en
tly
,
s
ca
le
-
in
v
ar
ian
t
f
ea
tu
r
e
tr
a
n
s
f
o
r
m
(
SIFT
)
was
em
p
lo
y
ed
i
n
th
e
r
e
co
n
s
tr
u
ctio
n
o
f
3
D
f
ac
es
[
1
7
]
.
Nav
ic
et
a
l.
[
1
8
]
p
r
o
p
o
s
ed
r
ep
r
o
d
u
cin
g
a
v
ir
tu
al
f
ac
e,
with
f
ac
ial
s
o
f
t
ti
s
s
u
e
th
ick
n
ess
id
en
tifie
d
as
a
k
ey
cr
iter
io
n
f
o
r
en
s
u
r
i
n
g
r
eliab
ilit
y
an
d
ac
c
u
r
ac
y
i
n
3
D
c
o
m
p
u
ter
ized
f
ac
ial
r
ec
o
n
s
tr
u
ctio
n
.
th
e
p
r
im
a
r
y
t
o
o
ls
in
3
D
m
o
d
elin
g
s
o
f
twar
e
an
d
co
m
p
u
ter
an
im
atio
n
b
len
d
e
r
,
a
u
to
d
esk
,
an
d
m
a
y
a
wer
e
em
p
l
o
y
ed
to
r
ec
o
n
s
tr
u
ct
th
e
3
D
v
ir
tu
al
f
ac
e.
Facial
r
ec
o
n
s
tr
u
ctio
n
in
3
D
co
m
p
u
ter
iza
tio
n
ty
p
ically
r
elies o
n
a
s
em
i
-
au
to
m
ated
m
eth
o
d
,
s
u
ch
as
3
D
m
o
d
elin
g
s
o
f
twar
es
,
s
u
ch
as
B
len
d
e
r
,
Z
b
r
u
s
h
,
Fre
ef
o
r
m
,
an
d
3
DStu
d
io
Ma
x
.
T
h
is
a
p
p
r
o
ac
h
s
till
in
co
r
p
o
r
ates
co
n
v
e
n
tio
n
al
m
a
n
u
al
tech
n
iq
u
es
a
n
d
r
e
q
u
ir
es
s
tr
o
n
g
u
n
d
er
s
tan
d
in
g
o
f
an
t
h
r
o
p
o
lo
g
y
an
d
an
ato
m
y
,
o
r
an
au
to
m
ated
ap
p
r
o
ac
h
u
s
ed
to
r
ec
o
n
s
tr
u
ctio
n
in
v
o
lv
es
ad
v
an
ce
d
c
o
m
p
u
te
r
-
b
ase
d
tech
n
iq
u
es
an
d
cr
an
io
f
ac
ial
d
ata
to
g
en
e
r
ate
f
ac
ial
r
ec
o
n
s
tr
u
ctio
n
[
1
8
]
.
3.
M
E
T
H
O
D
Face
r
ec
o
n
s
tr
u
ctio
n
is
u
s
ed
to
id
en
tify
th
e
in
d
iv
i
d
u
al
f
r
o
m
t
h
eir
s
k
u
ll.
Face
r
ec
o
n
s
tr
u
ctio
n
is
u
s
ed
i
n
f
o
r
en
s
ics,
ar
ch
ae
o
lo
g
y
,
an
d
a
n
th
r
o
p
o
l
o
g
y
.
W
e
p
r
o
p
o
s
e
a
tech
n
iq
u
e
o
f
a
C
NNSPS
m
o
d
el.
T
h
e
m
ain
p
r
o
c
ed
u
r
es
in
th
is
r
esear
ch
will
b
e
e
x
p
l
ain
ed
,
s
u
ch
as
p
r
e
p
r
o
ce
s
s
in
g
,
cr
o
p
p
in
g
p
h
o
to
g
r
ap
h
s
,
a
n
d
th
e
m
o
d
el
lay
e
r
s
em
p
lo
y
ed
,
f
o
llo
wed
b
y
t
h
e
o
u
t
p
u
t im
ag
e.
T
h
e
m
ain
p
r
o
ce
d
u
r
es a
r
e
s
h
o
wn
in
Fig
u
r
e
4
.
Fig
u
r
e
4
.
T
h
e
p
r
o
p
o
s
ed
C
NNSPS
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell,
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
3
8
5
8
-
3
8
6
8
3862
3
.
1
.
P
re
pro
ce
s
s
ing
T
h
is
s
tep
co
n
v
er
ts
th
e
in
p
u
t
im
ag
e
to
g
r
ay
s
ca
le
an
d
ap
p
lies
s
o
m
e
n
ec
ess
ar
y
im
ag
e
en
h
an
ce
m
en
t,
s
u
ch
as
n
o
is
e
r
em
o
v
al
u
s
in
g
f
astNlMea
n
s
Den
o
is
in
g
C
o
lo
r
ed
in
O
p
en
C
V
.
f
astNlMea
n
s
Den
o
is
in
g
C
o
lo
r
ed
u
s
es
th
e
non
-
lo
ca
l
m
ea
n
s
d
en
o
is
in
g
alg
o
r
ith
m
.
I
t
co
m
p
a
r
es
p
ix
el
n
eig
h
b
o
r
h
o
o
d
s
in
th
e
in
p
u
t
im
ag
e
to
a
tem
p
late
n
eig
h
b
o
r
h
o
o
d
,
id
e
n
tify
in
g
an
d
r
ed
u
cin
g
g
r
a
y
s
ca
le
an
d
co
l
o
r
n
o
is
e.
T
h
e
s
eg
m
e
n
tatio
n
te
ch
n
iq
u
e
is
o
lates
th
e
s
k
u
ll
r
eg
io
n
f
r
o
m
th
e
b
ac
k
g
r
o
u
n
d
u
s
in
g
th
e
Gr
a
b
C
u
t
alg
o
r
ith
m
th
at
ap
p
lies
to
iter
at
iv
ely
u
p
d
ate
th
e
m
ask
b
ased
o
n
co
lo
r
an
d
s
p
atial
in
f
o
r
m
at
io
n
.
Featu
r
e
ex
tr
a
ctio
n
is
u
s
ed
to
id
en
tify
lan
d
m
a
r
k
s
o
n
t
h
e
s
k
u
ll
b
o
n
e
af
ter
d
etec
tin
g
th
e
s
k
u
ll r
e
g
io
n
u
s
in
g
th
e
SIFT
.
T
h
e
p
r
o
m
in
e
n
t
lan
d
m
ar
k
s
d
et
ec
ted
o
n
th
e
s
k
u
ll
im
ag
e.
T
h
e
MU
G5
0
0
+
d
ataset
[
1
9
]
f
o
r
t
h
e
s
k
u
ll
is
u
s
ed
in
th
e
p
r
ep
r
o
ce
s
s
in
g
s
te
p
.
ar
e:
th
e
an
ter
io
r
n
asal
s
p
in
e
(
An
tNS),
Po
i
n
t
A(
A
)
,
lef
t
s
u
p
r
a
-
o
r
b
ital
in
cisu
r
a
(
L
s
u
p
O)
,
lef
t
in
f
e
r
io
r
later
al
o
r
b
ital
r
im
(
L
Or
b
i
)
,
lef
t
g
o
n
io
n
(
L
Go
n
i)
,
lef
t
p
o
r
i
o
n
(
L
Po
r
i)
,
r
ig
h
t
s
u
p
r
a
-
o
r
b
ital
in
cisu
r
a
(
R
s
u
p
O)
,
r
ig
h
t
in
f
er
i
o
r
later
al
o
r
b
ital
r
im
(
R
Or
b
i)
,
r
ig
h
t
g
o
n
io
n
(
R
Go
n
i)
,
r
ig
h
t
p
o
r
io
n
(
R
Po
r
i)
,
an
d
Me
n
to
n
(
Me
n
t
o
)
as sh
o
wn
in
Fig
u
r
e
5
.
Fig
u
r
e
5
.
T
h
e
p
r
o
m
in
en
t la
n
d
m
ar
k
s
o
n
th
e
s
k
u
ll im
ag
e
3
.
2
.
F
a
ce
a
nd
s
k
ull
s
eg
m
ent
a
t
io
n
T
h
is
s
tag
e
in
v
o
lv
es
s
eg
m
en
tin
g
th
e
s
k
u
ll
im
ag
e
in
to
p
ar
ts
lik
e
th
e
r
ig
h
t
an
d
lef
t
ey
es,
r
ig
h
t
an
d
lef
t
ch
ee
k
,
a
n
d
b
r
o
w.
T
h
ese
s
eg
m
en
ted
p
a
r
ts
ar
e
th
e
n
f
ed
i
n
to
t
h
e
C
NNSPS
m
o
d
el
f
o
r
f
u
r
th
e
r
p
r
o
ce
s
s
in
g
.
Dete
ct
th
e
f
ac
e
lan
d
m
ar
k
s
o
f
th
e
C
ele
b
AM
ask
-
Hq
d
ataset
in
[
2
0
]
co
n
s
is
ts
o
f
3
0
,
0
0
0
f
ac
e
im
ag
es o
f
f
am
o
u
s
p
eo
p
le
to
d
etec
t
th
e
6
8
p
o
in
ts
o
f
lan
d
m
a
r
k
s
,
as
s
h
o
wn
in
Fig
u
r
e
6
,
th
at
ar
e
u
s
ed
to
cr
o
p
th
e
f
ac
e
i
n
to
1
0
p
ar
ts
,
in
clu
d
in
g
r
ig
h
t
a
n
d
lef
t
ey
es,
r
ig
h
t
an
d
l
ef
t
ch
ee
k
,
a
n
d
b
r
o
w,
as
s
h
o
w
n
in
Fig
u
r
e
s
7
(
a)
an
d
7
(
b
)
.
T
h
e
n
,
tr
ain
i
n
g
m
o
d
el
to
g
et
th
e
m
o
s
t p
ar
t
m
atch
es with
th
e
p
ar
t o
f
th
e
f
ac
e
with
th
e
s
k
u
ll.
Fig
u
r
e
6
.
L
a
n
d
m
ar
k
s
o
f
th
e
f
a
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2
2
5
2
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
fo
r
fo
r
en
s
ic
fa
cia
l reco
n
s
tr
u
ctio
n
.
.
.
(
Do
a
a
M.
Mo
h
a
mme
d
)
3863
(
a)
(
b
)
Fig
u
r
e
7
.
T
h
e
s
eg
m
en
tatio
n
o
f
th
e
s
k
u
lls
an
d
f
ac
es (
a
)
th
e
s
e
g
m
en
ted
p
ar
ts
o
f
th
e
f
ac
e
an
d
(
b
)
th
e
s
eg
m
e
n
ted
p
ar
ts
o
f
th
e
s
k
u
ll
3
.
3
.
C
o
nv
o
lutio
n neura
l net
wo
rk
s
k
ull
pa
rt
wit
h si
f
t
mo
del
Use
th
e
o
u
tp
u
t
o
f
th
e
s
eg
m
en
tatio
n
o
f
th
e
f
ac
e
an
d
s
k
u
ll
as
in
p
u
t
to
th
e
m
o
d
el.
Dete
ct
th
e
p
r
im
ar
y
s
k
u
ll
lan
d
m
ar
k
s
o
f
th
e
im
a
g
e.
C
NN
(
also
C
o
n
v
Net)
is
a
d
ee
p
lear
n
in
g
-
b
ased
n
eu
r
al
n
etwo
r
k
with
an
in
h
er
e
n
t
m
ac
h
in
e
lear
n
i
n
g
tec
h
n
iq
u
e
.
C
NN
p
r
o
ce
s
s
es
d
ata,
s
u
ch
as
im
ag
es,
with
a
g
r
i
d
p
atter
n
.
T
h
e
m
o
d
el
ar
c
h
itectu
r
e
co
m
p
r
is
es
f
iv
e
co
n
v
o
lu
tio
n
al
lay
er
s
with
a
k
er
n
el
s
ize
o
f
3
×3
,
s
u
cc
ee
d
e
d
b
y
b
atch
n
o
r
m
aliza
tio
n
,
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
ac
tiv
atio
n
,
an
d
m
ax
p
o
o
lin
g
o
p
er
atio
n
s
.
T
h
e
f
ir
s
t
co
n
v
o
lu
tio
n
al
lay
er
co
m
p
r
is
es
th
r
ee
in
p
u
t
ch
an
n
els
(
wh
ich
r
ep
r
esen
t
R
GB
co
lo
r
s
)
an
d
1
6
o
u
t
p
u
t
ch
a
n
n
els.
T
h
e
s
u
b
s
eq
u
e
n
t
lay
er
s
f
o
llo
w
a
s
im
ilar
p
atter
n
,
with
3
2
,
6
4
,
1
2
8
,
an
d
2
2
4
o
u
tp
u
t
ch
an
n
els.
B
atch
n
o
r
m
al
izatio
n
is
u
s
ed
af
ter
ea
ch
co
n
v
o
lu
tio
n
al
lay
er
to
in
cr
ea
s
e
co
n
v
er
g
e
n
ce
a
n
d
tr
ai
n
in
g
s
tab
ilit
y
.
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
s
ad
d
n
o
n
lin
ea
r
ity
to
th
e
m
o
d
el,
h
elp
i
n
g
it
d
etec
t
co
m
p
licated
d
ata
p
att
er
n
s
.
Ma
x
-
p
o
o
lin
g
o
p
er
atio
n
s
m
in
im
ize
s
p
atial
d
im
e
n
s
io
n
s
,
allo
win
g
th
e
m
o
d
el
to
f
o
cu
s
o
n
ess
en
tial f
ea
tu
r
es wh
ile
r
em
o
v
in
g
d
u
p
licate
d
ata
as sh
o
wn
in
Fig
u
r
e
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell,
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
3
8
5
8
-
3
8
6
8
3864
Fig
u
r
e
8
.
C
NNSPS
ar
ch
itectu
r
e
m
o
d
el
3
.
3
.
1
.
Co
nv
o
lutio
n la
y
er
A
co
n
v
o
lu
tio
n
lay
e
r
is
a
k
ey
co
m
p
o
n
en
t
o
f
th
e
C
NN
an
d
is
r
esp
o
n
s
ib
le
f
o
r
ex
tr
ac
tin
g
f
ea
tu
r
es.
I
t
is
u
s
ed
to
ex
tr
ac
t
ess
en
tial
in
f
o
r
m
atio
n
.
T
h
e
co
n
v
o
lu
tio
n
al
lay
er
co
m
p
u
tes
th
e
co
n
v
o
lu
tio
n
al
o
p
er
atio
n
o
f
th
e
in
p
u
t
p
ictu
r
es
u
s
in
g
f
ilter
s
.
T
h
e
k
e
r
n
el
f
ilter
s
h
a
v
e
th
e
s
am
e
d
im
e
n
s
io
n
as
th
e
in
p
u
t
im
a
g
e
b
u
t
h
av
e
lo
wer
co
n
s
tan
t
p
ar
am
eter
s
[
1
6
]
.
T
h
e
co
n
v
o
lu
t
io
n
lay
er
is
ca
lcu
lated
in
(
1
)
[
2
1
]
:
w
out
=(
−
+
2
∗
)
+1
(
1
)
wh
er
e
wi
in
is
an
in
p
u
t size,
k
s
is
k
er
n
el
s
ize,
p
d
is
p
ad
d
in
g
,
a
n
d
s
t is st
r
id
e.
3
.
3
.
2
.
P
o
o
lin
g
la
y
er
A
p
o
o
lin
g
la
y
er
c
o
n
d
u
cts
a
s
t
an
d
ar
d
d
o
wn
s
am
p
lin
g
o
p
e
r
atio
n
,
d
ec
r
ea
s
in
g
th
e
f
ea
tu
r
e
m
a
p
s
'
s
p
atial
d
im
en
s
io
n
ality
.
T
h
is
is
d
o
n
e
t
o
in
tr
o
d
u
ce
tr
an
s
latio
n
i
n
v
ar
ia
n
ce
,
m
ak
in
g
th
e
n
etwo
r
k
less
s
en
s
itiv
e
to
s
lig
h
t
s
h
if
ts
an
d
d
is
to
r
tio
n
s
an
d
d
ec
r
ea
s
in
g
th
e
n
u
m
b
er
o
f
lear
n
ab
l
e
p
ar
am
eter
s
.
Ma
x
p
o
o
lin
g
is
th
e
m
o
s
t
c
o
m
m
o
n
p
r
o
ce
d
u
r
e,
tak
in
g
p
atch
es
f
r
o
m
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
s
.
I
t
o
u
tp
u
ts
th
e
m
o
s
t
s
ig
n
if
ican
t
v
alu
e
in
ea
ch
p
atch
an
d
d
is
ca
r
d
s
all
o
th
er
v
alu
es [
1
6
]
.
T
h
e
eq
u
atio
n
f
o
r
m
ax
p
o
o
lin
g
is
ca
lcu
lated
in
(
2
)
[
2
2
]
an
d
(
3
)
[
2
3
]
:
(
)=
(
2
)
r
elate
d
f
ilter
s
ar
e
u
s
ed
f
o
r
t
h
e
co
m
p
o
s
itio
n
o
f
m
th
m
a
x
p
o
o
l
in
g
b
an
d
=[
1,
,
….
,
,
,
….
,
,
]
:
,
= m
a
x(
ℎ
,(
−
1)
+
)
(
3
)
W
h
er
e
{
1
,
…,
}
is
ter
m
e
d
as
a
p
o
o
lin
g
s
h
if
t
th
at
allo
ws
e
n
ab
lin
g
o
v
er
la
p
with
in
c
o
n
ce
r
n
in
g
p
o
o
lin
g
r
e
g
io
n
s
wh
en
<
.
3
.
3
.
3
.
F
ull
y
co
nn
ec
t
ed
la
y
er
T
h
e
o
u
tp
u
t
f
ea
tu
r
e
m
ap
s
f
r
o
m
th
e
last
co
n
v
o
l
u
tio
n
o
r
p
o
o
li
n
g
lay
er
ar
e
u
s
u
ally
f
latten
ed
.
T
h
e
f
u
lly
co
n
n
ec
ted
lay
er
is
co
n
v
e
r
ted
in
to
a
o
n
e
-
d
im
e
n
s
io
n
al
(
1
D)
ar
r
ay
o
r
v
ec
to
r
o
f
v
alu
es.
T
h
en
,
it
is
lin
k
ed
to
o
n
e
o
r
m
o
r
e
f
u
lly
co
n
n
ec
ted
lay
er
s
,
o
f
ten
ca
lled
d
en
s
e
lay
er
s
.
I
n
th
e
s
e
d
en
s
e
lay
e
r
s
,
ev
e
r
y
in
p
u
t
is
co
n
n
ec
ted
to
ev
er
y
o
u
tp
u
t
th
r
o
u
g
h
ad
ju
s
tab
le
weig
h
ts
.
Af
ter
u
s
in
g
C
NN,
th
e
SIFT
alg
o
r
ith
m
d
etec
ts
d
is
tin
ctiv
e
k
ey
p
o
in
ts
o
r
f
ea
tu
r
es in
an
im
a
g
e
[
2
4
]
.
3
.
3
.
4
.
Sca
le
inv
a
ria
nt
f
e
a
t
ure
t
ra
ns
f
o
rm
T
h
e
SIFT
is
u
tili
ze
d
t
o
ex
t
r
ac
t
f
ea
tu
r
es
th
at
r
ed
u
ce
s
v
is
u
al
i
n
f
o
r
m
atio
n
to
a
s
et
o
f
p
o
in
ts
t
h
at
ca
n
b
e
u
s
ed
to
f
in
d
s
im
ilar
p
atter
n
s
in
o
th
er
im
ag
es.
T
h
is
ap
p
r
o
ac
h
i
s
co
m
m
o
n
ly
u
s
ed
in
co
m
p
u
ter
v
is
io
n
ap
p
licatio
n
s
lik
e
p
ictu
r
e
m
atch
in
g
a
n
d
o
b
ject
d
etec
tio
n
.
SIFT
h
as
m
an
y
ad
v
an
tag
es,
s
u
ch
as
lo
ca
lity
,
d
is
tin
ctiv
en
ess
,
q
u
an
tity
,
ef
f
icien
cy
,
a
n
d
e
x
ten
s
ib
len
ess
[
2
5
]
.
T
h
e
s
tep
s
o
f
SIFT
ex
ec
u
tio
n
ar
e
s
h
o
wn
in
Fig
u
r
e
9
.
Fig
u
r
e
9
.
SIFT
alg
o
r
ith
m
s
tep
s
B
u
ild
in
g
t
h
e
sc
ale
-
spac
e
Key
p
oi
n
t
locali
z
at
i
on
Orien
t
a
t
io
n
ass
ign
men
t
K
ey
p
oint
d
escr
i
p
to
r
&
K
ey
p
oint
m
atc
h
i
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2
2
5
2
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
fo
r
fo
r
en
s
ic
fa
cia
l reco
n
s
tr
u
ctio
n
.
.
.
(
Do
a
a
M.
Mo
h
a
mme
d
)
3865
3
.
3
.
5
.
B
uil
din
g
t
he
s
ca
le
-
s
pa
ce
T
h
e
'
Gau
s
s
ian
B
lu
r
'
ap
p
r
o
ac
h
d
is
tin
g
u
is
h
es
u
n
iq
u
e
s
p
o
ts
wh
ile
d
is
r
eg
ar
d
in
g
n
o
is
e.
T
h
e
o
u
t
p
u
t
im
ag
e
o
f
s
ca
le
-
s
p
ac
e
is
a
f
u
n
ctio
n
L
(
x
,
y
,
σ
)
[
1
6
]
.
T
h
e
b
lu
r
r
ed
im
ag
e
is
ca
lcu
lated
in
(
4
)
[
2
1
]
,
[
2
5
]
:
L(x
,
y,
σ)
= I
(
x,
y)
*
G(
x,
y,
σ)
(
4
)
Gau
s
s
ian
b
lu
r
o
p
er
at
o
r
is
ca
lc
u
lated
f
r
o
m
(
5
)
[
2
1
]
,
[
2
6
]
:
G(
x
,
y
,
σ
)
=
1
2
2
−
(
2
+
2
)
/
2
2
(
5
)
W
h
er
e
x
an
d
y
ar
e
t
h
e
lo
ca
tio
n
co
o
r
d
i
n
ates,
σ
is
th
e
"scale"
p
ar
am
eter
,
an
d
G
is
th
e
Gau
s
s
ian
b
lu
r
o
p
er
at
o
r
.
T
h
en
,
th
e
d
if
f
er
e
n
ce
o
f
Gau
s
s
ian
s
(
Do
G)
is
u
s
ed
to
lo
ca
te
r
el
ev
an
t k
ey
p
o
in
ts
in
a
n
im
ag
e
[
2
1
]
,
[
2
7
]
.
D(
x,
y,
σ)
= −L(
x,
y,
σ)
+ L(
x,
y,
kσ
)
(
6
)
W
h
er
e
σ
,
k
σ
ar
e
th
e
s
ca
les o
f
i
m
ag
es.
3
.
3
.
6
.
K
ey
po
int
lo
c
a
liza
t
io
n
Key
p
o
in
t
lo
ca
lizatio
n
is
u
s
ed
to
f
in
d
th
e
k
ey
p
o
in
ts
in
th
e
im
ag
e.
T
h
e
lo
w
-
co
n
tr
ast
f
ea
tu
r
es
an
d
th
o
s
e
p
lace
d
o
n
th
e
e
d
g
es
ar
e
r
em
o
v
ed
.
T
h
e
tech
n
iq
u
e
d
eter
m
i
n
es
th
e
p
r
ec
is
e
p
o
s
itio
n
s
o
f
th
e
k
ey
p
o
in
ts
b
y
in
ter
p
o
latin
g
n
eig
h
b
o
r
d
ata
p
o
in
ts
f
o
r
ea
c
h
k
ey
p
o
i
n
t.
I
t
ca
n
b
e
ac
h
iev
e
d
b
y
e
m
p
lo
y
i
n
g
t
h
e
T
ay
lo
r
ex
p
an
s
io
n
o
f
th
e
Do
G
im
a
g
e
[
2
8
]
.
3
.
3
.
7
.
O
rient
a
t
io
n a
s
s
ig
nm
en
t
Or
ien
tatio
n
ass
ig
n
m
en
t
is
u
s
e
d
to
en
s
u
r
e
im
ag
e
r
o
tatio
n
i
n
v
ar
ian
ce
.
E
ac
h
k
ey
p
o
in
t
is
a
s
s
ig
n
ed
an
o
r
ien
tatio
n
.
Dep
e
n
d
in
g
o
n
th
e
s
ize,
a
n
eig
h
b
o
r
h
o
o
d
is
d
r
awn
ar
o
u
n
d
th
e
k
e
y
p
o
in
t
lo
ca
tio
n
,
an
d
th
e
g
r
ad
ien
t
m
ag
n
itu
d
e
a
n
d
d
ir
ec
tio
n
ar
e
d
eter
m
in
ed
th
er
e
[
2
2
]
,
ca
lcu
late
d
in
(
7
)
an
d
(
8
)
[
2
9
]
.
Ma
g
n
itu
d
e=
√
(
G
a
2
+G
b
2
)
(
7
)
Ori
en
ta
tio
n
=a
r
cta
n
(
G
b
/G
a
)
(
8
)
W
h
er
e
G
a
an
d
G
b
r
e
p
r
esen
t th
e
g
r
ad
ien
t in
th
e
a
a
n
d
b
d
ir
ec
ti
o
n
r
esp
ec
tiv
ely
f
o
r
t
h
e
im
ag
e.
3
.
3
.
8
.
K
ey
po
int
des
cr
ipto
r
a
nd
k
ey
po
int
m
a
t
ching
F
i
r
s
t
,
t
o
b
u
il
d
a
k
e
y
p
o
i
n
t
d
e
s
c
r
ip
t
o
r
,
c
o
m
p
u
t
e
t
h
e
g
r
a
d
i
e
n
t
m
a
g
n
i
t
u
d
e
a
n
d
d
i
r
e
c
ti
o
n
f
o
r
e
a
c
h
i
m
a
g
e
s
a
m
p
l
e
p
o
i
n
t
i
n
a
n
a
r
e
a
s
u
r
r
o
u
n
d
i
n
g
t
h
e
k
e
y
p
o
i
n
t
p
o
s
it
i
o
n
.
T
h
e
s
e
a
r
e
w
e
i
g
h
t
e
d
b
y
a
G
a
u
s
s
i
a
n
w
i
n
d
o
w
,
r
e
p
r
e
s
e
n
t
e
d
b
y
t
h
e
o
v
e
r
l
a
i
d
c
i
r
c
l
e
,
a
s
s
h
o
w
n
i
n
Fi
g
u
r
e
1
0
[
3
0
]
.
K
e
y
p
o
i
n
t
m
a
t
c
h
i
n
g
i
s
d
e
r
i
v
e
d
f
r
o
m
t
h
e
k
e
y
p
o
i
n
t
d
e
s
c
r
i
p
t
o
r
.
Fig
u
r
e
1
0
.
E
x
am
p
le
o
f
th
e
SIFT
d
escr
ip
to
r
[
2
6
]
3
.
4
.
S
k
ull
-
to
-
f
a
ce
ma
pp
ing
S
k
u
l
l
-
to
-
f
a
c
e
m
a
p
p
i
n
g
p
r
o
p
o
s
e
s
a
m
a
t
c
h
e
d
i
m
a
g
e
t
h
a
t
e
n
t
a
i
l
s
s
u
p
e
r
i
m
p
o
s
i
n
g
a
p
h
o
t
o
g
r
a
p
h
o
f
a
p
e
r
s
o
n
'
s
f
a
c
e
o
n
t
o
a
2
D
m
o
d
e
l
o
f
t
h
e
i
r
s
k
u
l
l
.
T
h
i
s
p
r
o
c
e
d
u
r
e
e
n
a
b
l
e
s
f
o
r
e
n
s
i
c
a
n
t
h
r
o
p
o
l
o
g
i
s
t
s
,
f
o
r
e
n
s
i
c
a
r
t
i
s
t
s
,
a
n
d
p
l
a
s
t
i
c
s
u
r
g
e
o
n
s
t
o
m
a
t
c
h
t
h
e
s
k
u
l
l
'
s
f
e
a
t
u
r
e
s
d
i
r
e
c
t
l
y
t
o
t
h
o
s
e
o
f
t
h
e
i
n
d
i
v
i
d
u
a
l
w
h
i
l
e
t
h
e
y
a
r
e
a
l
i
v
e
.
T
h
e
f
i
r
s
t
s
t
e
p
i
s
u
s
i
n
g
i
m
a
g
e
s
i
n
t
h
e
C
e
l
e
b
A
m
a
s
k
H
q
a
v
a
i
l
a
b
l
e
d
a
t
a
s
e
t
.
T
h
e
s
e
c
o
n
d
s
t
e
p
i
s
u
s
i
n
g
t
h
e
s
k
u
l
l
i
m
a
g
e
s
i
n
t
h
e
M
U
G
5
0
0
+
a
v
a
i
l
a
b
l
e
d
a
t
a
s
e
t
.
T
h
i
r
d
,
m
a
t
c
h
t
h
e
p
h
o
t
o
s
t
o
t
h
e
2
D
m
o
d
e
l
o
f
t
h
e
s
k
u
l
l
.
T
h
i
s
e
n
t
a
i
l
s
r
e
s
i
z
i
n
g
a
n
d
a
l
i
g
n
i
n
g
t
h
e
p
h
o
t
o
g
r
a
p
h
t
o
m
a
t
c
h
t
h
e
s
i
z
e
a
n
d
p
o
s
i
t
i
o
n
o
f
t
h
e
s
k
u
l
l
.
T
h
e
f
o
u
r
t
h
s
t
e
p
i
s
i
d
e
n
t
i
f
y
i
n
g
m
a
j
o
r
a
n
a
t
o
m
i
c
a
l
l
a
n
d
m
a
r
k
s
i
n
p
h
o
t
o
g
r
a
p
h
s
a
n
d
t
h
e
s
k
u
l
l
.
T
h
e
s
e
l
a
n
d
m
a
r
k
s
m
a
y
i
n
c
l
u
d
e
t
h
e
e
y
e
s
,
n
o
s
e
,
m
o
u
t
h
,
c
h
i
n
,
a
n
d
o
t
h
e
r
l
o
c
a
t
i
o
n
s
a
l
o
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell,
Vo
l.
1
3
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
3
8
5
8
-
3
8
6
8
3866
t
h
e
f
a
c
e
'
s
c
o
n
t
o
u
r
s
.
T
h
e
f
i
n
a
l
s
t
e
p
i
s
t
o
p
r
o
j
e
c
t
t
h
e
f
e
a
t
u
r
e
s
f
r
o
m
t
h
e
i
m
a
g
e
o
n
t
o
t
h
e
a
p
p
r
o
p
r
i
a
t
e
r
e
g
i
o
n
s
o
n
t
h
e
s
k
u
l
l
(
t
h
e
i
m
a
g
e
o
n
t
h
e
l
e
f
t
d
e
p
i
c
t
s
t
h
e
s
k
u
l
l
,
w
h
i
l
e
t
h
e
i
m
a
g
e
o
n
t
h
e
r
i
g
h
t
i
l
l
u
s
t
r
a
t
e
s
t
h
e
f
a
c
e
)
a
s
s
h
o
w
n
i
n
F
i
g
u
r
e
1
1
.
Fig
u
r
e
1
1
.
Sk
u
ll
-
to
-
f
a
ce
m
ap
p
in
g
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
W
e
t
r
ain
th
e
m
o
d
el
o
n
Go
o
g
le
C
o
lab
.
T
h
e
s
y
s
tem
'
s
co
m
p
u
ta
tio
n
al
b
ac
k
b
o
n
e
is
a
Go
o
g
le
C
o
lab
T
PU
r
u
n
tim
e.
Featu
r
in
g
an
I
n
tel
X
eo
n
C
PU,
1
3
GB
R
AM
,
an
d
a
p
o
wer
f
u
l
clo
u
d
T
PU.
I
ts
s
o
f
t
war
e
s
tack
in
clu
d
es
Py
th
o
n
3
.
1
0
.
W
e
u
s
ed
Op
en
C
V
f
o
r
ad
v
a
n
ce
d
co
m
p
u
ter
v
is
i
o
n
task
s
an
d
lev
er
ag
e
d
th
e
ca
p
ab
ilit
ies o
f
Py
T
o
r
ch
f
o
r
cu
ttin
g
-
ed
g
e
d
ee
p
lea
r
n
in
g
ap
p
licatio
n
s
.
4
.
1
.
Da
t
a
s
et
s
T
h
e
p
r
o
p
o
s
ed
r
esear
ch
was
te
s
ted
o
n
th
e
MU
G5
0
0
+
[
3
1
]
an
d
C
eleb
AM
ask
-
HQ
d
atasets
.
MU
G5
0
0
+
is
a
co
llectio
n
o
f
5
0
0
h
ea
lth
y
s
k
u
lls
,
th
at
ar
e
s
eg
m
en
ted
f
r
o
m
h
ig
h
-
r
eso
lu
tio
n
h
ea
d
CT
im
ag
es.
C
eleb
AM
ask
-
HQ
h
as
ar
o
u
n
d
3
0
,
0
0
0
im
ag
es
o
f
ce
leb
r
ity
f
ac
es.
T
o
b
u
ild
an
au
to
m
ated
d
iag
n
o
s
is
to
o
l,
a
h
u
g
e
d
ataset
en
co
m
p
ass
in
g
o
v
er
3
0
0
,
0
0
0
h
ea
d
CT
im
ag
es
wa
s
g
ath
er
ed
r
etr
o
s
p
ec
tiv
ely
f
r
o
m
ar
o
u
n
d
2
0
ce
n
ter
s
in
I
n
d
ia
b
etwe
en
J
an
u
ar
y
1
,
2
0
1
1
an
d
J
u
n
e
1
,
2
0
1
7
.
I
t
in
cl
u
d
es
th
e
tr
ain
in
g
an
d
v
alid
atio
n
s
u
b
s
ets
(
Qu
r
e2
5
k
an
d
C
Q5
0
0
d
atasets
,
r
esp
ec
tiv
ely
)
,
th
e
o
r
ig
in
al
clin
ical
r
ad
io
lo
g
y
r
ep
o
r
t,
an
d
t
h
e
co
n
s
e
n
s
u
s
o
f
th
r
ee
in
d
ep
en
d
en
t
r
ad
io
lo
g
is
ts
,
wh
ich
ar
e
r
eg
a
r
d
e
d
as
th
e
g
o
ld
s
tan
d
ar
d
f
o
r
th
e
Q
u
r
e2
5
k
a
n
d
C
Q
5
0
0
d
atasets
.
T
h
is
d
ataset
is
m
ea
n
t
f
o
r
th
e
d
ev
elo
p
m
en
t
an
d
ass
ess
m
en
t
o
f
au
t
o
m
ated
d
ia
g
n
o
s
tic
s
y
s
tem
s
,
n
o
t
f
o
r
f
ac
e
a
p
p
r
o
x
im
atio
n
lear
n
in
g
[
1
2
]
.
Un
f
o
r
tu
n
atel
y
,
th
e
p
u
b
licly
ac
ce
s
s
ib
le
CT
d
ataset
i
s
n
o
t
s
u
ita
b
le
f
o
r
th
e
jo
b
o
f
s
k
u
ll
-
b
ased
f
ac
e
ap
p
r
o
x
im
atio
n
.
T
h
e
MU
G5
0
0
+
[
1
9
]
an
d
C
eleb
AM
ask
-
HQ
[
2
0
]
d
atasets
wer
e
ch
o
s
en
d
u
e
to
th
e
d
atasets
ar
e
av
ailab
le
to
d
o
wn
lo
ad
as
p
u
b
lic
d
atasets
.
I
t
is
d
if
f
icu
lt
to
cr
ea
te
d
atasets
o
f
3
D
h
u
m
an
s
k
u
lls
an
d
3
D
h
u
m
an
f
ac
es
s
in
ce
a
f
o
r
e
n
s
ic
ex
p
e
r
t
is
r
e
q
u
ir
ed
t
o
b
u
ild
3
D
f
ac
e
d
atasets
an
d
3
D
s
k
u
ll
d
atasets
.
Pre
v
io
u
s
r
e
s
ea
r
ch
cr
ea
ted
3
-
D
s
k
u
ll
an
d
f
ac
e
d
atasets
u
s
in
g
3
D
-
m
o
d
elin
g
s
o
f
twar
e
s
u
ch
as
3
DStu
d
io
Ma
x
,
Fre
ef
o
r
m
,
B
len
d
er
,
Z
b
r
u
s
h
1
5
,
clay
,
o
r
g
y
p
s
u
m
s
cu
lp
tu
r
e
t
o
cr
ea
te
d
atasets
an
d
r
eq
u
ir
es
a
g
o
o
d
k
n
o
wled
g
e
o
f
an
ato
m
y
,
s
p
ec
if
ically
f
ac
ial
an
ato
m
y
.
I
n
th
is
r
esear
ch
,
we
attem
p
ted
t
o
u
s
e
r
ea
l
d
ata
f
r
o
m
u
n
k
n
o
wn
h
u
m
an
s
k
u
lls
a
n
d
p
u
b
lic
2
D
f
ac
es.
Ho
wev
er
,
we
r
an
in
t
o
an
o
b
s
tacle
d
u
e
to
3
D
s
k
u
lls
an
d
3
D
f
ac
es
n
o
t
b
ein
g
a
v
ailab
le.
T
h
e
r
ec
o
n
s
tr
u
cted
f
ac
es
wer
e
cr
ea
ted
b
y
s
eg
m
en
tin
g
th
e
s
k
u
ll im
ag
e
in
to
ten
p
ar
ts
an
d
m
atch
in
g
th
em
to
p
ar
ts
o
f
th
e
f
ac
e,
as sh
o
wn
in
Fig
u
r
e
1
2
(
a
)
.
W
e
u
p
d
ated
th
e
r
ec
o
n
s
tr
u
cted
f
ac
e
b
y
clu
s
ter
in
g
th
e
C
eleb
AM
ask
HQ
in
to
m
ale
an
d
f
em
ale.
T
h
e
r
ec
o
n
s
tr
u
cted
o
u
t
p
u
t
f
ac
e
s
eg
m
en
ted
th
e
s
k
u
ll
im
ag
e
in
to
1
8
p
ar
ts
an
d
u
s
ed
a
m
atch
in
g
m
o
d
el
with
th
e
s
am
e
f
ac
es
as th
e
r
ec
o
n
s
tr
u
cted
f
ac
e
,
as sh
o
wn
in
Fig
u
r
e
1
2
(
b
)
.
4
.
2
.
E
x
perim
ent
a
l det
a
ils
T
h
e
in
p
u
t
im
ag
e
s
ize
is
2
2
4
×2
2
4
×3
to
th
e
C
NNSPS
m
o
d
el,
wh
ich
em
p
lo
y
s
co
n
v
o
lu
tio
n
2
D
with
m
an
y
in
p
u
t
ch
a
n
n
els,
m
an
y
o
u
t
p
u
t
ch
an
n
els
with
a
k
er
n
el
s
ize
o
f
3
×3
th
at
m
o
v
es
o
n
e
p
ix
el
at
a
tim
e
in
b
o
t
h
th
e
h
o
r
izo
n
tal
an
d
v
e
r
tical
d
ir
ec
tio
n
s
b
y
1
s
tr
id
e
a
n
d
1
p
ix
el
ad
d
e
d
to
ea
ch
s
id
e
o
f
t
h
e
in
p
u
t
i
m
a
g
e
b
y
p
ad
d
in
g
.
T
h
e
m
o
d
el
in
d
icate
s
th
at
th
e
b
atc
h
n
o
r
m
aliza
tio
n
lay
er
will
n
o
r
m
alize
th
e
n
u
m
b
er
o
f
o
u
tp
u
t
ch
an
n
els
f
r
o
m
th
e
p
r
ev
io
u
s
co
n
v
o
lu
tio
n
al
lay
er
.
T
h
e
m
o
d
el
u
s
es
b
atch
n
o
r
m
ali
za
tio
n
to
im
p
r
o
v
e
n
eu
r
al
n
etw
o
r
k
t
r
ain
in
g
s
tab
ilit
y
an
d
p
er
f
o
r
m
a
n
ce
.
I
t
n
o
r
m
alize
s
a
g
iv
en
lay
e
r
'
s
ac
tiv
atio
n
s
b
y
s
u
b
tr
ac
tin
g
th
e
m
ea
n
an
d
d
iv
i
d
in
g
b
y
th
e
s
tan
d
ar
d
d
ev
iatio
n
d
eter
m
in
e
d
ac
r
o
s
s
a
m
in
i
-
b
atch
o
f
s
am
p
les.
T
h
is
h
elp
s
to
r
ed
u
ce
in
ter
n
al
co
v
ar
iate
s
h
if
ts
an
d
ca
n
s
p
ee
d
u
p
n
etwo
r
k
co
n
v
er
g
e
n
c
e
d
u
r
in
g
tr
ain
in
g
.
Ma
x
p
o
o
lin
g
is
a
d
o
wn
-
s
am
p
lin
g
p
r
o
ce
d
u
r
e
u
s
ed
to
m
in
im
ize
th
e
s
p
atial
d
im
en
s
io
n
s
o
f
th
e
in
p
u
t
wh
ile
m
ain
tain
in
g
th
e
m
o
s
t
im
p
o
r
tan
t
ch
ar
ac
ter
is
tics
.
I
t
s
p
lits
th
e
in
p
u
t
in
to
non
-
o
v
er
lap
p
in
g
r
ec
tan
g
u
lar
r
eg
io
n
s
an
d
o
u
tp
u
ts
th
e
g
r
ea
test
v
alu
e
f
r
o
m
ea
ch
.
T
h
e
m
ax
p
o
o
lin
g
lay
e
r
u
s
es
a
2
×2
k
e
r
n
el
an
d
a
s
tr
id
e
o
f
2
.
T
h
e
m
o
d
el
u
s
es
a
lin
ea
r
tr
a
n
s
f
o
r
m
atio
n
to
th
e
in
p
u
t
d
ata
with
p
ar
am
eter
s
2
2
4
×
7
×7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2
2
5
2
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
fo
r
fo
r
en
s
ic
fa
cia
l reco
n
s
tr
u
ctio
n
.
.
.
(
Do
a
a
M.
Mo
h
a
mme
d
)
3867
(
a)
(
b
)
Fig
u
r
e
1
2
.
Face
r
ec
o
n
s
tr
u
ctio
n
(
a)
th
e
r
ec
o
n
s
tr
u
cted
f
ac
e
o
f
1
0
p
ar
ts
an
d
(
b
)
t
h
e
r
ec
o
n
s
tr
u
ct
ed
f
ac
e
o
f
1
8
p
ar
ts
5.
CO
NCLU
SI
O
N
An
th
r
o
p
o
lo
g
y
,
ar
c
h
ae
o
lo
g
y
,
a
n
d
f
o
r
en
s
ics
h
av
e
is
s
u
es
with
f
ac
e
a
p
p
r
o
x
im
atio
n
b
ased
o
n
th
e
s
k
u
ll
-
co
m
p
u
ter
-
aid
ed
f
ac
ial
r
ec
o
n
s
tr
u
ctio
n
tech
n
iq
u
es
b
ased
o
n
d
ig
itized
s
k
u
ll
im
ag
es
o
f
u
n
k
n
o
wn
p
eo
p
le.
C
o
m
p
u
ter
-
aid
ed
f
o
r
e
n
s
ics
p
r
ed
icted
th
e
v
ictim
'
s
n
ea
r
est
f
a
ce
b
y
n
o
t
em
p
l
o
y
i
n
g
s
tan
d
a
r
d
p
r
o
ce
d
u
r
es.
T
h
is
r
esear
ch
u
s
es
th
e
C
NNSP
S
m
o
d
el,
b
ased
o
n
b
r
ea
k
in
g
u
n
id
e
n
t
if
ied
s
k
u
lls
in
to
p
ar
ts
an
d
d
iv
id
in
g
k
n
o
wn
h
u
m
a
n
im
ag
es
in
to
p
ar
ts
to
p
r
ed
ict
f
a
ce
s
.
T
h
is
r
esear
ch
m
ak
es
u
s
e
o
f
p
u
b
licly
av
ailab
le
d
atasets
s
u
ch
as
MU
G5
0
0
+
an
d
C
eleb
AM
ask
-
HQ.
T
h
is
r
esear
ch
illu
s
tr
ates
th
e
an
ticip
at
ed
o
u
tco
m
e
o
f
u
n
i
d
en
tifie
d
s
k
u
lls
u
s
in
g
a
d
ataset
o
f
ce
leb
r
ity
p
er
s
o
n
s
.
Ou
r
f
u
n
d
am
en
tal
is
s
u
e
is
th
at
th
e
co
lle
ctio
n
o
f
k
n
o
wn
g
en
u
in
e
s
k
u
lls
an
d
k
n
o
w
n
p
e
o
p
le
d
o
es n
o
t e
x
is
t.
T
h
u
s
,
we
r
ely
o
n
p
u
b
licly
av
ailab
le
d
a
tasets
.
T
h
e
ch
allen
g
e
f
ac
in
g
u
s
in
th
is
r
esear
ch
is
th
at
t
h
e
r
ea
l d
ataset
is
n
o
t a
v
ailab
le,
an
d
it is
d
if
f
icu
lt to
h
a
v
e
it.
RE
F
E
R
E
NC
E
S
[
1
]
E.
A
.
A
.
El
-
D
i
n
,
“
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
i
n
f
o
r
e
n
s
i
c
sc
i
e
n
c
e
:
i
n
v
a
s
i
o
n
o
r
r
e
v
o
l
u
t
i
o
n
?
,
”
E
g
y
p
t
i
a
n
S
o
c
i
e
t
y
o
f
C
l
i
n
i
c
a
l
T
o
x
i
c
o
l
o
g
y
J
o
u
rn
a
l
,
v
o
l
.
1
0
,
n
o
.
2
,
p
p
.
2
0
–
3
2
,
2
0
2
2
,
d
o
i
:
1
0
.
2
1
6
0
8
/
e
sct
j
.
2
0
2
2
.
1
5
8
1
7
8
.
1
0
1
2
.
[
2
]
C
.
W
i
l
k
i
n
s
o
n
,
“
C
o
mp
u
t
e
r
i
z
e
d
f
o
r
e
n
s
i
c
f
a
c
i
a
l
r
e
c
o
n
s
t
r
u
c
t
i
o
n
:
a
r
e
v
i
e
w
o
f
c
u
r
r
e
n
t
sy
s
t
e
ms
,
”
F
o
re
n
si
c
S
c
i
e
n
c
e
,
M
e
d
i
c
i
n
e
,
a
n
d
Pa
t
h
o
l
o
g
y
,
v
o
l
.
1
,
n
o
.
3
,
p
p
.
1
7
3
–
1
7
7
,
2
0
0
5
,
d
o
i
:
1
0
.
1
3
8
5
/
F
S
M
P
:
1
:
3
:
1
7
3
.
[
3
]
S
.
N
.
B
y
e
r
s
a
n
d
C
.
A
.
J
u
a
r
e
z
,
I
n
t
r
o
d
u
c
t
i
o
n
t
o
f
o
r
e
n
s
i
c
a
n
t
h
ro
p
o
l
o
g
y
:
S
i
x
t
h
e
d
i
t
i
o
n
.
R
o
u
t
l
e
d
g
e
,
2
0
2
3
,
d
o
i
:
1
0
.
4
3
2
4
/
9
7
8
1
0
0
3
2
8
3
9
3
5
.
[
4
]
C
.
N
.
S
t
e
p
h
a
n
,
“
F
a
c
i
a
l
a
p
p
r
o
x
i
m
a
t
i
o
n
-
f
r
o
m
f
a
c
i
a
l
r
e
c
o
n
st
r
u
c
t
i
o
n
s
y
n
o
n
y
m
t
o
f
a
c
e
p
r
e
d
i
c
t
i
o
n
p
a
r
a
d
i
g
m,
”
J
o
u
rn
a
l
o
f
F
o
re
n
s
i
c
S
c
i
e
n
c
e
s
,
v
o
l
.
6
0
,
n
o
.
3
,
p
p
.
5
6
6
–
5
7
1
,
2
0
1
5
,
d
o
i
:
1
0
.
1
1
1
1
/
1
5
5
6
-
4
0
2
9
.
1
2
7
3
2
.
[
5
]
S
.
G
u
p
t
a
,
V
.
G
u
p
t
a
,
H
.
V
i
j
,
R
.
V
i
j
,
a
n
d
N
.
T
y
a
g
i
,
“
F
o
r
e
n
si
c
f
a
c
i
a
l
r
e
c
o
n
st
r
u
c
t
i
o
n
:
t
h
e
f
i
n
a
l
f
r
o
n
t
i
e
r
,
”
J
o
u
rn
a
l
o
f
C
l
i
n
i
c
a
l
a
n
d
D
i
a
g
n
o
s
t
i
c
Re
s
e
a
rc
h
,
v
o
l
.
9
,
n
o
.
9
,
p
p
.
2
6
–
2
8
,
2
0
1
5
,
d
o
i
:
1
0
.
7
8
6
0
/
J
C
D
R
/
2
0
1
5
/
1
4
6
2
1
.
6
5
6
8
.
[
6
]
M
.
D.
B
u
h
a
n
a
n
d
C
.
N
a
r
d
o
n
i
,
“
A
f
a
c
i
a
l
r
e
c
o
n
s
t
r
u
c
t
i
o
n
me
t
h
o
d
b
a
se
d
o
n
n
e
w
mesh
d
e
f
o
r
ma
t
i
o
n
t
e
c
h
n
i
q
u
e
s
,
”
F
o
re
n
si
c
S
c
i
e
n
c
e
s
Re
se
a
rc
h
,
v
o
l
.
3
,
n
o
.
3
,
p
p
.
2
5
6
–
2
7
3
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
8
0
/
2
0
9
6
1
7
9
0
.
2
0
1
8
.
1
4
6
9
1
8
5
.
[
7
]
P
.
M
e
se
j
o
,
R
.
M
a
r
t
o
s,
Ó
.
I
b
á
ñ
e
z
,
J.
N
o
v
o
,
a
n
d
M
.
O
r
t
e
g
a
,
“
A
s
u
r
v
e
y
o
n
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
t
e
c
h
n
i
q
u
e
s
f
o
r
b
i
o
me
d
i
c
a
l
i
ma
g
e
a
n
a
l
y
si
s
i
n
s
k
e
l
e
t
o
n
-
b
a
se
d
f
o
r
e
n
si
c
h
u
man
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
0
,
n
o
.
1
4
,
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
0
1
4
4
7
0
3
.
[
8
]
“
C
r
i
m
e
r
a
t
e
b
y
c
o
u
n
t
r
y
2
0
2
4
,
”
Wo
r
l
d
P
o
p
u
l
a
t
i
o
n
Re
v
i
e
w
,
2
0
2
4
.
A
c
c
e
sse
d
:
A
u
g
.
2
0
,
2
0
2
3
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
o
r
l
d
p
o
p
u
l
a
t
i
o
n
r
e
v
i
e
w
.
c
o
m
/
c
o
u
n
t
r
y
-
r
a
n
k
i
n
g
s
/
c
r
i
me
-
r
a
t
e
-
by
-
c
o
u
n
t
r
y
[
9
]
U
N
O
D
C
,
“
G
l
o
b
a
l
s
t
u
d
y
o
n
h
o
mi
c
i
d
e
:
U
n
i
t
e
d
N
a
t
i
o
n
s
o
f
f
i
c
e
o
n
d
r
u
g
s
a
n
d
c
r
i
me,
”
U
n
i
t
e
d
N
a
t
i
o
n
s
p
u
b
l
i
c
a
t
i
o
n
,
S
a
l
e
s
N
o
.
1
4
.
I
V
.
1
,
2
0
1
9
.
A
c
c
e
sse
d
:
A
u
g
.
2
0
,
2
0
2
3
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
w
w
w
.
u
n
o
d
c
.
o
r
g
/
u
n
o
d
c
/
e
n
/
d
a
t
a
-
a
n
d
-
a
n
a
l
y
s
i
s
/
g
l
o
b
a
l
-
st
u
d
y
-
on
-
h
o
m
i
c
i
d
e
-
2
0
1
9
.
h
t
ml
[
1
0
]
“
W
o
r
l
d
C
r
i
m
e
R
a
t
e
&
S
t
a
t
i
s
t
i
c
s
2
0
0
0
-
2
0
2
3
,
”
Ma
c
r
o
T
re
n
d
s
,
2
0
2
3
.
A
c
c
e
ss
e
d
:
A
u
g
.
2
5
,
2
0
2
3
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
ma
c
r
o
t
r
e
n
d
s.
n
e
t
/
c
o
u
n
t
r
i
e
s/
W
LD
/
w
o
r
l
d
/
c
r
i
me
-
r
a
t
e
-
s
t
a
t
i
st
i
c
s
[
1
1
]
H
.
Q
.
N
g
u
y
e
n
,
T.
N
.
N
g
u
y
e
n
,
V
.
D
.
Tr
a
n
,
a
n
d
T
.
T.
D
a
o
,
“
A
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
f
o
r
p
r
e
d
i
c
t
i
n
g
s
u
b
j
e
c
t
-
s
p
e
c
i
f
i
c
h
u
ma
n
s
k
u
l
l
sh
a
p
e
f
r
o
m
h
e
a
d
t
o
w
a
r
d
a
d
e
c
i
s
i
o
n
su
p
p
o
r
t
s
y
st
e
m
f
o
r
h
o
me
-
b
a
se
d
f
a
c
i
a
l
r
e
h
a
b
i
l
i
t
a
t
i
o
n
,
”
I
RB
M
,
v
o
l
.
4
4
,
n
o
.
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
r
b
m.
2
0
2
2
.
0
5
.
0
0
5
.
[
1
2
]
V
.
A
.
K
n
y
a
z
,
V
.
V
.
K
n
i
a
z
,
M
.
M
.
N
o
v
i
k
o
v
,
a
n
d
R
.
M
.
G
a
l
e
e
v
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
f
o
r
a
p
p
r
o
x
i
m
a
t
i
n
g
u
n
k
n
o
w
n
f
a
c
e
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
Arc
h
i
v
e
s
o
f
t
h
e
P
h
o
t
o
g
ra
m
m
e
t
ry
,
Re
m
o
t
e
S
e
n
si
n
g
a
n
d
S
p
a
t
i
a
l
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
-
I
S
PR
S
Ar
c
h
i
v
e
s
,
v
o
l
.
4
3
,
n
o
.
B
2
,
p
p
.
8
5
7
–
8
6
2
,
2
0
2
0
,
d
o
i
:
1
0
.
5
1
9
4
/
i
s
p
r
s
-
a
r
c
h
i
v
e
s
-
X
L
I
I
I
-
B2
-
2
0
2
0
-
857
-
2
0
2
0
.
[
1
3
]
V
.
K
n
i
a
z
,
V
.
K
n
y
a
z
,
a
n
d
V
.
M
i
z
g
i
n
o
v
,
“
S
y
n
t
h
e
si
s
a
n
d
v
i
su
a
l
i
z
a
t
i
o
n
o
f
p
h
o
t
o
r
e
a
l
i
st
i
c
t
e
x
t
u
r
e
s
f
o
r
3
d
f
a
c
e
r
e
c
o
n
st
r
u
c
t
i
o
n
o
f
p
r
e
h
i
s
t
o
r
i
c
h
u
ma
n
,
”
C
EU
R
Wo
r
k
sh
o
p
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
2
7
4
4
,
p
p
.
1
-
9
,
2
0
2
0
,
d
o
i
:
1
0
.
5
1
1
3
0
/
g
r
a
p
h
i
c
o
n
-
2
0
2
0
-
2
-
4
-
1
5
.
[
1
4
]
S
.
N
.
B
u
s
h
r
a
a
n
d
K
.
U
.
M
a
h
e
sw
a
r
i
,
“
C
r
i
me
i
n
v
e
st
i
g
a
t
i
o
n
u
si
n
g
D
C
G
A
N
b
y
f
o
r
e
n
si
c
s
k
e
t
c
h
-
to
-
f
a
c
e
t
r
a
n
sf
o
r
mat
i
o
n
(
S
TF)
-
a
r
e
v
i
e
w
,
”
i
n
5
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
i
n
g
Me
t
h
o
d
o
l
o
g
i
e
s
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
,
I
C
C
MC
2
0
2
1
,
I
EEE,
A
p
r
.
2
0
2
1
,
p
p
.
1
3
4
3
–
1
3
4
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
M
C
5
1
0
1
9
.
2
0
2
1
.
9
4
1
8
4
1
7
.
[
1
5
]
W
.
S
h
u
i
e
t
a
l
.
,
“
A
c
o
mp
u
t
e
r
i
z
e
d
c
r
a
n
i
o
f
a
c
i
a
l
r
e
c
o
n
st
r
u
c
t
i
o
n
met
h
o
d
f
o
r
a
n
u
n
i
d
e
n
t
i
f
i
e
d
sk
u
l
l
b
a
se
d
o
n
s
t
a
t
i
st
i
c
a
l
sh
a
p
e
mo
d
e
l
s,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
9
,
n
o
.
3
5
–
3
6
,
p
p
.
2
5
5
8
9
–
2
5
6
1
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
0
-
0
9
1
8
9
-
7.
[
1
6
]
J.
S
.
T
a
n
,
I
.
Y
.
L
i
a
o
,
I
.
V
e
n
k
a
t
,
B
.
B
e
l
a
t
o
n
,
a
n
d
P
.
T.
Ja
y
a
p
r
a
k
a
s
h
,
“
C
o
m
p
u
t
e
r
-
a
i
d
e
d
su
p
e
r
i
m
p
o
s
i
t
i
o
n
v
i
a
r
e
c
o
n
s
t
r
u
c
t
i
n
g
a
n
d
m
a
t
c
h
i
n
g
3
D
f
a
c
e
s
t
o
3
D
sk
u
l
l
s
f
o
r
f
o
r
e
n
s
i
c
c
r
a
n
i
o
f
a
c
i
a
l
i
d
e
n
t
i
f
i
c
a
t
i
o
n
s
,
”
Vi
s
u
a
l
C
o
m
p
u
t
e
r
,
v
o
l
.
3
6
,
n
o
.
9
,
p
p
.
1
7
3
9
–
1
7
5
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
3
7
1
-
0
1
9
-
0
1
7
6
7
-
7.
[
1
7
]
H
.
M
.
R
.
A
f
z
a
l
,
S
.
Lu
o
,
M
.
K
.
A
f
z
a
l
,
G
.
C
h
a
u
d
h
a
r
y
,
M
.
K
h
a
r
i
,
a
n
d
S
.
A
.
P
.
K
u
mar,
“
3
D
f
a
c
e
r
e
c
o
n
st
r
u
c
t
i
o
n
f
r
o
m
si
n
g
l
e
2
D
i
ma
g
e
u
si
n
g
d
i
s
t
i
n
c
t
i
v
e
f
e
a
t
u
r
e
s
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
8
0
6
8
1
–
1
8
0
6
8
9
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
3
0
2
8
1
0
6
.
[
1
8
]
P
.
N
a
v
i
c
,
P
.
P
a
l
e
e
,
S
.
P
r
a
p
a
y
a
sa
t
o
k
,
S
.
P
r
a
si
t
w
a
t
t
a
n
a
sere
e
,
A
.
S
i
n
t
h
u
b
u
a
,
a
n
d
P
.
M
a
h
a
k
k
a
n
u
k
r
a
u
h
,
“
T
h
e
d
e
v
e
l
o
p
me
n
t
a
n
d
t
e
s
t
i
n
g
o
f
T
h
a
i
f
a
c
i
a
l
so
f
t
t
i
ss
u
e
t
h
i
c
k
n
e
ss
d
a
t
a
i
n
t
h
r
e
e
-
d
i
me
n
si
o
n
a
l
c
o
m
p
u
t
e
r
i
z
e
d
f
o
r
e
n
s
i
c
f
a
c
i
a
l
r
e
c
o
n
s
t
r
u
c
t
i
o
n
,
”
M
e
d
i
c
i
n
e
,
S
c
i
e
n
c
e
a
n
d
Evaluation Warning : The document was created with Spire.PDF for Python.