T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
18
,
No
.
4
,
A
u
g
u
s
t
2020
,
p
p
.
2
1
6
9
~
2
1
7
6
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Ke
m
e
n
r
is
te
k
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
: 1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
KA
.
v
1
8
i4
.
1
4
8
3
6
2169
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Se
m
i
-
superv
ised
a
ut
o
-
enco
der
for
facia
l attribu
tes
r
ecog
nition
So
u
m
a
y
a
Z
a
g
hb
a
ni
1
,
No
ured
ine
B
o
ujn
e
h
2
,
M
ed
Sa
li
m
B
o
uh
lel
3
1
Na
ti
o
n
a
l
E
n
g
in
e
e
rin
g
S
c
h
o
o
l
o
f
G
a
b
è
s
,
Un
iv
e
rsit
y
o
f
G
a
b
è
s
,
T
u
n
isia
2
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s o
f
G
a
b
è
s
,
Un
iv
e
rsity
o
f
G
a
b
è
s
,
T
u
n
isia
3
Hig
h
In
stit
u
te o
f
Bio
tec
h
n
o
l
o
g
y
,
Un
iv
e
rsit
y
o
f
S
fa
x
,
T
u
n
isia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
6
,
2019
R
ev
i
s
ed
Ma
r
1
9
,
2
0
2
0
A
cc
ep
ted
A
p
r
7
,
2
0
2
0
T
h
e
p
a
rti
c
u
larity
o
f
o
u
r
f
a
c
e
s
e
n
c
o
u
ra
g
e
s
m
a
n
y
re
se
a
rc
h
e
rs
to
e
x
p
lo
it
th
e
ir
f
e
a
tu
re
s
in
d
if
f
e
re
n
t
d
o
m
a
in
s
su
c
h
a
s
u
se
r
id
e
n
ti
f
ica
ti
o
n
,
b
e
h
a
v
io
u
r
a
n
a
ly
sis,
c
o
m
p
u
ter
tec
h
n
o
lo
g
y
,
se
c
u
rit
y
,
a
n
d
p
sy
c
h
o
lo
g
y
.
In
th
is
p
a
p
e
r,
w
e
p
re
se
n
t
a
m
e
th
o
d
f
o
r
f
a
c
ial
a
tt
rib
u
tes
a
n
a
ly
sis.
T
h
e
w
o
rk
a
d
d
re
ss
e
d
to
a
n
a
ly
se
f
a
c
ia
l
im
a
g
e
s
a
n
d
e
x
trac
t
f
e
a
tu
re
s
in
th
e
p
u
r
p
o
se
to
re
c
o
g
n
ize
d
e
m
o
g
ra
p
h
ic
a
tt
rib
u
tes
:
a
g
e
,
g
e
n
d
e
r,
a
n
d
e
t
h
n
icity
(AG
E)
.
In
t
h
is
w
o
rk
,
we
e
x
p
lo
it
e
d
th
e
r
o
b
u
st
n
e
ss
o
f
d
e
e
p
lea
rn
in
g
(DL
)
u
sin
g
a
n
u
p
d
a
ti
n
g
v
e
rsio
n
o
f
a
u
to
e
n
c
o
d
e
rs
c
a
ll
e
d
th
e
d
e
e
p
sp
a
rse
a
u
to
e
n
c
o
d
e
r
(DSA
E).
In
th
is
w
o
rk
we
u
se
d
a
n
e
w
a
rc
h
it
e
c
tu
re
o
f
DSA
E
b
y
a
d
d
in
g
th
e
su
p
e
rv
i
sio
n
to
t
h
e
c
las
sic
m
o
d
e
l
a
n
d
w
e
c
o
n
tro
l
th
e
o
v
e
rf
it
ti
n
g
p
ro
b
lem
b
y
re
g
u
lariz
in
g
th
e
m
o
d
e
l.
T
h
e
p
a
ss
f
r
o
m
DS
A
E
to
t
h
e
se
m
i
-
su
p
e
rv
ise
d
a
u
to
e
n
c
o
d
e
r
(DSS
A
E)
f
a
c
il
it
a
tes
th
e
su
p
e
rv
isio
n
p
ro
c
e
ss
a
n
d
a
c
h
iev
e
s
a
n
e
x
c
e
ll
e
n
t
p
e
rf
o
rm
a
n
c
e
to
e
x
trac
t
f
e
a
tu
r
e
s.
In
th
is
w
o
rk
we
f
o
c
u
se
d
to
e
stim
a
te
AG
E
jo
in
tl
y
.
T
h
e
e
x
p
e
rim
e
n
t
re
su
lt
s
sh
o
w
th
a
t
DSS
A
E
is
c
re
a
ted
to
re
c
o
g
n
ize
f
a
c
ial
f
e
a
tu
re
s
w
i
th
h
ig
h
p
re
c
isio
n
.
T
h
e
w
h
o
le
s
y
ste
m
a
c
h
iev
e
s
g
o
o
d
p
e
rf
o
rm
a
n
c
e
a
n
d
i
m
p
o
rtan
t
ra
t
e
s
in
AG
E
u
sin
g
th
e
M
OR
P
H II
d
a
tab
a
se
K
ey
w
o
r
d
s
:
Ag
e
esti
m
at
io
n
Dee
p
lear
n
in
g
Gen
d
er
r
ec
o
g
n
itio
n
So
f
t
m
ax
cla
s
s
i
f
ier
Su
p
er
v
i
s
ed
au
to
en
co
d
er
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
:
So
u
m
a
y
a
Z
a
g
h
b
a
n
i
,
Natio
n
al
E
n
g
i
n
ee
r
in
g
Sc
h
o
o
l o
f
Gab
ès
,
Un
i
v
er
s
it
y
o
f
Gab
ès,
T
u
n
i
s
ia.
E
m
ail:
s
o
u
m
a
y
a.
za
g
h
h
b
an
i@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
R
ec
en
t
l
y
w
i
th
t
h
e
g
r
o
w
t
h
an
d
t
h
e
d
ev
elo
p
m
e
n
t
o
f
tec
h
n
o
lo
g
i
es,
in
tell
ig
e
n
t
an
d
r
ec
o
m
m
e
n
d
ed
s
y
s
te
m
s
b
ec
am
e
in
th
e
ce
n
ter
o
f
r
esea
r
ch
es.
W
e
ar
e
in
th
e
ep
o
ch
wh
en
s
y
s
te
m
s
m
u
s
t
k
n
o
w
a
n
d
u
n
d
er
s
ta
n
d
h
is
/
h
er
cu
r
r
en
t
u
s
er
.
No
w
ad
a
y
s
u
s
er
i
d
en
tific
atio
n
is
a
n
ess
e
n
ti
al
ch
allen
g
e
f
o
r
r
esear
ch
er
s
;
it
p
r
esen
ts
a
n
ec
e
s
s
it
y
i
n
n
e
w
tec
h
n
o
lo
g
ies
i
n
m
an
y
d
o
m
ai
n
s
s
u
ch
as
s
ec
u
r
it
y
,
co
m
p
u
ter
ac
ce
s
s
co
n
tr
o
l,
e
-
co
m
m
er
ce
,
b
an
k
in
g
,
h
u
m
a
n
m
ac
h
in
e
i
n
ter
ac
tio
n
(
HM
I
)
,
m
ed
icin
e,
s
o
cial
m
ed
ia,
ap
p
lican
t
id
en
tif
ica
tio
n
,
civ
i
l
p
r
o
tectio
n
,
c
r
i
m
es,
ter
r
o
r
is
m
,
an
d
m
o
s
t r
ec
en
tl
y
t
h
e
f
i
g
h
t
a
g
ain
s
t
s
o
cial
f
r
a
u
d
,
etc.
[
1
]
.
I
n
HM
I
,
th
e
r
esp
o
n
s
e
f
o
r
th
e
q
u
esti
o
n
“w
h
o
is
t
h
e
u
s
er
?
”
is
v
er
y
co
m
p
licated
an
d
n
ee
d
a
n
ex
ce
lle
n
t
p
r
ec
is
io
n
b
ec
au
s
e
it
is
d
if
f
er
e
n
t
f
r
o
m
an
ar
ea
to
an
o
th
er
.
G
en
er
all
y
,
in
u
s
er
a
n
al
y
s
i
s
in
cl
u
d
in
g
u
s
er
p
r
o
f
ilin
g
an
d
u
s
er
m
o
d
eli
n
g
,
w
e
n
ee
d
to
k
n
o
w
th
e
u
s
er
’
ag
e,
g
e
n
d
e
r
,
r
ac
e,
em
o
tio
n
,
ac
tu
al
b
eh
a
v
io
r
,
cu
ltu
r
al
le
v
el,
s
en
s
o
r
y
ab
ilit
ie
s
,
a
n
d
h
i
s
/
h
er
e
x
p
er
ien
ce
s
.
I
t
is
a
b
ig
an
d
m
u
lt
id
is
cip
lin
ar
y
c
h
alle
n
g
e
b
u
t
v
er
y
i
m
p
o
r
tan
t
b
ec
a
u
s
e
we
n
ee
d
to
m
a
k
e
m
ac
h
i
n
es
ab
le
to
u
n
d
er
s
ta
n
d
,
an
al
y
ze
t
h
e
u
s
er
’
s
n
ee
d
s
an
d
al
s
o
to
ad
ap
t
its
el
f
to
h
i
s
n
ee
d
s
an
d
ca
p
ab
ilit
ies
[
2
]
.
T
h
e
ch
all
en
g
e
o
f
t
h
ese
n
e
w
tech
n
o
lo
g
i
es
is
to
i
n
cr
ea
s
e
th
e
e
f
f
ec
ti
v
e
n
es
s
an
d
r
o
b
u
s
t
n
e
s
s
an
d
g
iv
e
t
h
e
p
r
ec
is
e,
r
ig
h
t
an
d
ex
ac
t
r
esp
o
n
s
e
in
t
h
e
r
ig
h
t m
o
m
en
t
ev
en
w
er
e
th
e
co
n
d
itio
n
s
.
T
o
k
n
o
w
t
h
i
s
u
s
er
an
d
u
n
d
er
s
tan
d
h
i
s
/h
er
s
en
s
o
r
ial
ca
p
ab
ilit
ies,
p
h
y
s
ica
l
ab
ilit
i
es,
af
f
ec
ti
v
e
s
tate,
s
o
cial
a
n
d
c
u
lt
u
r
al
le
v
el,
m
a
n
y
r
esear
ch
er
s
u
s
ed
f
ac
e
s
as
a
n
i
m
p
o
r
ta
n
t
s
t
i
m
u
l
u
s
f
o
r
th
eir
wo
r
k
s
d
u
e
to
th
e
p
ar
tic
u
lar
it
y
an
d
d
is
ti
n
cti
v
en
e
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
18
,
No
.
4
,
A
u
g
u
s
t 2
0
2
0
:
2
1
6
9
-
2
1
7
6
2170
f
ea
t
u
r
es
[
3
]
.
Facial
f
ea
tu
r
es
h
a
v
e
th
e
ad
v
an
tag
e
o
f
b
ein
g
u
n
i
q
u
e
an
d
p
er
m
a
n
en
t
a
n
d
ca
n
n
o
t
b
e
f
alsi
f
ied
u
n
li
k
e
co
n
v
e
n
tio
n
al
m
ea
n
s
s
u
c
h
as
p
ass
w
o
r
d
s
o
r
b
ad
g
es.
T
h
e
f
ac
e
is
a
n
es
s
en
t
ial
i
n
f
o
r
m
er
o
f
id
en
tit
y
a
n
d
a
b
asi
s
f
o
r
id
en
ti
f
y
i
n
g
p
eo
p
le
(
id
en
tity
p
h
o
to
g
r
ap
h
y
,
an
th
r
o
p
o
m
etr
y
,
f
ac
ial
r
ec
o
g
n
itio
n
.
I
n
th
i
s
w
o
r
k
w
e
p
r
esen
t
a
n
alg
o
r
ith
m
f
o
r
u
s
er
id
en
ti
f
ica
t
io
n
f
o
cu
s
es
o
n
d
eter
m
i
n
in
g
th
e
A
GE
.
T
h
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
w
s
:
in
t
h
e
n
ex
t
s
ec
tio
n
,
r
elate
d
wo
r
k
s
ar
e
d
escr
ib
ed
.
I
n
s
ec
tio
n
3
,
w
e
p
r
esen
t
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
.
I
n
s
ec
tio
n
4
,
p
er
f
o
r
m
a
n
ce
a
n
d
r
esu
lt
s
ar
e
r
ep
o
r
ted
.
Fin
all
y
,
co
n
cl
u
s
io
n
s
ar
e
d
r
aw
n
i
n
s
ec
tio
n
5.
2.
RE
L
AT
E
D
WO
RK
S
I
n
th
is
s
ec
tio
n
,
w
e
r
ev
ie
w
t
h
e
ex
is
t
in
g
A
GE
r
ec
o
g
n
itio
n
w
o
r
k
s
.
A
ll th
e
p
r
esen
ted
p
r
ev
io
u
s
w
o
r
k
cited
in
th
is
s
ec
tio
n
b
ased
o
n
th
e
u
s
e
o
f
DL
ar
c
h
itect
u
r
es.
I
n
ag
e
esti
m
at
io
n
r
esear
ch
e
s
,
p
o
s
tu
r
e
v
o
ca
b
u
lar
y
a
n
d
in
to
n
at
io
n
p
r
esen
t
s
ig
n
i
f
ica
n
t
ele
m
e
n
ts
to
p
r
ed
i
ct
th
e
ag
e
o
f
in
ter
lo
cu
to
r
,
b
u
t
f
ac
e
s
till
t
h
e
m
o
s
t
i
m
p
o
r
tan
t
s
o
u
r
ce
o
f
in
f
o
r
m
atio
n
to
esti
m
ate
t
h
e
r
ea
l
ag
e;
w
e
ca
n
e
x
tr
ac
t
an
e
f
f
i
cien
t
m
o
d
u
lat
io
n
o
f
t
h
e
i
n
d
iv
i
d
u
al
j
u
s
t
b
y
lo
o
k
in
g
to
h
is
f
ac
e.
I
n
HC
I
,
ag
e
p
la
y
s
a
n
i
m
p
o
r
tan
t
r
o
le
in
p
r
o
d
u
cin
g
e
f
f
ec
tiv
e
a
n
d
r
o
b
u
s
t
in
ter
f
ac
e
s
in
th
e
r
ec
o
m
m
en
d
ed
s
y
s
te
m
,
ad
ap
tiv
e
in
ter
f
ac
e,
s
m
ar
t
tech
n
o
lo
g
ie
s
an
d
e
m
b
o
d
ied
r
ec
o
g
n
itio
n
.
Gen
d
er
r
ec
o
g
n
itio
n
is
also
an
i
m
p
o
r
tan
t
f
ac
to
r
in
u
s
er
id
en
tific
atio
n
,
an
d
m
a
n
y
r
e
s
ea
r
ch
er
s
ex
p
lo
it
d
if
f
er
en
t
b
io
m
etr
ic
tech
n
iq
u
es
f
o
r
g
en
d
er
id
en
ti
f
icatio
n
.
Gen
d
er
r
ec
o
g
n
itio
n
,
b
ased
o
n
2
D
o
r
3
D
i
m
a
g
es,
is
p
ar
t
o
f
b
io
m
etr
i
c
tech
n
o
lo
g
ies
t
h
at
ca
n
b
e
ef
f
icie
n
t
i
n
f
o
r
m
atio
n
t
o
p
r
ec
is
e
th
e
in
d
i
v
id
u
al
id
en
ti
t
y
.
S
u
ch
a
s
ag
e
a
n
d
g
e
n
d
er
r
ec
o
g
n
itio
n
,
eth
n
ici
t
y
p
r
esen
ts
a
n
i
m
p
o
r
tan
t a
t
tr
ib
u
te
in
u
s
er
id
en
ti
f
icatio
n
in
m
an
y
t
y
p
es o
f
r
esear
c
h
,
esp
ec
iall
y
i
n
s
ec
u
r
it
y
.
T
h
e
n
o
tio
n
o
f
eth
n
ic
it
y
w
a
s
u
s
ed
f
r
o
m
th
e
eig
h
tee
n
t
h
y
ea
r
s
to
d
if
f
er
en
tia
te
in
d
iv
id
u
al
g
r
o
u
p
s
h
a
v
i
n
g
d
if
f
er
e
n
t p
h
y
s
ical
cr
iter
ia.
I
n
l
i
ter
atu
r
e,
m
a
n
y
r
esear
ch
er
s
e
x
p
lo
it f
ac
ial
f
ea
t
u
r
es to
e
s
ti
m
ate
A
GE
,
f
o
r
ex
a
m
p
le
in
t
h
eir
ar
ticle,
J
o
r
d
i
et
al
.
[
1
]
p
r
esen
ted
a
n
o
v
el
m
eth
o
d
f
o
r
g
e
n
d
er
id
en
ti
f
icatio
n
u
s
i
n
g
t
h
e
d
ee
p
n
eu
r
al
n
et
w
o
r
k
(
DNN)
,
th
e
n
e
w
ar
ch
i
tectu
r
e
p
r
o
p
o
s
ed
in
th
eir
w
o
r
k
b
ased
o
n
th
e
u
s
e
o
f
lo
ca
l f
ea
tu
r
e
s
u
s
in
g
s
m
all
o
v
er
lap
p
in
g
r
eg
io
n
.
T
h
e
L
o
ca
l D
NN
w
as t
ested
o
n
L
FW
an
d
Galla
g
h
er
’
s
d
atab
ase
an
d
g
iv
es a
n
i
m
p
o
r
t
an
t r
esu
lt e
s
p
ec
iall
y
u
s
i
n
g
f
o
u
r
la
y
er
s
; th
e
d
if
f
er
en
ce
w
a
s
s
u
b
s
ta
n
tial c
o
m
p
ar
ed
w
it
h
t
h
e
n
et
w
o
r
k
w
i
th
o
n
e
la
y
er
.
I
n
2
0
1
6
Ma
n
ep
ali
et
al
.
p
r
esen
ted
a
n
o
v
el
m
et
h
o
d
o
f
ag
e
esti
m
at
io
n
w
i
th
a
r
ea
l
i
m
a
g
e,
d
if
f
er
e
n
t
p
o
s
es
a
n
d
d
if
f
er
e
n
t
e
m
o
t
io
n
s
u
s
i
n
g
L
FW
,
Gr
o
u
p
s
,
an
d
FER
E
T
d
atasets
.
I
n
th
is
m
et
h
o
d
,
a
d
ictio
n
ar
y
is
p
r
o
d
u
ce
d
f
r
o
m
th
e
tr
ain
i
n
g
p
h
a
s
e,
an
d
m
a
tch
i
n
g
i
s
co
m
p
leted
b
y
r
eb
u
ild
in
g
t
h
e
te
s
ti
n
g
i
m
ag
e
u
s
in
g
a
s
p
ar
s
e
d
ictio
n
ar
y
.
K
ay
a
e
t
al
.
[
2
]
p
r
es
en
t
e
d
a
n
al
g
o
r
ith
m
o
f
A
G
E
r
ec
o
g
n
it
io
n
f
o
r
c
h
i
l
d
r
en
th
r
o
w
s
p
e
e
ch
th
ey
u
s
e
d
a
d
a
t
as
e
t
c
o
n
t
a
in
s
th
e
s
e
q
u
en
c
e
f
o
r
ch
il
d
r
en
w
ith
ag
es
b
e
tw
ee
n
th
r
e
e
an
d
s
ev
en
y
e
a
r
s
in
a
d
if
f
e
r
en
t
em
o
ti
o
n
a
l
s
t
at
e
(
c
o
m
f
o
r
t
,
d
is
c
o
m
f
o
r
t
an
d
n
e
u
t
r
a
l
)
.
T
h
e
class
if
icat
i
o
n
p
r
o
ce
s
s
w
a
s
ap
p
lied
u
s
in
g
ex
tr
e
m
e
m
ac
h
in
e
lear
n
i
n
g
(
E
M
L
)
w
i
th
a
s
in
g
le
la
y
e
r
f
ee
d
f
o
r
w
ar
d
n
et
w
o
r
k
(
L
FN)
.
I
n
th
eir
ar
ticle
A
n
tip
o
v
et
al
.
[
3
]
p
r
esen
t
an
alg
o
r
ith
m
o
f
ag
e
an
d
g
en
d
er
class
i
f
icatio
n
u
s
in
g
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
(
C
NN)
;
t
h
e
y
u
s
ed
th
r
ee
p
o
p
u
lar
b
en
ch
m
ar
k
s
L
FW
,
FG
-
NE
T
,
an
d
MO
R
P
H
f
o
r
th
e
tr
ai
n
i
n
g
p
r
o
ce
s
s
.
I
n
2
0
1
7
L
ei
C
ai
et
al
.
[
4
]
p
r
esen
t
a
n
e
w
ar
ch
i
tectu
r
e
f
o
r
g
en
d
er
r
ec
o
g
n
i
tio
n
f
o
r
p
ed
estrian
s
;
to
ad
d
r
ess
t
h
e
p
r
o
b
le
m
o
f
ill
u
m
in
a
tio
n
s
,
o
cc
lu
s
io
n
a
n
d
poor
-
q
u
alit
y
r
esear
ch
er
s
u
s
ed
a
n
ef
f
ec
ti
v
e
H
OG
-
ass
is
ted
d
ee
p
f
ea
t
u
r
e
lear
n
i
n
g
(
HD
FL
)
.
T
h
e
y
ex
p
lo
it
t
h
e
d
ee
p
-
lear
n
ed
a
n
d
w
e
ig
h
ted
HOG
f
ea
t
u
r
e
ex
tr
ac
tio
n
b
r
an
c
h
es
s
i
m
u
ltan
eo
u
s
l
y
o
n
t
h
e
i
n
p
u
t i
m
a
g
es.
3.
P
RO
P
O
SE
D
M
E
T
H
O
D
3
.
1
.
O
v
er
v
ie
w
o
f
t
he
pro
po
s
ed
cla
s
s
if
ica
t
io
n a
lg
o
rit
h
m
I
n
th
e
a
g
e
e
s
ti
m
atio
n
p
r
o
ce
s
s
,
o
u
r
g
o
al
is
n
o
t
to
f
in
d
t
h
e
e
x
ac
t
ag
e
b
u
t
to
f
i
n
d
t
h
e
ag
e
g
r
o
u
p
.
T
h
er
ef
o
r
e
w
e
d
escr
ib
e
th
r
ee
a
g
e
g
r
o
u
p
s
;
y
o
u
t
h
(
1
6
-
3
0
)
,
s
en
io
r
(
3
1
-
5
0
)
an
d
eld
er
l
y
(
5
1
-
o
v
er
)
.
Fo
r
th
e
eth
n
icit
y
p
r
o
ce
s
s
,
w
e
cla
s
s
i
f
y
t
h
e
r
ac
e
in
to
t
w
o
class
es
:
C
a
u
ca
s
ia
n
a
n
d
n
o
t
C
a
u
ca
s
ia
n
.
W
e
h
a
v
e
t
h
r
ee
class
e
s
f
o
r
ag
e
s
,
t
w
o
f
o
r
eth
n
icit
y
a
n
d
t
w
o
f
o
r
g
e
n
d
er
.
T
h
e
n
u
m
b
er
o
f
f
i
n
al
clas
s
es
is
1
2
as
d
escr
ib
ed
in
F
ig
u
r
e
1
an
d
o
r
g
an
ized
as
f
o
llo
w
:
n
o
t
C
a
u
ca
s
ia
n
f
e
m
ale
(
N
C
F)
f
r
o
m
1
6
to
3
0
,
n
o
t
C
au
ca
s
ia
n
f
r
o
m
3
1
to
5
0
,
n
o
t
C
au
ca
s
ia
n
f
e
m
ale
m
o
r
e
t
h
an
5
0
,
ca
u
ca
s
ia
n
f
e
m
ale
(
C
F)
f
r
o
m
1
6
to
3
0
,
C
au
ca
s
ian
f
e
m
ale
f
r
o
m
3
1
to
5
0
,
C
au
ca
s
ian
f
e
m
ale
m
o
r
e
th
a
n
5
0
,
No
t
C
au
ca
s
ian
m
ale
(
N
C
M)
f
r
o
m
1
6
to
3
0
,
n
o
t
C
au
ca
s
ia
n
m
ale
f
r
o
m
3
1
to
5
0
,
n
o
t
C
a
u
ca
s
ia
n
m
ale
m
o
r
e
t
h
an
5
0
,
C
au
ca
s
ian
m
ale
(C
M)
f
r
o
m
1
6
to
3
0
,
C
au
ca
s
ian
m
ale
f
r
o
m
3
1
to
5
0
,
C
au
ca
s
ia
n
m
ale
m
o
r
e
th
an
5
0
.
I
n
th
is
w
o
r
k
,
w
e
s
tar
t
b
y
d
ata
p
r
e
-
p
r
o
ce
s
s
i
n
g
:
T
h
e
f
ir
s
t
s
tep
i
n
o
u
r
wo
r
k
is
to
ex
tr
ac
t
th
e
f
ac
e
f
r
o
m
th
e
i
m
a
g
es,
f
o
r
th
i
s
r
ea
s
o
n
,
w
e
u
s
ed
t
h
e
A
d
aB
o
o
s
t f
r
a
m
e
w
o
r
k
o
f
Vio
la
P
.
an
d
J
o
n
es [
5
]
p
u
b
lis
h
ed
o
n
J
u
l
y
1
3
2
0
0
1
.
T
h
e
s
ec
o
n
d
s
tep
is
to
cr
o
p
u
p
f
ac
e
s
.
Fi
n
all
y
,
an
in
-
p
la
n
e
r
o
tatio
n
is
ap
p
lied
to
ad
j
u
s
t
th
e
h
ea
d
o
r
ien
tatio
n
b
ec
au
s
e
it
co
u
ld
in
f
lu
e
n
ce
th
e
alg
o
r
it
h
m
p
er
f
o
r
m
a
n
ce
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
ap
p
lied
t
o
th
e
tw
o
f
r
a
m
e
w
o
r
k
s
A
GE
R
a
n
d
E
R
.
I
n
m
ac
h
in
e
lear
n
in
g
p
r
o
ce
s
s
,
th
e
m
ai
n
p
r
o
b
lem
o
f
class
if
ica
t
io
n
is
to
d
is
tin
g
u
i
s
h
to
w
h
ic
h
o
f
a
s
et
o
f
g
r
o
u
p
s
a
n
e
w
s
a
m
p
le
b
elo
n
g
s
,
b
y
e
x
tr
ac
t
in
g
f
ea
t
u
r
es
o
f
a
tr
ain
in
g
s
et
o
f
d
ata
w
h
ic
h
co
n
tain
s
o
m
e
o
b
s
er
v
atio
n
s
w
h
o
s
e
cla
s
s
m
e
m
b
er
s
h
ip
i
s
alr
ea
d
y
k
n
o
w
n
.
I
n
t
h
is
a
x
i
s
,
th
e
r
e
v
o
lu
tio
n
t
h
at
ar
ti
f
icia
l
in
telli
g
e
n
c
e
an
d
esp
ec
iall
y
m
a
ch
in
e
lear
n
i
n
g
ar
e
e
x
p
er
ien
ci
n
g
to
d
a
y
,
D
L
is
at
t
h
e
f
o
r
ef
r
o
n
t.
Me
th
o
d
s
ex
p
lo
iti
n
g
th
e
t
h
eo
r
y
o
f
D
NN
i
n
au
to
m
ati
c
lear
n
in
g
h
av
e
p
r
o
v
ed
t
h
eir
r
o
b
u
s
t
n
ess
o
n
co
m
p
le
x
tas
k
s
in
t
h
e
f
ie
ld
s
o
f
i
m
a
g
i
n
g
an
d
ac
o
u
s
tic
p
r
o
ce
s
s
i
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
S
emi
-
s
u
p
ervis
ed
a
u
to
-
en
c
o
r
d
e
r
fo
r
fa
cia
l a
ttr
ib
u
tes r
ec
o
g
n
itio
n
(
S
o
u
ma
y
a
Za
g
h
b
a
n
i
)
2171
I
n
i
t
i
a
l
I
m
a
g
e
N
C
F
f
r
o
m
1
6
t
o
3
0
y
e
a
r
s
F
e
a
t
u
r
e
s
e
x
t
r
a
c
t
i
o
n
N
C
F
f
r
o
m
3
1
t
o
5
0
N
C
F
f
r
o
m
3
1
t
o
5
0
C
M
f
r
o
m
3
1
t
o
5
0
N
C
M
f
r
o
m
3
1
t
o
5
0
N
C
F
m
o
r
e
t
h
a
n
5
0
C
M
f
r
o
m
1
6
t
o
3
0
N
C
M
f
r
o
m
1
6
t
o
3
0
C
F
f
r
o
m
1
6
t
o
3
0
F
a
c
i
a
l
i
m
a
g
e
C
F
m
o
r
e
t
h
a
n
5
0
N
C
M
m
o
r
e
t
h
a
n
5
0
C
M
m
o
r
e
t
h
a
n
5
0
Fig
u
r
e
1
.
Dif
f
er
en
t c
la
s
s
es
u
s
e
d
f
o
r
ag
e,
g
en
d
er
an
d
eth
n
ici
t
y
r
ec
o
g
n
itio
n
3
.
2
.
Aut
o
enco
der
m
o
del
T
h
e
au
to
en
co
d
er
s
p
r
esen
t
a
r
o
b
u
s
t
ar
ch
itect
u
r
e
o
f
DL
to
d
ay
,
as
d
escr
ib
ed
in
[
6
]
,
au
to
en
co
d
er
s
ar
e
s
tr
u
ct
u
r
es
co
m
p
o
s
ed
o
f
t
w
o
p
ar
ts
:
an
en
co
d
er
an
d
a
d
ec
o
d
er
,
th
e
y
ar
e
b
u
ilt
w
i
th
d
ee
p
ar
ch
ite
ctu
r
es.
T
h
e
n
u
m
b
er
o
f
n
e
u
r
o
n
s
in
th
e
la
s
t d
ec
o
d
er
la
y
er
is
eq
u
al
t
o
t
h
e
s
ize
o
f
t
h
e
n
et
w
o
r
k
i
n
p
u
t.
T
h
e
p
u
r
p
o
s
e
o
f
a
n
au
to
e
n
co
d
er
is
to
f
in
d
a
co
d
ed
r
ep
r
esen
tatio
n
o
f
an
i
n
p
u
t
t
h
at
ca
n
b
e
d
ec
o
d
ed
ac
cu
r
atel
y
.
Su
c
h
a
n
et
w
o
r
k
is
d
r
iv
e
n
to
f
i
n
d
a
r
ep
r
esen
tatio
n
o
f
t
h
e
i
n
p
u
t
d
ata
an
d
to
lear
n
th
e
co
n
n
ec
tio
n
b
et
w
ee
n
a
n
en
tr
y
a
n
d
it
s
h
id
d
en
r
ep
r
esen
tatio
n
.
W
e
co
n
s
id
er
th
e
in
p
u
t v
ec
to
r
X
tr
an
s
f
o
r
m
ed
in
to
a
h
id
d
en
r
ep
r
esen
tatio
n
a
s
f
o
llo
w
s
:
z=
f
(
x
)
=
s
i
g
m
(
+b
)
(
1
)
w
h
er
e
W
an
d
b
p
r
esen
ts
th
e
weig
h
t
m
atr
i
x
an
d
th
e
b
ias
b
et
w
ee
n
th
e
i
n
p
u
t
a
n
d
h
id
d
en
la
y
er
,
r
esp
ec
tiv
el
y
an
d
th
e
m
ap
p
in
g
w
as
t
h
r
o
u
g
h
s
i
g
m
o
id
f
u
n
ctio
n
o
r
tan
h
f
u
n
ctio
n
(
s
i
g
m
o
id
f
u
n
ctio
n
is
u
s
u
al
l
y
u
s
ed
a
s
an
ac
tiv
a
tio
n
f
u
n
ctio
n
)
:
Ɵ
=sig
m
(
y
)
=
1
1
+
ex
p
(
−
y
)
,
o
r
(
2
)
Ɵ
=ta
n
h
(
y
)
=
ex
p
(
y
)
−
ex
p
(
−
y
)
ex
p
(
y
)
+
ex
p
(
−
y
)
f
o
r
th
e
d
ec
o
d
in
g
s
tag
e,
t
h
e
h
id
d
en
r
ep
r
esen
tatio
n
i
s
m
ap
p
ed
b
ac
k
to
th
e
f
ir
s
t r
ep
r
esen
tatio
n
as f
o
llo
w
i
n
g
:
X
̂
=
g
(
z)
=
s
ig
m
(
W
’
z
+
b
’
)
(
3
)
w
h
er
e
W
’
an
d
b
’
d
en
o
te
th
e
w
ei
g
h
t
m
atr
i
x
an
d
th
e
b
ias
b
et
wee
n
th
e
h
id
d
en
an
d
o
u
tp
u
t
la
y
er
r
esp
ec
tiv
el
y
.
T
h
e
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
is
d
ef
in
ed
b
y
m
i
n
i
m
izi
n
g
th
e
E
u
clid
ea
n
co
s
t:
a
r
gmin
W
,
w
′
‖
X
−
X
̂
‖
2
2
(
4
)
T
h
e
p
u
r
p
o
s
e
o
f
an
au
to
en
co
d
er
is
to
estab
lis
h
s
o
m
e
co
r
r
elatio
n
f
o
r
th
e
in
p
u
t
d
ata
f
o
r
d
im
e
n
s
io
n
alit
y
r
ed
u
ctio
n
af
ter
t
h
at
a
clas
s
if
i
ca
tio
n
p
r
o
ce
s
s
is
estab
li
s
h
ed
.
I
n
liter
atu
r
e,
m
a
n
y
t
y
p
es
o
f
au
to
en
co
d
er
s
w
er
e
p
r
esen
ted
an
d
d
is
c
u
s
s
ed
;
f
o
r
ex
a
m
p
le,
p
r
o
g
r
es
s
iv
e,
s
p
ar
s
e
an
d
d
en
o
is
i
n
g
au
to
e
n
co
d
er
s
.
T
h
e
s
p
ar
s
e
co
d
in
g
m
et
h
o
d
p
r
esen
ts
a
g
o
o
d
r
ep
r
e
s
en
tat
io
n
a
n
d
p
er
f
o
r
m
a
n
ce
.
I
n
f
ac
t,
t
h
e
ter
m
s
p
ar
s
e
i
s
u
s
ed
to
in
d
icate
t
h
at
w
e
n
ee
d
h
id
d
en
n
e
u
r
o
n
s
w
ith
t
h
e
s
a
m
e
p
r
o
b
ab
ilit
y
o
f
ac
tiv
a
tio
n
.
T
h
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
o
f
th
e
h
id
d
en
la
y
er
s
i
s
s
m
al
ler
th
a
n
t
h
at
o
f
t
h
e
i
n
p
u
t
a
n
d
o
u
tp
u
t
la
y
er
s
.
I
t
w
il
l
b
e
i
m
p
o
r
tan
t
to
co
m
p
r
ess
d
ata
an
d
tr
y
to
f
i
n
d
a
co
r
r
elatio
n
b
et
w
ee
n
t
h
e
d
ata
a
n
d
t
h
er
ef
o
r
e
class
i
f
y
t
h
e
m
ac
co
r
d
in
g
to
t
h
is
co
r
r
elatio
n
.
T
h
e
m
i
n
i
m
ize
d
f
u
n
ctio
n
o
f
s
p
ar
s
e
au
to
en
co
d
er
p
r
esen
ted
as f
o
llo
w
i
n
g
[
6
,
7
]
:
a
r
gmin
W
,
W
′
‖
X
−
W
′
Ɵ
(
WX
)
‖
F
2
+
β
∑
KL
(
ρ
ǁ
ρ
̂
j
)
m
j
=
1
(
5
)
w
h
er
e
m
is
t
h
e
n
u
m
b
er
o
f
h
id
d
en
n
o
d
es,
a
n
d
β
i
s
a
c
o
ef
f
i
ci
en
t
th
a
t
d
et
e
r
m
in
es
th
e
w
ei
g
h
t
o
f
s
p
a
r
s
i
ty
p
en
a
lty
i
t
em
.
ρ
is
th
e
s
p
a
r
s
i
ty
p
a
r
am
et
e
r
,
i
t
p
r
e
s
en
ts
th
e
ta
r
g
e
t
av
e
r
ag
e
ac
t
iv
a
t
i
o
n
o
f
h
i
d
d
e
n
u
n
it
s
,
w
h
i
ch
is
g
en
e
r
a
lly
a
s
m
a
l
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
18
,
No
.
4
,
A
u
g
u
s
t 2
0
2
0
:
2
1
6
9
-
2
1
7
6
2172
v
a
lu
e
n
e
a
r
i
n
g
z
e
r
o
:
ρ
̂
j
=
1
N
∑
h
i
(
x
i
)
N
i
=
1
d
e
n
o
te
s
th
e
a
v
e
r
ag
e
a
ct
iv
a
ti
o
n
o
f
h
i
d
d
en
n
o
d
e
j,
a
n
d
t
h
e
Ku
ll
b
a
ck
-
L
e
i
b
l
e
r
d
i
v
e
r
g
en
c
e
ca
n
b
e
d
ef
i
n
e
d
b
y
:
KL
(
ρ
ǁ
ρ
̂
j
)
=
ρ
l
og
ρ
ρ
̂
j
+
(
1
−
ρ
)
l
og
1
−
ρ
1
−
ρ
̂
j
T
h
e
p
u
r
p
o
s
e
o
f
tr
ain
in
g
a
s
p
ar
s
e
au
to
en
co
d
er
is
to
lear
n
th
e
alg
o
r
ith
m
to
au
to
m
a
ticall
y
e
x
tr
a
ct
f
ea
tu
r
e
s
f
r
o
m
u
n
lab
eled
d
ata.
R
ec
e
n
tl
y
,
m
a
n
y
r
e
s
ea
r
ch
er
s
co
n
s
id
e
r
ed
th
at
au
to
en
co
d
er
s
m
a
y
b
e
s
e
m
i
-
s
u
p
er
v
i
s
ed
b
ec
au
s
e
t
h
e
y
co
u
ld
b
e
m
o
r
e
p
er
f
o
r
m
an
t
w
h
en
th
e
y
s
p
ec
if
y
c
lass
es,
th
eir
r
e
s
ea
r
ch
es
d
en
o
te
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
is
n
e
w
ar
ch
itect
u
r
e
a
n
d
r
esu
lts
s
h
o
w
n
t
h
e
r
o
b
u
s
t
n
es
s
o
f
s
u
p
er
v
is
ed
au
to
e
n
co
d
er
s
th
a
n
m
an
y
o
th
er
o
r
d
in
ar
y
n
et
w
o
r
k
s
[
8
-
1
1
]
.
I
n
th
is
w
o
r
k
w
e
u
s
ed
th
e
s
p
ar
s
e
au
to
en
co
d
er
in
a
s
em
i
-
s
u
p
er
v
i
s
ed
m
a
n
n
er
to
p
r
e
d
ict
ag
e,
g
en
d
er
,
eth
n
icit
y
f
r
o
m
f
ac
ial
i
m
ag
e
s
.
I
n
f
ac
t,
th
e
id
ea
o
f
s
e
m
i
s
u
p
er
v
is
ed
au
to
en
co
d
er
s
o
r
class
s
p
ec
if
ic
au
to
en
co
d
er
is
to
in
co
r
p
o
r
ate
in
f
o
r
m
atio
n
ab
o
u
t
cla
s
s
es
i
n
t
h
e
b
asic
ar
ch
itec
tu
r
e
o
f
a
u
to
en
c
o
d
er
s
in
t
h
e
p
u
r
p
o
s
e
to
im
p
r
o
v
e
th
e
s
u
p
er
v
is
io
n
p
r
o
ce
s
s
.
W
e
k
ee
p
th
e
s
a
m
e
ar
ch
itect
u
r
e
o
f
u
n
-
s
u
p
er
v
i
s
io
n
p
r
o
ce
s
s
an
d
w
e
ad
d
lab
eled
d
ata
to
s
u
p
er
v
is
e
th
e
au
to
en
co
d
er
an
d
lear
n
f
ea
tu
r
es
s
u
c
h
th
at
t
h
e
illu
s
tr
atio
n
s
o
f
s
a
m
p
les
r
ef
er
r
in
g
to
a
class
ar
e
eq
u
al
to
t
h
e
m
ea
n
r
ep
r
esen
tatio
n
o
f
t
h
e
s
a
m
e
cl
ass
.
Fro
m
t
h
e
f
ir
s
t
r
esear
c
h
er
s
w
h
o
w
o
r
k
ed
w
it
h
au
to
en
co
d
er
s
in
a
s
u
p
er
v
is
ed
m
an
n
er
w
e
cite
Gao
et
al
.
in
2
0
1
5
[
1
0
]
;
th
ey
m
o
d
i
f
ied
d
en
o
is
in
g
au
to
en
co
d
d
er
s
to
o
p
tim
ize
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
id
en
ti
f
icat
io
n
.
Sp
ec
if
icall
y
,
t
h
e
id
ea
o
f
s
e
mi
-
s
u
p
er
v
i
s
ed
au
to
en
co
d
er
to
m
an
u
a
l
ly
s
p
e
c
if
y
t
h
e
in
p
u
t
f
e
a
tu
r
es
x
g
iv
en
t
o
t
h
e
al
g
o
r
ith
m
.
O
n
c
e
a
g
o
o
d
f
ea
t
u
r
e
r
e
p
r
es
en
t
a
ti
o
n
i
s
g
iv
en
,
a
l
e
a
r
n
in
g
a
lg
o
r
it
h
m
c
an
d
o
w
e
ll
.
T
h
e
c
l
a
s
s
-
s
p
es
if
i
c
a
u
t
o
en
c
o
d
e
r
m
ak
e
s
th
e
f
e
a
tu
r
es
c
o
n
s
i
s
t
t
o
th
e
s
am
e
cl
a
s
s
s
im
i
l
a
r
a
ty
,
i
n
f
a
c
t
,
it
ex
t
r
ac
ts
m
o
r
e
e
f
f
i
ci
en
t f
e
a
tu
r
es
f
o
r
t
h
e
s
am
e
g
r
o
u
p
r
e
p
r
e
s
en
t
at
i
o
n
.
A
s
d
escr
ib
ed
p
r
ev
io
u
s
l
y
,
i
n
th
e
f
ir
s
t
s
tep
,
w
e
s
u
g
g
est
to
lear
n
f
ea
t
u
r
es
s
u
c
h
u
s
th
e
y
h
a
v
e
th
e
s
a
m
e
s
p
ar
s
it
y
s
ig
n
atu
r
e
ac
r
o
s
s
e
v
er
y
cla
s
s
.
Fo
r
a
g
iv
e
n
i
n
p
u
t
X,
t
h
e
lo
s
s
f
u
n
ct
i
o
n
in
t
h
e
tr
ad
itio
n
a
l
ar
ch
itect
u
r
e
o
f
au
to
en
co
d
er
is
g
iv
e
n
as m
en
tio
n
ed
in
t
h
e
(
4
)
.
I
n
th
eir
w
o
r
k
Ma
j
u
m
d
ar
et
al
.
p
r
o
p
o
s
e
[
1
2
,
1
3
]
1
1
n
o
r
m
f
o
r
r
eg
u
lar
izatio
n
p
r
esen
ted
as f
o
llo
w
:
a
r
gmin
W
,
W
′
‖
X
−
W
′
Ɵ
(
WX
)
‖
F
2
+λ
|
|
WX
|
|
1
(
6
)
t
o
in
c
o
r
p
o
r
a
t
e
th
e
s
u
p
e
r
v
i
s
i
o
n
in
t
o
th
e
cl
as
s
ic
a
r
ch
it
e
c
tu
r
e
o
f
au
t
o
en
c
o
d
e
r
s
w
e
h
av
e
t
o
p
as
s
f
r
o
m
th
e
u
n
s
u
p
e
r
v
is
in
g
t
o
t
h
e
s
u
p
e
r
v
is
in
g
b
y
l
a
b
e
l
li
n
g
d
a
t
a
,
h
o
w
ev
e
r
t
h
e
t
r
a
in
in
g
d
a
ta
c
o
u
l
d
b
e
p
r
e
s
e
n
t
e
d
u
s
f
o
ll
o
w
in
g
:
X=
[
x
1
,
1
|
.
.
|
x
1
,
n
1
x
2
,
1
|
.
.
|
x
2
,
n
2
.
.
x
c
,
1
|
.
.
|
x
c
,
n
c
]
w
h
er
e
th
e
tr
ain
i
n
g
d
ata
is
d
iv
id
ed
in
to
class
es
(
c)
.
T
h
e
id
ea
is
to
lear
n
f
ea
tu
r
es
i
n
to
co
m
m
o
n
s
p
ar
s
e
s
u
p
p
o
r
t.
an
d
WX
i
w
i
ll b
e
r
o
w
s
p
ar
s
e.
T
h
is
is
ac
h
ie
v
ed
b
y
in
co
r
p
o
r
atin
g
l
2
,
1
n
o
r
m
r
eg
u
lar
izatio
n
as f
o
llo
w
i
n
g
:
a
r
gmin
W
,
W
′
‖
X
−
W
′
Ɵ
(
WX
)
‖
F
2
+λ
∑
|
|
W
X
c
|
|
2
,
1
c
c
=
1
+
β
∑
KL
(
ρ
ǁ
ρ
̂
c
)
c
c
=
1
(
7
)
w
h
er
e
‖
v
‖
2
,
1
=
∑
‖
v
j
→
‖
2
j
is
th
e
s
u
m
o
f
l
2
-
n
o
r
m
s
o
f
t
h
e
r
o
w
s
.
W
e
n
o
te
t
h
at
t
h
e
i
n
p
u
t
X
ap
p
ar
tain
to
cla
s
s
c
d
u
r
in
g
th
e
cla
s
s
i
f
icatio
n
p
h
ases
,
th
e
n
t
ak
in
g
in
to
co
n
s
ed
er
aio
n
a
ll
cla
s
s
es
w
e
o
p
ti
m
ize
w
an
d
w
’
lea
d
in
g
to
a
m
i
n
u
m
u
m
o
f
o
b
j
ec
t
if
f
u
n
c
tio
n
.
T
h
e
in
n
e
r
l
2
-
n
o
r
m
p
r
o
m
o
tes
a
d
en
s
e
(
n
o
n
-
ze
r
o
)
s
o
lu
tio
n
w
ith
in
t
h
e
s
elec
ted
r
o
w
s
,
b
u
t
th
e
o
u
ter
l1
-
n
o
r
m
(
s
u
m
)
en
f
o
r
ce
s
s
p
ar
s
it
y
i
n
s
elec
ti
n
g
th
e
r
o
w
s
[
6
,
1
3
]
.
T
h
e
p
r
o
p
o
s
ed
s
o
l
u
tio
n
s
h
o
w
n
in
(
7
)
i
m
p
r
o
u
v
e
r
o
w
-
s
p
ar
s
it
y
w
it
h
i
n
ev
er
y
g
r
o
u
p
.
T
h
e
ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
is
d
es
cr
ib
ed
in
F
ig
u
r
e
2
.
4.
E
XP
E
R
I
M
E
NT
A
L
RE
SUL
T
S
4
.
1
.
M
O
RP
H
I
I
da
t
a
s
et
Fo
r
A
GE
w
e
u
s
ed
th
e
MO
R
P
H
I
I
an
d
FG
-
NE
T
d
atab
ase.
M
OR
P
H
I
I
b
en
ch
m
ar
k
h
o
ld
s
5
5
,
0
0
0
im
ag
e
s
o
f
m
o
r
e
th
an
1
3
,
0
0
0
v
o
lu
n
teer
s
in
d
iv
id
u
als,
p
ass
o
v
er
f
r
o
m
2
0
0
3
to
late
2
0
0
7
.
A
g
es
g
r
o
u
p
f
r
o
m
1
6
to
7
7
w
it
h
a
m
ed
ia
n
ag
e
o
f
3
3
.
T
h
e
av
er
a
g
e
ti
m
e
ta
k
en
b
et
w
ee
n
p
h
o
to
s
is
1
6
4
d
ay
s
an
d
th
e
av
er
a
g
e
n
u
m
b
er
o
f
i
m
a
g
es
p
er
in
d
iv
id
u
al
i
s
4
(
t
h
e
m
i
n
i
m
u
m
b
ein
g
o
n
e
d
a
y
an
d
t
h
e
m
a
x
i
m
u
m
is
1
6
8
1
d
ay
s
)
.
T
h
e
s
tan
d
ar
d
d
ev
iatio
n
o
f
d
a
y
s
b
et
w
ee
n
i
m
a
g
es
is
1
8
0
.
I
n
liter
atu
r
e,
MO
R
P
H
I
I
d
atab
ase
o
f
f
er
s
t
h
e
m
o
s
t
ex
ten
s
i
v
e
d
atab
ase
th
at
co
m
b
in
es
A
GE
.
T
h
e
d
is
tr
ib
u
tio
n
s
o
f
t
h
e
f
ac
e
i
m
a
g
es
ar
e
s
h
o
w
n
i
n
T
ab
le
1
.
Dif
f
er
en
tl
y
,
o
f
p
r
ev
io
u
s
s
t
u
d
ies
o
n
MO
R
P
H
I
I
,
th
at
d
iv
id
e
th
e
wh
o
le
MO
R
P
H
I
I
d
atab
ase
W
i
n
to
th
r
ee
s
et
s
,
S1
,
S2
,
an
d
S3
,
w
e
s
p
lit
t
h
e
f
ac
ile
i
m
a
g
es
i
n
to
1
2
class
es
as
s
h
o
wn
in
Fig
u
r
e
2
.
So
m
e
ex
a
m
p
le
s
o
f
f
ac
ial
i
m
a
g
es
e
x
tr
ac
ted
f
r
o
m
MO
R
P
H
I
I
d
ataset
ar
e
s
h
o
w
n
i
n
Fi
g
u
r
e
3
,
f
r
o
m
d
if
f
er
en
t
a
g
es
a
n
d
eth
n
icitie
s
.
I
n
ev
er
y
i
m
ag
e,
w
e
d
etec
t
a
n
d
cr
o
p
th
e
f
ac
e
ar
ea
,
an
d
th
e
i
m
ag
e
is
d
i
m
e
n
s
io
n
ed
in
to
3
2
*
3
2
.
On
l
y
t
w
o
eth
n
ic
s
g
r
o
u
p
s
ar
e
u
s
ed
f
o
r
th
is
w
o
r
k
(
C
au
ca
s
ian
a
n
d
n
o
t
C
au
ca
s
ian
w
e
n
o
te
t
h
at
f
o
r
eld
er
l
y
clas
s
es
w
e
ad
d
ed
m
o
r
e
th
an
3
5
0
i
m
a
g
es
f
o
r
ea
ch
clas
s
b
ec
au
s
e
o
f
th
e
litt
l
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
S
emi
-
s
u
p
ervis
ed
a
u
to
-
en
c
o
r
d
e
r
fo
r
fa
cia
l a
ttr
ib
u
tes r
ec
o
g
n
itio
n
(
S
o
u
ma
y
a
Za
g
h
b
a
n
i
)
2173
n
u
m
b
er
o
f
i
m
a
g
es
f
o
r
eld
er
ly
i
n
co
m
p
ar
is
o
n
w
it
h
o
th
er
g
r
o
u
p
s
.
T
h
e
illu
s
tr
atio
n
o
f
f
ac
ial
i
m
a
g
es
u
s
ed
in
t
h
e
test
p
h
ase
f
o
r
th
e
M
OR
P
HI
I
d
atab
ases
is
d
escr
ib
ed
in
T
ab
le
1
.
+
1
.
.
.
.
.
.
.
.
.
+
1
S
o
f
t
m
a
x
R
e
g
r
e
s
s
i
o
n
E
m
o
t
i
o
n
R
e
c
o
g
n
i
t
i
o
n
I
n
i
t
i
a
l
i
m
a
g
e
F
a
c
e
d
e
t
e
c
t
i
o
n
.
.
.
x
1
W
i
̂
1
+
1
.
.
.
.
.
.
.
.
.
x
2
x
3
x
4
x
n
+
1
b
i
2
3
4
…
W
i
+
1
.
.
.
.
.
.
.
.
.
x
2
x
3
x
4
x
n
b
i
2
3
4
x
n
+
1
.
.
.
.
.
.
x
2
x
3
x
4
x
n
b
i
W
i
I
n
p
u
t
F
e
a
t
u
r
e
1
I
n
p
u
t
O
u
t
p
u
t
O
u
t
p
u
t
I
n
p
u
t
F
e
a
t
u
r
e
s
2
O
u
t
p
u
t
F
e
a
t
u
r
e
2
S
o
f
t
m
a
x
c
l
a
s
s
i
f
i
e
r
Fig
u
r
e
2
.
P
ip
elin
e
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
Fig
u
r
e
3.
E
x
a
m
p
les o
f
f
ac
ia
l i
m
ag
e
s
o
f
MO
R
P
H
I
I
d
atab
ase
.
T
ab
le
1
.
T
h
e
illu
s
tr
atio
n
o
f
f
ac
ial
i
m
a
g
es
u
s
ed
in
t
h
e
te
s
t p
h
a
s
e
f
o
r
th
e
MO
R
P
HI
I
d
atab
ases
M
e
t
h
o
d
Y
e
a
r
M
A
E(
M
O
R
P
H
)
F
l
e
x
i
b
l
e
o
v
e
r
l
a
p
p
e
d
A
A
M
+
L
P
Q
[
6
]
2
0
1
5
5
.
6
8
O
D
L
F
[
1
2
]
2
0
1
7
3
.
1
2
C
S
C
+
S
T
D
P
o
o
l
i
n
g
[
1
4
]
2
0
1
7
3
.
6
6
C
S
C
+
M
a
x
P
o
o
l
i
n
g
[
1
4
]
2
0
1
7
3
.
7
8
GA
-
D
F
L
[
1
5
]
2
0
1
6
3
.
3
7
D
2
C
[
1
6
]
2
0
1
7
3
.
0
6
N
e
t
V
G
G
[
17
]
2
0
1
7
2
.
9
6
M
o
h
a
mm
e
d
e
t
a
l
.
[
18
]
2
0
1
9
3
.
1
7
T
a
h
e
r
i
e
t
a
l
.
[
19
]
2
0
1
9
2
.
8
1
P
r
o
p
o
se
d
me
t
h
o
d
-
3
.
2
6
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
18
,
No
.
4
,
A
u
g
u
s
t 2
0
2
0
:
2
1
6
9
-
2
1
7
6
2174
4
.
2
.
Ag
e
,
g
ender
a
nd
et
hn
icit
y
re
co
g
nitio
n (
AG
E
R)
T
h
is
s
ec
tio
n
i
s
d
ed
icate
d
to
ev
alu
a
te
th
e
p
r
ese
n
ted
s
t
u
d
y
o
n
A
GE
a
n
d
ca
r
r
y
o
u
t
t
h
e
v
ar
io
u
s
co
m
p
ar
is
o
n
s
w
ith
o
th
er
r
ec
o
g
n
i
tio
n
m
et
h
o
d
s
.
R
e
s
u
lt
s
o
f
A
GE
p
r
ese
n
ted
s
ep
ar
atel
y
in
T
ab
le
2
,
T
ab
le
3
an
d
T
a
b
le
4
.
T
h
e
ac
cu
r
ac
ies
o
f
A
GE
est
i
m
a
tio
n
j
o
in
tl
y
p
r
esen
t
s
7
3
.
5
%.
T
h
e
b
est
r
ates
ar
e
g
iv
en
f
o
r
NC
M
(
3
0
-
5
0
)
,
C
M
(
5
0
+)
an
d
NC
F
(
5
0
+)
w
ith
r
esp
ec
ti
v
el
y
7
8
.
7
%,
8
3
.
1
%
an
d
8
4
.
4
%.
T
h
e
d
if
f
ic
u
lt
y
in
t
h
e
co
m
p
ar
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
(
A
GE
R
j
o
in
tl
y
)
w
it
h
o
th
er
s
is
th
e
la
ck
o
f
m
o
d
els
esti
m
ati
n
g
t
h
e
AGE
jo
in
tl
y
.
W
e
u
s
ed
th
e
D
SS
A
E
al
s
o
to
esti
m
ate
A
GE
s
ep
ar
atel
y
o
n
t
h
e
MO
R
P
H
I
I
d
atab
ase,
an
d
w
e
co
m
p
ar
e
w
it
h
o
th
er
D
L
m
o
d
el
s
,
as
w
e
ll
as
w
it
h
s
o
m
e
o
th
er
ap
p
r
o
ac
h
es
o
n
MO
R
P
H
I
I
an
d
o
th
er
b
en
ch
m
ar
k
s
.
L
et
u
s
lo
o
k
at
th
e
ag
e
esti
m
atio
n
r
es
u
lts
s
h
o
w
n
in
T
ab
le
2
.
Firstl
y
,
f
o
r
test
in
g
an
d
e
v
alu
a
tin
g
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
,
w
e
u
s
ed
t
h
e
Me
an
A
b
s
o
l
u
te
E
r
r
o
r
(
MA
E
)
to
d
ef
in
e
th
e
al
g
o
r
ith
m
p
er
f
o
r
m
a
n
ce
in
ag
e
e
s
ti
m
a
tio
n
it i
s
ca
lc
u
lat
ed
as f
o
llo
w
:
Ε
=
‖
y
′
−
y
‖
2
N
(
8
)
w
h
er
e
y
’
an
d
y
p
r
ese
n
t
t
h
e
p
r
ed
icted
an
d
r
ea
l
ag
e
v
al
u
e
r
esp
ec
ti
v
el
y
an
d
N
d
en
o
tes
t
h
e
n
u
m
b
er
o
f
th
e
test
in
g
f
ac
ial
i
m
ag
e
s
.
T
h
e
p
u
r
p
o
s
e
f
r
o
m
o
u
r
w
o
r
k
i
s
n
o
t
to
ex
tr
ac
t
ex
ac
t
l
y
t
h
e
ag
e
b
u
t
w
e
l
o
o
k
to
j
u
s
t
class
i
f
y
ag
es
in
to
t
h
r
ee
r
a
n
g
e
s
,
y
o
u
t
h
(
1
6
-
3
0
)
,
s
en
io
r
(
3
1
-
5
0
)
an
d
eld
er
ly
(
5
1
-
o
v
er
)
.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
o
b
tain
s
an
M
A
E
o
f
3
.
2
6
y
ea
r
s
,
w
h
ic
h
is
co
n
s
id
er
ab
le
v
er
y
i
m
p
o
r
ta
n
t
co
m
p
ar
ed
w
it
h
o
th
er
m
e
th
o
d
s
.
T
h
is
lo
w
M
A
E
is
ev
en
s
m
aller
co
n
s
id
er
in
g
t
h
e
ag
e
esti
m
atio
n
is
p
er
f
o
r
m
ed
o
n
a
v
er
y
lar
g
e
d
atab
ase,
th
is
er
r
o
r
r
e
d
u
ctio
n
r
ate
is
s
tatis
t
icall
y
s
i
g
n
i
f
ican
t.
T
h
e
ap
p
r
o
ac
h
s
h
o
w
n
th
e
p
er
f
o
r
m
a
n
ce
o
n
y
o
u
n
g
a
n
d
s
e
n
io
r
g
r
o
u
p
s
,
th
e
M
A
E
p
r
esen
ts
th
e
b
est r
es
u
lt
w
it
h
les
s
th
a
n
a
n
d
3
.
2
6
.
T
ab
le
2
.
A
cc
u
r
ac
y
o
f
a
g
e
r
ec
o
g
n
i
tio
n
f
o
r
MO
R
P
H
I
I
d
ataset
(
%)
M
e
t
h
o
d
Y
e
a
r
M
A
E(
M
O
R
P
H
)
F
l
e
x
i
b
l
e
o
v
e
r
l
a
p
p
e
d
A
A
M
+
L
P
Q
[
6
]
2
0
1
5
5
.
6
8
O
D
L
F
[
1
2
]
2
0
1
7
3
.
1
2
C
S
C
+
S
T
D
P
o
o
l
i
n
g
[
1
4
]
2
0
1
7
3
.
6
6
C
S
C
+
M
a
x
P
o
o
l
i
n
g
[
1
4
]
2
0
1
7
3
.
7
8
GA
-
D
F
L
[
1
5
]
2
0
1
6
3
.
3
7
D
2
C
[
1
6
]
2
0
1
7
3
.
0
6
N
e
t
V
G
G
[
1
7
]
2
0
1
7
2
.
9
6
M
o
h
a
mm
e
d
e
t
a
l
.
[
18
]
2
0
1
9
3
.
1
7
T
a
h
e
r
i
e
t
a
l
.
[
19
]
2
0
1
9
2
.
8
1
P
r
o
p
o
se
d
me
t
h
o
d
-
3
.
2
6
T
ab
le
3
.
C
o
m
p
ar
is
o
n
o
f
g
e
n
d
er
ac
cu
r
ac
y
w
it
h
th
e
s
tate
-
of
-
t
h
e
-
ar
t
m
et
h
o
d
s
(
%)
M
e
t
h
o
d
Y
e
a
r
A
c
c
u
r
a
c
y
D
u
a
n
e
t
a
l
.
[
8
]
2
0
1
7
8
8
.
2
0
G
u
o
e
t
a
l
.
[
1
0
]
2
0
1
4
9
8
.
4
0
D
h
o
mn
e
e
t
a
l
.
[
1
6
]
2
0
1
8
9
5
.
0
0
S
r
i
n
i
v
a
s
e
t
a
l
.
[
1
7
]
2
0
1
7
8
4
.
7
0
L
e
e
e
t
a
l
.
[
20
]
2
0
1
7
8
8
.
5
0
H
u
a
n
g
e
t
a
l
.
[
21
]
2
0
1
7
8
9
.
6
0
B
e
n
i
n
i
e
t
a
l
.
[2
2
]
2
0
1
9
9
8
.
5
9
F
a
n
g
e
t
a
l
.
[2
3
]
2
0
1
9
9
8
.
8
0
P
r
o
p
o
se
d
me
t
h
o
d
-
9
5
.
0
0
T
ab
le
4
.
A
cc
u
r
ac
y
o
f
et
h
n
icit
y
r
ec
o
g
n
itio
n
f
o
r
MO
R
P
H
I
I
d
a
taset (
%)
M
e
t
h
o
d
Y
e
a
r
A
c
c
u
r
a
c
y
G
u
o
e
t
a
l
[
1
0
]
2
0
1
4
9
9
.
0
0
U
d
d
i
n
e
t
a
l
[
1
3
]
2
0
1
6
9
5
.
4
0
S
r
i
n
i
v
a
s
e
t
a
l
[
1
7
]
2
0
1
7
3
3
.
3
3
M
o
h
a
mm
e
d
e
t
a
l
[
18
]
2
0
1
9
9
3
.
3
H
o
c
q
u
e
t
e
t
a
l
[
2
4
]
2
0
1
6
9
7
.
5
0
M
o
h
a
mm
e
d
e
t
a
l
[
2
5
]
2
0
1
7
9
4
.
6
0
P
r
o
p
o
se
d
me
t
h
o
d
-
9
8
.
2
0
I
n
o
u
r
s
t
u
d
y
,
w
e
clas
s
i
f
ied
g
e
n
d
er
w
i
th
a
n
d
w
it
h
o
u
t
et
h
n
icit
y
co
n
s
id
er
atio
n
.
T
h
e
ac
cu
r
ac
y
o
f
g
e
n
d
er
r
ec
o
g
n
itio
n
w
it
h
o
u
t
et
h
n
icit
y
co
n
s
id
er
atio
n
w
it
h
t
h
e
DS
S
AE
is
9
5
%,
th
is
ac
c
u
r
ac
y
is
co
n
s
id
er
ed
an
e
x
citi
n
g
r
esu
lt
f
o
r
t
w
o
r
ea
s
o
n
s
.
Firstl
y
,
th
e
s
tate
o
f
ar
t
s
d
em
o
n
s
tr
ate
s
th
at
th
i
s
r
ate
is
s
ig
n
i
f
ica
n
t
in
co
m
p
ar
is
o
n
w
it
h
o
th
er
m
e
th
o
d
s
,
s
u
c
h
as
m
e
n
ti
o
n
ed
in
T
ab
le
3
.
Seco
n
d
ly
,
d
u
e
to
th
e
co
m
p
o
s
itio
n
o
f
MO
R
P
H
I
I
b
en
ch
m
ar
k
,
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
S
emi
-
s
u
p
ervis
ed
a
u
to
-
en
c
o
r
d
e
r
fo
r
fa
cia
l a
ttr
ib
u
tes r
ec
o
g
n
itio
n
(
S
o
u
ma
y
a
Za
g
h
b
a
n
i
)
2175
th
is
d
atab
ase
co
n
tain
s
f
ac
ial
i
m
ag
e
s
o
f
d
if
f
er
e
n
t
p
eo
p
le
w
it
h
d
if
f
er
e
n
t
lo
o
k
s
s
u
c
h
as
a
m
an
w
i
th
lo
n
g
h
air
f
e
m
ales
w
it
h
s
h
o
r
t
h
air
.
T
ab
l
e
5
s
h
o
w
s
th
e
co
n
f
u
s
io
n
m
at
r
ix
o
f
g
e
n
d
er
r
ec
o
g
n
itio
n
w
it
h
co
n
s
id
er
atio
n
o
f
eth
n
icit
y
.
Facial
i
m
ag
e
s
w
er
e
d
ev
is
ed
i
n
to
f
o
u
r
clas
s
es;
NC
F,
C
F,
NC
F
an
d
C
M.
t
h
e
ac
c
u
r
ac
y
i
n
cr
ea
s
ed
b
y
ab
o
u
t
2
%
in
co
m
p
ar
is
o
n
w
it
h
g
e
n
d
er
r
ec
o
g
n
itio
n
w
it
h
o
u
t
e
th
n
icit
y
co
n
s
id
er
atio
n
.
Fo
r
eth
n
ic
it
y
es
ti
m
atio
n
,
w
e
o
n
l
y
r
ep
o
r
ted
ac
cu
r
ac
ies
f
o
r
th
e
C
au
ca
s
ia
n
an
d
No
t
C
au
ca
s
ian
,
s
i
n
ce
o
th
er
r
ac
e
g
r
o
u
p
s
lik
e
Asi
a
n
an
d
I
n
d
ian
r
ac
e
w
er
e
n
o
t
u
s
ed
i
n
tr
ai
n
in
g
b
ec
au
s
e
t
h
e
n
u
m
b
er
o
f
i
m
ag
e
s
is
v
er
y
s
m
all.
T
h
e
clas
s
i
f
icati
o
n
g
r
o
u
p
s
i
n
to
t
w
o
eth
n
icit
y
c
lass
e
s
.
T
h
e
ac
cu
r
ac
y
o
b
tain
ed
is
9
8
.
2
%;
th
i
s
r
ate
is
in
ter
e
s
tin
g
i
n
co
m
p
ar
is
o
n
w
it
h
o
t
h
er
m
et
h
o
d
s
as d
escr
ib
ed
in
T
ab
le
4
.
I
n
th
i
s
ar
ticle,
w
e
u
s
ed
th
e
DSS
A
E
as
a
cla
s
s
i
f
ier
;
w
e
m
ea
s
u
r
ed
th
e
ac
c
u
r
ac
y
t
h
at
o
cc
u
r
s
w
h
e
n
a
class
i
f
ier
is
te
s
ted
w
it
h
d
if
f
e
r
en
t
h
id
d
en
la
y
er
s
.
T
h
e
p
ar
a
m
eter
s
o
f
th
e
d
ee
p
n
e
u
r
al
n
e
t
w
o
r
k
ar
e
in
v
esti
g
ated
b
y
alter
i
n
g
th
e
n
u
m
b
er
o
f
h
id
d
en
la
y
er
s
,
t
h
e
n
u
m
b
er
o
f
n
e
u
r
o
n
s
an
d
t
h
e
s
ize
o
f
th
e
tr
ai
n
in
g
s
et.
W
e
ca
r
r
ied
o
u
t
ex
ten
s
i
v
e
ex
p
er
i
m
e
n
ts
to
d
eter
m
i
n
e
th
e
o
p
ti
m
u
m
p
ar
a
m
eter
s
.
T
h
e
n
u
m
b
er
o
f
la
y
er
s
in
th
e
DNN
is
cr
u
cial
f
o
r
ab
o
u
t
1
5
.
0
0
0
im
a
g
e
s
.
I
n
th
is
ar
ticle,
w
e
u
s
ed
t
w
o
m
o
d
els
w
it
h
a
d
if
f
er
en
t
n
u
m
b
er
o
f
h
i
d
d
en
lay
er
s
,
an
d
w
e
ex
a
m
in
e
t
h
e
f
in
al
r
es
u
lt
s
:
w
e
h
av
e
th
e
f
ir
s
t
m
o
d
el
ca
lled
m
o
d
1
:
w
e
h
a
v
e
t
w
o
h
id
d
en
la
y
er
s
an
d
th
e
s
ec
o
n
d
o
n
e
ca
lled
m
o
d
2
w
it
h
T
h
r
ee
lay
er
s
.
T
h
e
r
esu
lts
ar
e
s
u
m
m
ar
ized
in
T
ab
le
6
.
T
ab
le
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
g
en
d
er
th
r
o
w
s
et
h
n
icit
y
r
ec
o
g
n
itio
n
(
MO
R
P
H
I
I
)
NC
F
CF
N
C
M
CM
N
o
t
C
a
u
c
a
si
a
n
F
e
mal
e
9
2
.
4
2
.
0
4
.
2
2
.
0
C
a
u
c
a
s
i
a
n
F
e
mal
e
2
.
1
9
4
.
6
0
.
0
3
.
2
N
o
t
C
a
u
c
a
si
a
n
M
a
l
e
3
.
9
0
.
0
9
1
.
5
0
.
0
C
a
u
c
a
s
i
a
n
M
a
l
e
1
.
6
3
.
4
4
.
3
9
4
.
8
T
ab
le
6
.
P
ar
am
eter
s
u
s
ed
f
o
r
t
h
e
DSS
A
E
ar
c
h
itect
u
r
e
P
a
r
a
me
t
e
r
s
L
a
y
e
r
s
si
z
e
R
e
g
u
l
a
r
i
z
a
t
i
o
n
t
e
r
m (λ
)
S
p
a
r
si
t
y
p
a
r
a
me
t
e
r
(
ρ
)
W
e
i
g
h
t
s
p
a
r
s
i
t
y
p
e
n
a
l
t
y
(
β
)
A
c
c
u
r
a
c
y
A
g
e
,
g
e
n
d
e
r
a
n
d
e
t
h
n
i
c
i
t
y
j
o
i
n
t
l
y
L
a
y
e
r
1
50
0
.
0
0
2
0
.
5
5
7
3
.
5
%
L
a
y
e
r
2
25
0
.
0
0
3
0
.
8
5
T
h
e
p
r
esen
ted
w
o
r
k
s
h
o
w
ed
an
i
m
p
r
o
v
e
m
en
t
p
er
f
o
r
m
a
n
ce
u
s
i
n
g
DS
S
A
E
.
T
h
is
p
er
f
o
r
m
a
n
ce
i
s
ex
p
lain
ed
b
y
t
h
e
t
w
o
i
m
p
o
r
ta
n
t
m
e
tr
ics:
f
ir
s
tl
y
;
t
h
e
u
s
e
o
f
s
u
p
er
v
i
s
io
n
u
n
d
er
clas
s
es
i
m
p
r
o
v
ed
th
e
ac
cu
r
ac
y
.
R
es
u
lts
i
n
o
u
r
w
o
r
k
an
d
in
l
iter
atu
r
e
s
h
o
w
n
th
a
t
u
s
i
n
g
th
e
A
E
is
v
er
y
in
ter
es
tin
g
i
n
s
u
p
er
v
is
io
n
m
a
n
n
er
.
T
h
e
s
ec
o
n
d
m
etr
ic
is
th
e
u
s
e
o
f
L
1
an
d
L
2
n
o
r
m
to
r
eg
u
lar
i
ze
th
e
m
o
d
el.
T
h
is
m
etr
ic
h
el
p
ed
u
s
to
r
eg
u
lar
ize
th
e
m
o
d
el
an
d
r
ed
u
ce
th
e
o
v
er
f
it
tin
g
p
r
o
b
le
m
.
I
n
o
th
er
h
an
d
,
t
h
e
s
o
lu
tio
n
u
s
ed
i
n
t
h
is
w
o
r
k
,
e
n
h
a
n
ce
s
th
e
s
p
ar
s
it
y
i
n
e
v
er
y
cla
s
s
a
n
d
co
n
s
eq
u
e
n
tl
y
it
i
m
p
r
o
v
e
s
th
e
g
en
er
aliza
t
io
n
o
f
f
ea
t
u
r
es
f
o
r
ev
er
y
clas
s
.
T
h
e
m
aj
o
r
p
r
o
b
lem
en
co
u
n
ter
ed
in
t
h
i
s
w
o
r
k
is
th
e
n
u
m
b
er
o
f
s
a
m
p
les
u
s
ed
u
n
d
er
e
v
er
y
c
las
s
,
b
ec
au
s
e
d
iv
id
i
n
g
s
a
m
p
les b
et
w
ee
n
t
h
e
th
r
ee
attr
ib
u
tes
w
il
l d
ec
r
ea
s
e
th
e
to
tal
n
u
m
b
er
i
n
ev
er
y
cla
s
s
.
T
h
is
ex
p
lai
n
th
e
lo
w
r
ate
f
o
u
n
d
in
eld
er
l
y
clas
s
w
h
er
e
w
e
h
av
e
v
er
y
lo
w
n
u
m
b
er
o
f
s
a
m
p
les
m
a
k
e
th
e
m
o
d
el
u
n
ab
le
to
ex
tr
ac
t
m
o
r
e
f
ea
tu
r
es
an
d
g
e
n
er
alize
.
T
h
is
w
h
y
w
e
ad
d
ed
s
am
p
les
in
eld
er
l
y
class
es
as
ex
p
lain
ed
i
n
s
ec
tio
n
(
4
.
1
)
an
d
th
e
ac
c
u
r
ac
y
w
as
in
cr
ea
s
ed
f
r
o
m
5
2
.
2
5
%
to
6
3
.
5
7
%
(
N
C
M
5
0
+)
.
I
t
ca
n
b
e
ex
p
lain
ed
as
k
i
n
d
o
f
d
ata
au
g
m
en
tatio
n
m
etr
ic.
I
n
th
e
y
o
u
n
g
an
d
s
en
io
r
clas
s
es
w
e
h
a
v
e
f
o
u
n
d
a
r
ate
f
o
r
m
o
r
e
th
an
8
0
%
f
o
r
all
attr
ib
u
tes.
I
n
t
h
ese
cla
s
s
e
s
w
e
h
av
e
lar
g
e
n
u
m
b
er
o
f
s
a
m
p
les
p
er
f
o
r
m
t
h
e
m
o
d
el
to
g
e
n
er
alize
.
B
u
t in
co
m
p
ar
is
o
n
o
f
t
h
ese
r
at
es
w
ith
f
o
u
n
d
r
esu
lts
f
o
r
f
ac
ial
attr
ib
u
tes s
ep
ar
atel
y
is
m
o
r
e
i
n
ter
esti
n
g
.
5.
CO
NCLU
SI
O
N
I
n
th
is
ar
ticle,
w
e
p
r
esen
t
a
m
et
h
o
d
f
o
r
A
GE
.
W
e
u
s
ed
au
to
en
co
d
er
s
f
o
r
class
i
f
icatio
n
.
T
h
e
w
o
r
k
co
n
s
is
ts
o
f
u
s
i
n
g
f
ac
e
s
f
r
o
m
MO
R
P
H
I
I
d
atab
ases
to
r
ec
o
g
n
ize
A
GE
j
o
in
tl
y
b
y
clas
s
i
f
y
i
n
g
f
ac
ial
i
m
a
g
es
i
n
to
1
2
class
es
to
f
i
n
d
th
r
ee
d
e
m
o
g
r
ap
h
ic
attr
ib
u
te
s
(
ag
e,
g
e
n
d
er
an
d
eth
n
icit
y
)
.
T
h
e
clas
s
i
f
icatio
n
m
o
d
el
w
as
b
ased
o
n
u
p
d
ati
n
g
v
er
s
io
n
o
f
a
u
to
en
co
d
er
ca
lled
DSS
A
E
.
I
n
th
i
s
m
o
d
el
w
e
ar
e
tr
y
i
n
g
to
ex
p
lo
i
t
th
e
ad
v
a
n
tag
e
s
o
f
s
u
p
er
v
i
s
ed
an
d
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
in
t
h
e
s
a
m
e
t
i
m
e.
T
h
e
ex
p
er
i
m
e
n
ts
ar
e
co
n
d
u
ct
ed
o
n
an
e
x
te
n
s
iv
e
d
atab
ase
co
n
tain
in
g
m
o
r
e
th
an
5
5
,
0
0
0
f
ac
e
im
a
g
es.
A
n
d
th
e
y
s
h
o
w
th
e
r
o
b
u
s
t
n
es
s
o
f
o
u
r
m
et
h
o
d
as
class
i
f
icatio
n
m
o
d
el
to
f
i
n
d
th
e
th
r
ee
at
tr
ib
u
tes
s
ep
ar
atel
y
.
RE
F
E
R
E
NC
E
S
[1
]
M
.
Jo
rd
i
e
t
a
l.
,
“
L
o
c
a
l
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
g
e
n
d
e
r
re
c
o
g
n
it
io
n
,
”
Pa
tt
e
rn
Rec
o
g
n
i
ti
o
n
L
e
tt
e
rs
.,
v
o
l
7
0
,
p
p
.
8
0
-
8
6
,
Ja
n
u
a
ry
2
0
1
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
18
,
No
.
4
,
A
u
g
u
s
t 2
0
2
0
:
2
1
6
9
-
2
1
7
6
2176
[2
]
H.
Ka
y
a
,
e
t
a
l.
,
“
E
m
o
ti
o
n
,
a
g
e
,
a
n
d
g
e
n
d
e
r
c
las
sif
ic
a
ti
o
n
in
c
h
il
d
re
n
’s
sp
e
e
c
h
b
y
h
u
m
a
n
s
a
n
d
m
a
c
h
i
n
e
s,
”
Co
mp
u
ter
S
p
e
e
c
h
&
L
a
n
g
u
a
g
e
,
v
o
l
.
4
6
,
p
p
.
2
6
8
-
2
8
3
,
No
v
e
m
b
e
r
2
0
1
7
.
[3
]
G
.
A
n
ti
p
o
v
,
e
t
a
l.
,
“
Ef
fe
c
ti
v
e
train
in
g
o
f
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
t
w
o
rk
s
f
o
r
fa
c
e
-
b
a
se
d
g
e
n
d
e
r
a
n
d
a
g
e
p
re
d
ictio
n
,
”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l
.
7
2
,
p
p
.
1
5
-
2
6
,
De
c
e
m
b
e
r
2
0
1
7
.
[4
]
L
.
Ca
i,
e
t
a
l.
,
“
HOG
-
a
ss
isted
d
e
e
p
f
e
a
tu
re
lea
rn
in
g
f
o
r
p
e
d
e
strian
g
e
n
d
e
r
re
c
o
g
n
it
i
o
n
,
”
J
o
u
rn
a
l
o
f
th
e
Fra
n
k
li
n
In
stit
u
te
,
v
o
l.
3
5
5
,
n
o
.
4
,
p
p
.
1
9
9
1
-
2
0
0
8
,
M
a
rc
h
2
0
1
8
.
[
5
]
P
.
V
i
o
l
a
,
e
t
a
l
.
,
“
R
o
b
u
s
t
r
e
a
l
-
t
i
m
e
f
a
c
e
d
e
t
e
c
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
V
i
s
i
o
n
,
v
o
l
.
5
7
,
p
p
.
1
3
7
-
1
5
4
,
2
0
0
4
.
[6
]
S
.
Zag
h
b
a
n
i
e
t
a
l.
,
“
A
g
e
Esti
m
a
ti
o
n
u
si
n
g
De
e
p
L
e
a
rn
in
g
,
”
Co
mp
u
ter
s
&
El
e
c
trica
l
En
g
i
n
e
e
rin
g
,
v
o
l.
6
8
,
p
p
.
3
3
7
-
3
4
7
,
M
a
y
2
0
1
8
.
[7
]
M
.
S
i
n
g
h
,
e
t
a
l
.,
“
A
re
y
o
u
e
li
g
ib
le?
P
re
d
icti
n
g
a
d
u
lt
h
o
o
d
f
ro
m
fa
c
e
i
m
a
g
e
s
v
ia
c
las
s
sp
e
c
i
f
ic
m
e
a
n
a
u
to
e
n
c
o
d
e
r,
”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
L
e
tt
e
rs
,
v
o
l
.
1
1
9
,
p
p
.
1
2
1
-
1
3
0
,
M
a
rc
h
2
0
1
9
.
[8
]
M
.
Du
a
n
,
e
t
a
l.
,
“
A
H
y
b
rid
De
e
p
L
e
a
rn
in
g
CNN
-
E
L
M
f
o
r
A
g
e
a
n
d
G
e
n
d
e
r
Clas
siÞ
c
a
ti
o
n
,
”
N
e
u
ro
c
o
mp
u
ti
n
g
,
v
o
l.
2
7
5
,
p
p
.
4
4
8
-
4
6
1
,
Ja
n
u
a
ry
2
0
1
8
.
[9
]
S
.
Zag
h
b
a
n
i
e
t
a
l.
,
“
F
a
c
ial
Em
o
ti
o
n
Re
c
o
g
n
it
i
o
n
f
o
r
A
d
a
p
ti
v
e
In
t
e
rfa
c
e
s
Us
in
g
W
rin
k
les
a
n
d
W
a
v
e
let
Ne
tw
o
rk
,
”
2
0
1
7
IEE
E
/A
CS
1
4
t
h
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Co
m
p
u
ter
S
y
ste
ms
a
n
d
A
p
p
li
c
a
ti
o
n
s
(
AICCS
A)
,
p
p
.
3
4
2
-
3
4
9
,
2
0
1
7
.
[1
0
]
G
.
G
u
o
,
e
t
a
l.
,
”
A
F
ra
m
e
w
o
rk
f
o
r
Jo
in
t
Esti
m
a
ti
o
n
o
f
A
g
e
,
G
e
n
d
e
r
a
n
d
E
th
n
icity
o
n
A
L
a
r
g
e
Da
tab
a
se
,
”
Ima
g
e
a
n
d
Vi
sio
n
Co
mp
u
ti
n
g
,
v
o
l.
3
2
,
n
o
.
1
0
,
p
p
.
7
6
1
-
7
7
0
,
Oc
t
o
b
e
r
2
0
1
4
.
[1
1
]
H.
G
a
o
,
e
t
a
l.
,
“
S
in
g
le
S
a
m
p
le
F
a
c
e
Re
c
o
g
n
it
io
n
v
ia
L
e
a
rn
in
g
De
e
p
S
u
p
e
rv
ise
d
A
u
to
e
n
c
o
d
e
rs,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
I
n
fo
rm
a
ti
o
n
F
o
re
n
sic
s A
n
d
S
e
c
u
rity
,
v
o
l.
1
0
,
no
.
1
0
,
p
p
.
2
1
0
8
-
2
1
1
8
,
Oc
t
2
0
1
5
.
[1
2
]
S
.
Zag
h
b
a
n
i
,
e
t
a
l.
,
“
Re
a
l
ti
m
e
h
a
n
d
g
e
stu
re
re
c
o
g
n
it
io
n
u
si
n
g
f
e
a
t
u
re
s
e
x
trac
ti
o
n
,
”
In
ter
n
a
t
io
n
a
l
c
o
n
fer
e
n
c
e
ICM
V
(
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
Vi
si
o
n
)
,
2
0
1
7
.
[1
3
]
A
.
M
a
ju
m
d
a
r,
e
t
a
l.
,
“
F
a
c
e
V
e
rif
ica
ti
o
n
v
ia
c
las
s
sp
a
rsit
y
b
a
se
d
su
p
e
rv
ise
d
e
n
c
o
d
in
g
,
”
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
r
n
An
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
tell
ig
e
n
c
e
,
v
o
l.
3
9
,
n
o
.
6
,
p
p
.
1
2
7
3
-
1
2
8
0
,
J
u
n
e
2
0
1
7
.
[1
4
]
K.
Jh
o
n
y
,
e
t
a
l.
,
"
A
F
lex
ib
le
Hie
ra
rc
h
ica
l
A
p
p
ro
a
c
h
F
o
r
F
a
c
ial
A
g
e
Esti
m
a
ti
o
n
Ba
se
d
o
n
M
u
lt
ip
le
F
e
a
tu
re
s,"
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
5
4
,
p
p
.
3
4
-
5
1
,
J
u
n
e
2
0
1
6
.
[1
5
]
K.
L
i,
e
t
a
l.
,
“
D2
C:
De
e
p
c
u
m
u
lat
iv
e
l
y
a
n
d
c
o
m
p
a
r
a
ti
v
e
l
y
le
a
rn
in
g
f
o
r
h
u
m
a
n
a
g
e
e
sti
m
a
ti
o
n
,
”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
6
6
,
p
p
.
9
5
-
1
0
5
,
Ju
n
e
2
0
1
7
.
[1
6
]
A
.
Dh
o
m
n
e
,
e
t
a
l.
,
"
G
e
n
d
e
r
re
c
o
g
n
it
io
n
th
ro
u
g
h
f
a
c
e
u
sin
g
d
e
e
p
l
e
a
rn
in
g
,
"
Pro
c
e
d
ia
C
o
mp
u
ter
S
c
i
e
n
c
e
,
v
o
l.
1
3
2
,
p
p
.
2
-
1
0
,
2
0
1
8
.
[1
7
]
N.
S
ri
n
iv
a
s,
e
t
a
l.
,
“
A
g
e
,
Ge
n
d
e
r,
a
n
d
F
i
n
e
-
G
ra
in
e
d
e
th
n
icity
p
re
d
ictio
n
u
si
n
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
th
e
e
a
st
a
sia
n
f
a
c
e
d
a
tas
e
t
,
”
I
EE
E
1
2
th
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
A
u
t
o
ma
ti
c
F
a
c
e
&
Ge
stu
re
Rec
o
g
n
it
i
o
n
,
p
p
.
9
5
3
-
9
6
0
,
2
0
1
7
.
[1
8
]
N.
M
o
h
a
m
m
e
d
,
e
t
a
l.
,
“
T
w
o
-
sta
g
e
s b
a
se
d
f
a
c
ial
d
e
m
o
g
ra
p
h
ic attri
b
u
tes
c
o
m
b
in
a
ti
o
n
f
o
r
a
g
e
e
stim
a
ti
o
n
,
“
J
o
u
rn
a
l
o
f
Vi
su
a
l
Co
mm
u
n
ica
ti
o
n
a
n
d
Im
a
g
e
Rep
re
se
n
ta
ti
o
n
,
v
o
l
.
6
1
,
p
p
.
2
3
6
-
2
4
9
,
M
a
y
2
0
1
9
.
[1
9
]
S
.
T
a
h
e
ri
,
e
t
a
l.
,
"
On
th
e
u
se
o
f
D
AG
-
CNN
a
r
c
h
it
e
c
tu
re
f
o
r
a
g
e
e
stim
a
ti
o
n
w
it
h
m
u
lt
i
-
sta
g
e
f
e
a
tu
re
s
f
u
sio
n
,
"
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
3
2
9
,
p
p
.
3
0
0
-
3
1
0
,
F
e
b
r
u
a
ry
2019.
[2
0
]
S.
H.
L
e
e
,
e
t
a
l.
,
“
A
g
e
a
n
d
g
e
n
d
e
r
e
sti
m
a
ti
o
n
u
sin
g
d
e
e
p
re
sid
u
a
l
l
e
a
rn
in
g
n
e
tw
o
rk
,
”
2
0
1
8
In
ter
n
a
ti
o
n
a
l
W
o
rk
sh
o
p
o
n
A
d
v
a
n
c
e
d
Ima
g
e
T
e
c
h
n
o
lo
g
y
(
IW
AIT
)
,
p
p
.
1
-
3
,
2
0
1
8
.
[2
1
]
C.
Hu
a
n
g
,
e
t
a
l.
,
"
Ag
e
/
g
e
n
d
e
r
c
las
sif
ic
a
ti
o
n
w
it
h
w
h
o
le
-
c
o
m
p
o
n
e
n
t
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
,
"
An
n
u
a
l
S
u
mm
it
a
n
d
Co
n
fer
e
n
c
e
,
p
p
.
1
2
8
2
-
1
2
8
5
,
2
0
1
7
.
[2
2
]
S
.
Be
n
i
n
i
,
e
t
a
l.
,
"
F
a
c
e
a
n
a
l
y
sis
t
h
ro
u
g
h
se
m
a
n
ti
c
f
a
c
e
se
g
m
e
n
tatio
n
,
"
S
i
g
n
a
l
Pr
o
c
e
ss
in
g
:
Ima
g
e
C
o
mm
u
n
ic
a
ti
o
n
,
v
o
l
7
4
,
p
p
.
2
1
-
3
1
,
M
a
y
2
0
1
9
.
[2
3
]
J.
F
a
n
,
e
t
a
l
.
,
"
M
u
ti
-
sta
g
e
lea
rn
in
g
f
o
r
g
e
n
d
e
r
a
n
d
a
g
e
p
re
d
icti
o
n
,
"
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
ol
.
3
3
4
,
p
p
.
1
1
4
-
1
2
4
,
M
a
rc
h
2
0
1
9
.
[2
4
]
G
.
P
.
M
a
b
u
z
a
a
-
Ho
c
q
u
e
t,
e
t
a
l.
,
“
Et
h
n
ici
ty
p
re
d
ictio
n
a
n
d
c
las
sif
ica
ti
o
n
f
ro
m
iri
s
tex
tu
re
p
a
tt
e
rn
s:
a
su
rv
e
y
o
n
re
c
e
n
t
a
d
v
a
n
c
e
s
,
”
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
C
o
mp
u
ta
ti
o
n
a
l
S
c
ien
c
e
a
n
d
Co
m
p
u
t
a
ti
o
n
a
l
In
telli
g
e
n
c
e
,
p
p
.
8
1
8
-
8
2
3
,
2
0
1
6
.
[2
5
]
A
.
S
.
M
o
h
a
m
m
a
d
,
e
t
a
l.
,
“
T
o
w
a
rd
s
Et
h
n
icity
De
tec
ti
o
n
u
sin
g
L
e
a
rn
in
g
Ba
se
d
Clas
sif
iers
,
”
2
0
1
7
9
th
C
o
mp
u
ter
S
c
ien
c
e
a
n
d
El
e
c
tro
n
ic E
n
g
i
n
e
e
rin
g
(
CEE
C)
,
p
p
.
2
1
9
-
2
2
4
,
2
0
1
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.