I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
1
6
,
No
.
1
,
Feb
r
u
ar
y
20
2
6
,
p
p
.
395
~
403
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
1
.
pp
3
9
5
-
4
0
3
395
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Facia
l emo
tion re
co
g
nition under
f
a
ce mas
k occlus
io
n using
v
isio
n t
ra
nsfo
rme
rs
Ash
ra
f
Yuni
s
M
a
g
ha
ri,
Am
ee
r
M
.
T
elba
ni
F
a
c
u
l
t
y
o
f
I
n
f
o
r
ma
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
I
sl
a
m
i
c
U
n
i
v
e
r
s
i
t
y
o
f
G
a
z
a
,
G
a
z
a
,
P
a
l
e
st
i
n
e
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
2
2
,
2
0
2
5
R
ev
is
ed
Sep
2
4
,
2
0
2
5
Acc
ep
ted
No
v
2
3
,
2
0
2
5
F
a
c
ial
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
(
F
ER
)
sy
ste
m
s
fa
c
e
sig
n
ifi
c
a
n
t
c
h
a
ll
e
n
g
e
s
wh
e
n
in
d
i
v
id
u
a
ls
we
a
r
fa
c
e
m
a
sk
s,
a
s
c
rit
ica
l
fa
c
ial
re
g
io
n
s
a
re
o
c
c
l
u
d
e
d
.
Th
is
p
a
p
e
r
a
d
d
re
ss
e
s
th
is
li
m
it
a
ti
o
n
b
y
e
m
p
lo
y
in
g
v
isi
o
n
tran
sf
o
rm
e
rs
(ViT)
,
wh
ich
o
ffe
r
a
p
ro
m
isin
g
a
lt
e
r
n
a
ti
v
e
with
re
d
u
c
e
d
c
o
m
p
u
tatio
n
a
l
c
o
m
p
lex
it
y
c
o
m
p
a
re
d
to
trad
it
io
n
a
l
d
e
e
p
le
a
rn
in
g
m
e
th
o
d
s.
We
p
r
o
p
o
se
a
ViT
-
b
a
se
d
F
ER
fra
m
e
wo
rk
th
a
t
fi
n
e
-
tu
n
e
s
a
p
re
-
train
e
d
ViT
a
rc
h
it
e
c
tu
re
t
o
e
n
h
a
n
c
e
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
u
n
d
e
r
m
a
sk
-
in
d
u
c
e
d
o
c
c
lu
sio
n
.
T
h
e
m
o
d
e
l
is
fin
e
-
tu
n
e
d
a
n
d
e
v
a
lu
a
ted
o
n
t
h
e
Affe
c
tNe
t
d
a
tas
e
t,
wh
ich
o
rig
in
a
ll
y
re
p
re
se
n
ts
e
ig
h
t
e
m
o
ti
o
n
c
a
teg
o
r
ies
.
Th
e
se
c
a
te
g
o
ries
a
re
re
stru
c
tu
re
d
in
to
fi
v
e
b
ro
a
d
e
r
c
las
se
s
to
m
it
ig
a
te
th
e
imp
a
c
t
o
f
o
c
c
lu
d
e
d
fe
a
tu
re
s.
Th
e
m
o
d
e
l’s
p
e
rfo
rm
a
n
c
e
is
a
ss
e
ss
e
d
u
si
n
g
sta
n
d
a
r
d
m
e
tri
c
s,
i
n
c
lu
d
in
g
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
sc
o
re
.
Ex
p
e
rime
n
tal
re
su
lt
s
d
e
m
o
n
stra
t
e
th
a
t
t
h
e
p
ro
p
o
se
d
fra
m
e
wo
rk
a
c
h
iev
e
s
a
n
a
c
c
u
ra
c
y
o
f
8
1
%
,
o
u
t
p
e
rfo
rm
i
n
g
se
v
e
ra
l
sta
te
-
of
-
th
e
-
a
rt
a
p
p
ro
a
c
h
e
s.
Th
e
s
e
fin
d
in
g
s
h
ig
h
li
g
h
t
t
h
e
p
o
ten
t
ial
o
f
v
isi
o
n
tran
sfo
rm
e
r
s
in
re
c
o
g
n
izin
g
e
m
o
t
io
n
s
u
n
d
e
r
m
a
sk
e
d
c
o
n
d
it
i
o
n
s
a
n
d
su
p
p
o
rt
th
e
d
e
v
e
l
o
p
m
e
n
t
o
f
m
o
re
r
o
b
u
st
F
ER
sy
ste
m
s
fo
r
re
a
l
-
wo
rl
d
a
p
p
li
c
a
ti
o
n
s
in
h
e
a
lt
h
c
a
re
,
su
rv
e
il
lan
c
e
,
a
n
d
h
u
m
a
n
–
c
o
m
p
u
ter
i
n
tera
c
ti
o
n
.
Th
is
wo
rk
in
tro
d
u
c
e
s
a
sc
a
lab
le
a
n
d
e
ffe
c
ti
v
e
a
p
p
ro
a
c
h
t
h
a
t
in
teg
ra
tes
se
lf
-
a
tt
e
n
ti
o
n
,
sy
n
th
e
ti
c
m
a
sk
a
u
g
m
e
n
tati
o
n
,
a
n
d
e
m
o
t
io
n
c
las
s
re
stru
c
tu
ri
n
g
t
o
imp
r
o
v
e
e
m
o
ti
o
n
re
c
o
g
n
it
io
n
u
n
d
e
r
fa
c
ial
o
c
c
lu
sio
n
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
E
m
o
tio
n
r
ec
o
g
n
itio
n
Face
m
ask
Occ
lu
s
io
n
h
an
d
lin
g
Vis
io
n
tr
an
s
f
o
r
m
er
s
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
:
Ash
r
af
Yu
n
is
Ma
g
h
ar
i
Facu
lty
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
I
s
lam
ic
Un
iv
er
s
ity
o
f
G
az
a
P.O.
B
o
x
1
0
8
,
Gaz
a,
Palest
in
e
E
m
ail: a
m
ag
h
ar
i@
iu
g
az
a.
ed
u
.
p
s
1.
I
NT
RO
D
UCT
I
O
N
Facial
ex
p
r
ess
io
n
s
ar
e
f
u
n
d
a
m
en
tal
to
h
u
m
an
c
o
m
m
u
n
ic
atio
n
an
d
em
o
tio
n
u
n
d
er
s
tan
d
in
g
[
1
]
.
R
ec
en
t
ad
v
an
ce
s
in
co
m
p
u
ter
v
is
io
n
h
av
e
in
teg
r
ated
f
ac
ial
ex
p
r
ess
io
n
s
in
to
m
an
y
ap
p
licatio
n
s
lik
e
v
ir
tu
al
r
ea
lity
(
VR
)
an
d
au
g
m
en
ted
r
e
ality
(
AR
)
,
s
ec
u
r
ity
s
y
s
tem
s
,
an
d
h
u
m
an
-
c
o
m
p
u
ter
in
ter
ac
tio
n
(
HC
I
)
[
2
]
.
Facial
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
(
FER
)
s
y
s
tem
s
clas
s
if
y
em
o
tio
n
s
s
u
ch
as
h
ap
p
in
ess
,
s
ad
n
ess
,
an
g
er
,
f
ea
r
,
s
u
r
p
r
is
e,
an
d
d
is
g
u
s
t
f
r
o
m
s
tatic
im
ag
es
an
d
v
id
eo
s
tr
ea
m
s
[
3
]
.
FER
m
o
d
els
ar
e
tr
ain
ed
o
n
n
u
m
er
o
u
s
im
ag
es
co
n
v
ey
in
g
em
o
tio
n
s
lik
e
h
ap
p
in
ess
,
s
ad
n
ess
,
an
g
er
,
f
ea
r
,
s
u
r
p
r
is
e,
an
d
d
is
g
u
s
t.
Ho
wev
er
,
em
o
tio
n
e
x
p
r
ess
io
n
v
ar
ies
d
u
e
to
ag
e,
cu
ltu
r
e,
an
d
g
en
d
er
[
4
]
,
r
aisi
n
g
ac
cu
r
ac
y
ch
allen
g
es a
n
d
eth
ical
co
n
ce
r
n
s
.
T
h
i
s
s
t
u
d
y
a
i
m
s
t
o
e
n
h
a
n
c
e
FE
R
u
n
d
e
r
m
a
s
k
-
i
n
d
u
c
e
d
o
c
c
l
u
s
i
o
n
,
u
s
i
n
g
a
v
i
s
i
o
n
t
r
a
n
s
f
o
r
m
e
r
s
(
V
i
T
)
b
a
s
e
d
m
o
d
e
l
t
r
a
i
n
e
d
o
n
s
y
n
t
h
et
i
c
a
ll
y
m
a
s
k
e
d
d
a
t
a
.
T
r
a
d
i
t
i
o
n
a
l
F
E
R
r
el
i
es
o
n
m
a
n
u
a
l
f
e
a
t
u
r
e
e
n
g
i
n
e
e
r
i
n
g
w
it
h
p
r
e
p
r
o
c
e
s
s
i
n
g
,
f
ea
t
u
r
e
e
x
t
r
ac
t
i
o
n
,
a
n
d
cl
a
s
s
i
f
ic
a
t
i
o
n
[
2
]
,
w
h
i
l
e
d
e
e
p
l
ea
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
s
u
c
h
a
s
d
e
e
p
b
e
l
i
e
f
n
e
t
w
o
r
k
s
(
DB
Ns
)
,
l
o
n
g
s
h
o
r
t
-
t
e
r
m
m
e
m
o
r
y
n
e
t
w
o
r
k
s
(
L
S
T
Ms
)
,
g
e
n
e
r
a
t
i
v
e
ad
v
e
r
s
a
r
i
a
l
n
e
tw
o
r
k
s
(
G
A
Ns
)
,
a
n
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
C
NN
s
)
h
a
v
e
i
m
p
r
o
v
e
d
p
e
r
f
o
r
m
a
n
c
e
[
5
]
.
T
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
f
u
r
t
h
e
r
i
n
c
r
e
a
s
e
d
a
cc
u
r
a
c
y
u
s
i
n
g
p
r
e
t
r
a
i
n
e
d
m
o
d
e
l
s
s
u
c
h
a
s
R
es
N
et5
0
,
M
o
b
i
l
e
N
e
t
,
a
n
d
VG
G
1
9
[
6
]
,
[
7
]
.
F
a
c
e
m
a
s
k
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
1
,
Feb
r
u
ar
y
20
2
6
:
3
9
5
-
403
396
w
i
d
e
l
y
u
s
e
d
d
u
r
i
n
g
C
O
V
I
D
-
1
9
,
o
b
s
c
u
r
e
k
e
y
r
e
g
i
o
n
s
l
i
k
e
t
h
e
m
o
u
t
h
,
m
a
k
i
n
g
e
m
o
t
i
o
n
r
e
co
g
n
i
t
i
o
n
h
a
r
d
e
r
[
8
]
.
R
es
e
a
r
c
h
e
r
s
h
a
v
e
a
p
p
l
i
e
d
at
t
e
n
t
i
o
n
h
e
a
t
m
a
p
s
[
9
]
,
c
r
o
p
p
e
d
e
y
e
r
e
g
i
o
n
s
[
1
0
]
,
o
r
r
e
t
a
i
n
e
d
f
u
ll
im
a
g
e
s
[
1
1
]
,
[
1
2
]
t
o
a
d
d
r
e
s
s
o
c
cl
u
s
i
o
n
.
S
i
n
c
e
n
o
e
m
o
t
i
o
n
-
l
a
b
e
l
e
d
m
as
k
e
d
d
a
t
as
e
ts
e
x
i
s
t
,
s
y
n
t
h
e
ti
c
a
p
p
r
o
a
c
h
e
s
a
r
e
u
s
e
d
.
T
h
r
o
u
g
h
s
e
l
f
-
a
t
t
e
n
t
i
o
n
,
V
i
T
m
o
d
e
ls
c
a
p
t
u
r
e
g
l
o
b
a
l
d
e
p
e
n
d
e
n
c
i
e
s
a
n
d
c
o
n
t
e
x
t
f
r
o
m
i
m
a
g
e
p
a
t
c
h
es
,
w
h
ic
h
a
l
l
o
ws
V
i
T
s
t
o
e
x
t
r
a
c
t
c
o
n
t
e
x
t
u
al
f
e
at
u
r
e
s
f
r
o
m
p
a
r
t
i
a
l
l
y
o
cc
l
u
d
e
d
f
a
c
es
.
T
h
i
s
w
o
r
k
p
r
o
p
o
s
es
a
Vi
T
-
b
a
s
e
d
F
E
R
f
r
a
m
e
w
o
r
k
u
s
i
n
g
a
s
y
n
t
h
e
t
i
c
al
l
y
m
a
s
k
e
d
A
f
f
e
c
t
N
et
d
a
t
a
s
e
t
a
n
d
r
e
s
t
r
u
c
tu
r
e
d
e
m
o
t
i
o
n
c
l
a
s
s
es
.
F
o
ll
o
w
in
g
M
a
g
h
e
r
i
n
i
e
t
a
l
.
[
1
1
]
,
o
u
r
a
p
p
r
o
a
c
h
d
e
m
o
n
s
t
r
a
t
es
i
m
p
r
o
v
e
d
r
e
c
o
g
n
i
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
u
n
d
e
r
o
c
c
l
u
s
i
o
n
.
T
h
e
r
e
m
ain
d
er
o
f
th
is
p
ap
er
is
o
r
g
an
ize
d
as
f
o
llo
ws:
s
ec
tio
n
2
r
ev
iews
r
elate
d
s
tu
d
ies.
Sectio
n
3
d
escr
ib
es th
e
ViT
-
b
ased
FER m
o
d
el.
Sectio
n
4
p
r
esen
ts
r
esu
lts
an
d
d
is
cu
s
s
io
n
.
Sectio
n
5
c
o
n
clu
d
es th
e
p
ap
er
.
2.
RE
L
AT
E
D
WO
RK
T
h
e
v
i
s
i
o
n
t
r
a
n
s
f
o
r
m
e
r
(
Vi
T
)
[
1
3
]
h
a
s
s
h
o
w
n
p
r
o
m
i
s
i
n
g
r
e
s
u
l
ts
i
n
v
a
r
i
o
u
s
c
o
m
p
u
t
e
r
v
is
i
o
n
t
as
k
s
,
i
n
c
l
u
d
i
n
g
F
E
R
[
1
4
]
.
I
t
i
s
b
u
il
t
o
n
t
r
a
n
s
f
o
r
m
e
r
a
r
c
h
i
t
e
ct
u
r
e
w
h
i
c
h
i
s
i
n
i
ti
a
l
l
y
d
es
i
g
n
e
d
f
o
r
N
L
P
.
Vi
T
e
m
p
l
o
y
s
m
u
l
t
i
-
h
e
a
d
s
e
l
f
-
a
t
te
n
t
i
o
n
a
n
d
im
a
g
e
p
a
t
c
h
p
r
o
c
e
s
s
i
n
g
.
H
u
a
n
g
e
t
a
l
.
[
1
5
]
u
t
i
li
z
e
d
V
iT
w
it
h
a
St
a
r
G
A
N
f
r
a
m
e
w
o
r
k
f
o
r
d
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
i
n
F
E
R
.
S
q
u
e
e
z
e
Vi
T
wa
s
p
r
o
p
o
s
e
d
t
o
c
o
m
b
i
n
e
g
l
o
b
a
l
a
n
d
l
o
ca
l
f
e
a
t
u
r
e
s
w
i
t
h
f
ew
e
r
d
i
m
e
n
s
i
o
n
s
[
1
6
]
.
Fa
t
i
m
a
et
a
l
.
[
1
7
]
d
e
m
o
n
s
t
r
a
t
e
d
t
h
e
v
a
l
u
e
o
f
s
e
l
f
-
a
t
t
e
n
ti
o
n
i
n
V
i
T
f
o
r
e
m
o
ti
o
n
r
e
c
o
g
n
i
t
i
o
n
.
Stu
d
ies
h
av
e
also
ad
d
r
ess
ed
FER
u
n
d
er
p
a
r
tial
o
cc
lu
s
io
n
.
T
ec
h
n
iq
u
es
lik
e
Gab
o
r
wav
elet
tex
tu
r
e
an
aly
s
is
,
DNM
F
d
ec
o
m
p
o
s
itio
n
,
an
d
lan
d
m
a
r
k
-
b
ased
s
h
ap
e
an
aly
s
is
h
av
e
b
ee
n
a
p
p
lied
t
o
s
ep
ar
ate
o
cc
lu
d
e
d
ar
ea
s
an
d
ex
tr
ac
t
d
is
cr
im
in
an
t
f
ea
tu
r
es
[
1
8
]
.
Oth
er
s
tu
d
ies
co
n
s
id
er
ed
clo
th
in
g
-
b
ased
o
cc
l
u
s
io
n
,
s
u
ch
as
h
ijab
d
etec
tio
n
u
s
in
g
tr
an
s
f
er
lear
n
i
n
g
[
1
9
]
,
wh
ich
also
d
em
o
n
s
tr
ates
th
e
im
p
ac
t
o
f
p
ar
tial
co
v
e
r
in
g
o
n
r
ec
o
g
n
itio
n
p
er
f
o
r
m
an
ce
.
I
n
ad
d
itio
n
to
o
cc
lu
s
io
n
h
an
d
lin
g
,
o
th
er
wo
r
k
s
f
o
cu
s
ed
o
n
im
p
r
o
v
in
g
d
ata
q
u
ality
an
d
m
o
d
el
r
o
b
u
s
tn
ess
.
Fen
g
an
d
Sh
ao
[
2
0
]
e
n
h
an
ce
d
d
ata
q
u
ality
u
s
in
g
p
r
ep
r
o
ce
s
s
in
g
(
e.
g
.
,
h
is
to
g
r
am
e
q
u
aliza
tio
n
,
af
f
in
e
tr
an
s
f
o
r
m
s
)
an
d
u
s
ed
I
n
ce
p
tio
n
-
v
3
with
tr
a
n
s
f
er
lear
n
in
g
to
ac
h
iev
e
h
i
g
h
ac
c
u
r
ac
y
o
n
C
K+
an
d
J
af
f
e
d
atasets
.
Oth
er
m
eth
o
d
s
ex
p
an
d
ed
class
ic
C
NNs,
s
u
ch
as
L
eNe
t
-
5
[
2
1
]
,
b
y
d
ee
p
e
n
in
g
co
n
v
o
lu
tio
n
an
d
p
o
o
lin
g
lay
er
s
to
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
u
n
d
er
o
cc
lu
s
io
n
.
C
h
en
et
a
l.
[
2
2
]
p
r
o
p
o
s
ed
e
f
f
ic
ien
t
atten
tio
n
-
b
ased
E
R
FNet
en
h
an
ce
m
en
ts
u
s
in
g
g
r
o
u
p
co
n
v
o
lu
tio
n
s
an
d
r
esid
u
al
m
o
d
u
les.
Ma
s
k
-
awa
r
e
FE
R
ap
p
r
o
ac
h
es
h
a
v
e
em
er
g
ed
r
ec
en
tly
.
On
e
s
tu
d
y
u
s
ed
C
NNs
o
n
s
y
n
th
etica
lly
m
a
s
k
ed
Af
f
ec
tNet
d
ata,
m
er
g
in
g
em
o
tio
n
class
es
to
ad
d
r
ess
o
cc
lu
s
io
n
a
n
d
ac
h
iev
ed
9
6
%
tr
ai
n
in
g
ac
c
u
r
ac
y
an
d
7
0
%
v
alid
atio
n
ac
cu
r
ac
y
[
1
1
]
.
AC
NN
[
2
3
]
was
in
tr
o
d
u
ce
d
to
ass
ig
n
ad
ap
tiv
e
weig
h
ts
to
f
ac
ial
r
eg
io
n
o
f
i
n
ter
est
s
(
R
OI
s
)
,
with
v
ar
ian
ts
lik
e
p
AC
NN
an
d
g
AC
NN
in
teg
r
atin
g
lo
ca
l a
n
d
g
lo
b
al
f
ea
tu
r
es.
R
ec
en
t w
o
r
k
c
o
m
b
in
ed
f
ac
e
p
ar
s
in
g
with
a
ViT
-
b
ased
class
if
ier
u
s
in
g
cr
o
s
s
-
atten
tio
n
to
d
if
f
e
r
en
tiate
m
ask
ed
an
d
v
is
ib
le
r
e
g
io
n
s
,
o
u
tp
e
r
f
o
r
m
in
g
o
th
e
r
m
eth
o
d
s
o
n
d
atasets
lik
e
M
-
L
FW
-
FER
an
d
M
-
F
E
R
-
2013
[
2
4
]
.
Ou
r
p
a
p
er
u
tili
ze
s
th
e
ca
p
ab
ilit
ies
o
f
ViT
’
s
s
elf
-
atten
tio
n
to
im
p
r
o
v
e
FER
u
n
d
er
m
ask
o
c
clu
s
io
n
,
u
s
in
g
s
y
n
th
etica
lly
m
ask
ed
Af
f
ec
tNet
d
ata
an
d
c
lass
r
ec
ateg
o
r
izatio
n
f
o
llo
win
g
Ma
g
h
er
in
i
et
a
l.
[
1
1
]
.
3.
ViT
-
B
A
SE
D
F
E
R
F
RA
M
E
WO
RK
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
e
r
all
f
r
am
ewo
r
k
o
f
th
e
p
r
o
p
o
s
ed
ViT
-
b
ased
FER
m
o
d
el
f
o
r
m
a
s
k
ed
f
ac
ial
im
ag
es.
T
h
e
f
r
am
ewo
r
k
i
n
clu
d
es
d
ata
c
o
llectio
n
f
r
o
m
t
h
e
Af
f
ec
tNet
d
ataset,
p
r
ep
r
o
ce
s
s
in
g
u
s
in
g
t
h
e
m
ask
-
th
e
-
f
ac
e
(
MT
F)
to
o
l,
f
i
n
e
-
tu
n
in
g
o
f
th
e
p
r
e
-
tr
ai
n
ed
ViT
m
o
d
el,
an
d
f
in
al
ev
alu
at
io
n
u
s
in
g
s
tan
d
ar
d
m
etr
ics (
p
r
ec
is
io
n
,
r
ec
all,
ac
c
u
r
ac
y
,
an
d
F1
s
co
r
e)
.
Fig
u
r
e
1
.
W
o
r
k
f
lo
w
o
f
t
h
e
p
r
o
p
o
s
ed
FER f
r
am
ewo
r
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
cia
l e
mo
tio
n
r
ec
o
g
n
itio
n
u
n
d
er fa
ce
ma
s
k
o
cc
lu
s
io
n
u
s
in
g
visi
o
n
…
(
A
s
h
r
a
f Yu
n
is
Ma
g
h
a
r
i
)
397
3
.
1
.
Da
t
a
c
o
llect
io
n
Af
f
ec
tNet
co
n
tain
s
o
v
er
o
n
e
m
illi
o
n
f
ac
ial
im
ag
es
co
llected
f
r
o
m
th
e
I
n
ter
n
et
b
y
q
u
er
y
in
g
th
r
ee
s
ea
r
ch
en
g
in
es.
Ab
o
u
t
h
al
f
o
f
th
e
r
etr
iev
e
d
im
ag
es,
a
r
o
u
n
d
4
5
0
th
o
u
s
an
d
s
wer
e
m
a
n
u
ally
an
n
o
tated
f
o
r
elev
en
ca
teg
o
r
ies;
n
eu
tr
al,
h
a
p
p
y
,
s
ad
,
s
u
r
p
r
is
e,
f
ea
r
,
an
g
e
r
,
d
is
g
u
s
t,
co
n
tem
p
t,
n
o
n
e,
u
n
c
er
tain
,
an
d
n
o
n
-
f
ac
e
(
h
e
No
n
e
(
No
n
e
o
f
th
e
ei
g
h
t e
m
o
tio
n
s
)
[
2
5
]
.
T
h
ese
ca
teg
o
r
ie
s
ar
e
s
h
o
wn
in
Fig
u
r
e
2
.
Af
f
ec
tNet
is
wid
ely
u
s
ed
in
f
ac
ial
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
d
u
e
to
its
s
ca
le
an
d
d
iv
er
s
it
y
,
o
f
f
er
in
g
ar
o
u
n
d
4
5
0
K
im
ag
es
[
2
5
]
f
ilt
er
ed
f
r
o
m
1
2
0
GB
o
f
d
ata.
W
ith
f
ac
e
m
ask
s
o
cc
lu
d
in
g
k
e
y
f
ea
tu
r
es,
em
o
tio
n
class
if
icatio
n
b
ec
o
m
es
d
if
f
icu
lt.
T
h
er
ef
o
r
e
,
f
iv
e
class
es
(
An
g
er
-
Dis
g
u
s
t,
Fear
-
Su
r
p
r
is
e,
H
ap
p
in
ess
,
Sad
n
ess
,
an
d
Neu
tr
al)
wer
e
cr
ea
ted
b
y
m
er
g
in
g
s
im
ilar
ex
p
r
ess
io
n
s
,
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
h
is
r
ec
lass
if
icatio
n
im
p
r
o
v
es
r
ec
o
g
n
itio
n
u
n
d
e
r
m
ask
o
cc
lu
s
io
n
.
T
h
e
f
in
al
d
is
tr
ib
u
tio
n
o
f
th
ese
f
iv
e
class
es
is
s
h
o
wn
in
T
ab
le
1
,
an
d
s
tr
atif
ied
s
am
p
lin
g
e
n
s
u
r
ed
b
alan
ce
ac
r
o
s
s
tr
ain
in
g
,
v
a
lid
atio
n
,
an
d
test
s
ets.
I
t
is
w
o
r
th
n
o
tin
g
th
at,
i
n
ad
d
itio
n
to
Af
f
ec
tNet,
o
th
er
d
atasets
lik
e
FER2
0
1
3
,
J
AFF
E
,
an
d
C
K+
ca
n
also
b
e
co
n
s
id
er
ed
f
o
r
em
o
tio
n
r
ec
o
g
n
itio
n
.
T
h
ese
d
atasets
m
ay
b
e
u
s
ed
al
o
n
e
o
r
c
o
m
b
in
e
d
with
Af
f
ec
tNet
to
im
p
r
o
v
e
m
o
d
el
p
e
r
f
o
r
m
an
ce
.
I
n
th
is
s
tu
d
y
,
Af
f
ec
tNet
was
ch
o
s
en
d
u
e
to
its
s
ig
n
if
ican
tly
l
ar
g
er
n
u
m
b
e
r
o
f
f
ac
ial
im
ag
es
,
m
o
r
e
d
iv
er
s
e
an
d
f
in
e
-
g
r
ain
e
d
em
o
tio
n
lab
els
(
s
ev
en
em
o
tio
n
s
:
an
g
er
,
d
is
g
u
s
t,
f
ea
r
,
h
a
p
p
in
ess
,
s
ad
n
ess
,
s
u
r
p
r
is
e,
an
d
n
e
u
tr
al)
,
an
d
r
ea
l
-
wo
r
ld
im
ag
e
co
n
d
iti
o
n
s
.
T
h
ese
ad
v
an
tag
es
m
ak
e
it
m
o
r
e
s
u
itab
le
f
o
r
b
u
ild
in
g
r
o
b
u
s
t
an
d
s
ca
lab
le
FER s
y
s
tem
s
.
Fig
u
r
e
2
.
Sam
p
le
im
a
g
es f
r
o
m
th
e
Af
f
ec
tNet
d
ataset
[
2
5
]
Fig
u
r
e
3
.
R
eo
r
g
an
ized
em
o
tio
n
class
es: An
g
er
+D
is
g
u
s
t a
n
d
Fear
+Su
r
p
r
is
e
T
ab
le
1
.
Sam
p
les p
er
class
af
ter
m
er
g
in
g
Af
f
ec
tNet
ca
teg
o
r
ies
Ex
p
r
e
ssi
o
n
C
a
t
e
g
o
r
y
N
u
mb
e
r
N
e
u
t
r
a
l
8
0
2
7
6
H
a
p
p
y
1
4
6
1
9
8
S
a
d
2
9
4
8
7
F
e
a
r
-
S
u
r
p
r
i
se
2
4
4
7
9
A
n
g
e
r
-
D
i
sg
u
st
3
3
3
9
4
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
1
,
Feb
r
u
ar
y
20
2
6
:
3
9
5
-
403
398
3
.
2
.
Da
t
a
p
re
pro
ce
s
s
ing
T
h
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
f
o
c
u
s
es
o
n
s
im
u
latin
g
r
ea
l
-
wo
r
ld
m
a
s
k
s
ce
n
ar
io
s
b
y
m
ask
in
g
th
e
f
ac
e
im
ag
es
in
th
e
d
ataset.
T
h
is
s
tep
i
s
cr
u
cial
f
o
r
ad
ap
tin
g
th
e
m
o
d
el
to
em
o
tio
n
r
ec
o
g
n
itio
n
u
n
d
er
p
a
r
tial f
ac
ial
o
cc
lu
s
io
n
ca
u
s
ed
b
y
m
ask
u
s
ag
e.
On
e
c
o
m
m
o
n
ly
u
s
ed
to
o
l is th
e
"m
as
k
th
e
f
ac
e"
(
MT
F)
to
o
l
[
2
6
]
,
d
ep
icted
in
Fig
u
r
e
4
.
Fig
u
r
e
4
.
W
o
r
k
f
lo
w
o
f
th
e
“M
ask
th
e
FAC
E
”
(
MT
F)
to
o
l
3
.
3
.
ViT
m
o
del
f
inet
un
ing
I
n
th
is
s
tep
,
a
p
r
e
-
tr
ain
e
d
ViT
m
o
d
el
is
f
in
e
-
tu
n
e
d
u
s
in
g
th
e
au
g
m
en
te
d
Af
f
ec
tNet
d
at
aset
with
s
y
n
th
etic
m
ask
s
.
Fin
e
-
tu
n
in
g
in
v
o
lv
es f
u
r
th
e
r
tr
ain
in
g
t
h
e
m
o
d
el
o
n
th
e
m
ask
ed
im
ag
es
to
r
ef
in
e
its
ca
p
a
b
ilit
y
to
th
e
m
ask
ed
em
o
tio
n
r
ec
o
g
n
itio
n
task
.
3
.
3
.
1
.
Vis
io
n
t
ra
ns
f
o
rm
er
a
r
chit
ec
t
ure
T
h
e
v
is
io
n
tr
an
s
f
o
r
m
e
r
(
ViT
)
is
a
d
ee
p
lear
n
in
g
m
o
d
el
th
at
h
as
s
h
o
wn
s
tate
-
of
-
th
e
-
ar
t
p
er
f
o
r
m
an
ce
in
class
if
icatio
n
ta
s
k
s
[
1
3
]
.
Or
ig
in
ally
d
ev
elo
p
ed
f
o
r
NL
P,
ViT
later
p
r
o
v
ed
ef
f
icien
t
in
v
is
io
n
task
s
[
1
4
]
,
[
2
7
]
.
I
t
co
n
s
is
ts
o
f
a
p
atch
em
b
ed
d
in
g
m
o
d
u
le
th
at
s
p
lits
th
e
im
ag
e
in
to
p
atc
h
es
an
d
f
latte
n
s
th
em
in
to
to
k
e
n
s
,
f
o
llo
wed
b
y
a
tr
a
n
s
f
o
r
m
e
r
e
n
c
o
d
er
co
m
p
o
s
ed
o
f
m
u
lti
-
h
ea
d
s
elf
-
atten
tio
n
an
d
f
ee
d
f
o
r
war
d
lay
er
s
.
E
ac
h
p
atch
is
lin
ea
r
ly
p
r
o
jecte
d
,
an
d
t
h
e
a
tten
tio
n
m
ec
h
an
is
m
c
o
m
p
u
tes
a
weig
h
ted
s
u
m
ac
r
o
s
s
p
atch
es.
On
e
k
ey
ad
v
an
tag
e
o
f
ViT
is
lear
n
in
g
d
ir
ec
tly
f
r
o
m
d
ata
with
o
u
t
m
an
u
al
f
ea
tu
r
e
e
n
g
in
ee
r
in
g
.
I
t
h
as
d
em
o
n
s
tr
ated
s
tr
o
n
g
p
er
f
o
r
m
an
ce
o
n
d
atasets
lik
e
I
m
ag
eNe
t
[
2
8
]
,
m
a
k
in
g
it
a
co
m
p
e
titi
v
e
alter
n
ativ
e
to
tr
ad
itio
n
al
d
ee
p
lear
n
in
g
m
o
d
els.
Fo
r
FER,
th
e
s
elf
-
atten
tio
n
m
ec
h
an
is
m
is
b
en
ef
icial
in
ca
p
tu
r
in
g
f
ea
tu
r
es
f
r
o
m
p
ar
tially
o
cc
lu
d
ed
f
ac
ial
im
ag
es,
s
u
ch
as th
o
s
e
with
f
ac
e
m
ask
s
.
3
.
3
.
2
.
F
ine
-
t
un
ing
s
t
eps
A
p
r
e
-
tr
ai
n
ed
ViT
m
o
d
el
was
f
in
e
-
tu
n
ed
u
s
in
g
th
e
m
ask
ed
Af
f
ec
tNet
d
ataset
to
r
ec
o
g
n
iz
e
em
o
tio
n
s
u
n
d
er
o
cc
lu
s
io
n
.
T
h
e
s
tep
s
in
c
lu
d
e:
−
Mo
d
el
s
elec
tio
n
: A
p
r
e
-
tr
ain
e
d
ViT
m
o
d
el
was c
h
o
s
en
b
ased
o
n
a
r
ch
itectu
r
e
a
n
d
p
r
io
r
p
er
f
o
r
m
an
ce
.
−
I
n
itializatio
n
:
T
h
e
m
o
d
el'
s
p
ar
am
eter
s
,
lear
n
ed
f
r
o
m
lar
g
e
d
atasets
lik
e
I
m
ag
eNe
t,
wer
e
u
s
ed
as
a
s
tar
tin
g
p
o
in
t.
−
Hy
p
er
p
ar
a
m
eter
tu
n
in
g
: L
ea
r
n
in
g
r
ate,
b
atc
h
s
ize,
an
d
r
eg
u
la
r
izatio
n
wer
e
ad
ju
s
ted
e
x
p
er
i
m
en
tally
.
−
F
i
n
e
-
t
u
n
i
n
g
:
T
h
e
m
o
d
e
l
w
a
s
t
r
a
i
n
e
d
o
n
t
h
e
m
a
s
k
e
d
d
a
t
a
s
e
t
u
s
i
n
g
b
a
c
k
p
r
o
p
a
g
a
t
i
o
n
t
o
o
p
t
i
m
i
z
e
c
l
a
s
s
i
f
i
ca
t
i
o
n
a
c
c
u
r
a
c
y
.
T
h
i
s
p
r
o
c
e
s
s
h
e
l
p
s
t
h
e
V
i
T
m
o
d
e
l
l
e
a
r
n
t
h
e
l
i
n
k
b
e
t
w
e
e
n
m
a
s
k
e
d
f
a
c
i
a
l
f
e
a
t
u
r
e
s
a
n
d
t
h
e
r
e
c
a
t
e
g
o
r
i
z
e
d
e
m
o
t
i
o
n
s
.
An
ex
am
p
le
o
f
f
ea
t
u
r
e
e
x
tr
ac
tio
n
b
y
th
e
f
in
e
-
tu
n
e
d
ViT
m
o
d
el
is
s
h
o
wn
in
Fig
u
r
e
5
.
T
h
is
atten
tio
n
m
ap
was
ex
tr
ac
ted
f
r
o
m
th
e
f
i
n
al
s
elf
-
atten
tio
n
lay
er
o
f
th
e
ViT
m
o
d
el
u
s
in
g
v
is
u
aliza
tio
n
to
o
ls
p
r
o
v
id
ed
b
y
th
e
h
u
g
g
in
g
f
ac
e
tr
a
n
s
f
o
r
m
er
s
lib
r
ar
y
,
an
d
it
h
ig
h
lig
h
ts
th
e
f
ac
ial
r
eg
io
n
s
(
p
r
im
ar
ily
th
e
e
y
e
ar
ea
)
t
h
at
m
o
s
t
in
f
lu
en
ce
th
e
m
o
d
el’
s
d
ec
is
io
n
-
m
ak
in
g
p
r
o
ce
s
s
.
3
.
4
.
E
v
a
lua
t
i
o
n
T
h
e
ev
alu
atio
n
p
r
o
ce
s
s
ty
p
ically
co
m
p
r
is
es th
e
f
o
llo
win
g
k
e
y
co
m
p
o
n
en
t:
−
T
est d
ataset: I
n
clu
d
es d
iv
er
s
e
f
ac
ial
ex
p
r
ess
io
n
s
with
m
ask
s
,
s
im
u
latin
g
r
ea
l
-
wo
r
ld
s
ce
n
ar
i
o
s
.
−
Pre
d
ictio
n
s
: T
h
e
m
o
d
el
o
u
tp
u
ts
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
s
o
v
e
r
em
o
tio
n
class
es.
−
Me
tr
ics:
Pre
cisi
o
n
,
R
ec
all,
Ac
cu
r
ac
y
,
a
n
d
F1
Sco
r
e
.
−
Mo
d
el
co
m
p
ar
is
o
n
:
T
h
e
ViT
-
b
ased
FER
m
o
d
el
is
co
m
p
ar
e
d
with
b
aselin
e
an
d
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
to
v
alid
ate
im
p
r
o
v
em
en
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
cia
l e
mo
tio
n
r
ec
o
g
n
itio
n
u
n
d
er fa
ce
ma
s
k
o
cc
lu
s
io
n
u
s
in
g
visi
o
n
…
(
A
s
h
r
a
f Yu
n
is
Ma
g
h
a
r
i
)
399
Usi
n
g
th
ese
ev
alu
atio
n
m
etr
ics
ca
n
q
u
an
titativ
ely
m
ea
s
u
r
e
th
e
ac
cu
r
ac
y
an
d
ef
f
ec
tiv
e
n
ess
o
f
th
e
f
in
e
-
tu
n
ed
ViT
m
o
d
el
in
r
ec
o
g
n
izin
g
em
o
tio
n
s
u
n
d
e
r
f
ac
e
m
ask
.
Fig
u
r
e
5
.
Atten
tio
n
m
ap
f
r
o
m
th
e
f
in
al
lay
er
o
f
th
e
ViT
m
o
d
el,
h
ig
h
lig
h
tin
g
f
ac
ial
r
e
g
io
n
s
(
m
ain
ly
th
e
e
y
es)
th
at
g
u
id
ed
th
e
p
r
e
d
ictio
n
.
Vis
u
alize
d
u
s
in
g
Hu
g
g
in
g
Face
to
o
ls
3
.
5
.
E
x
perim
ent
a
l
env
ir
o
nm
ent
s
et
up
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
was
im
p
lem
en
ted
in
th
e
Go
o
g
le
C
o
lab
Pro
en
v
ir
o
n
m
en
t
wit
h
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
ac
ce
l
er
atio
n
(
T
esla
T
4
)
.
A
n
o
v
er
v
iew
o
f
th
e
co
m
p
u
tatio
n
al
s
et
u
p
is
illu
s
tr
ated
in
Fig
u
r
e
6
,
wh
ile
th
e
d
etailed
d
e
s
cr
ip
tio
n
o
f
th
e
ex
p
e
r
im
en
tal
e
n
v
ir
o
n
m
en
t is p
r
o
v
id
ed
i
n
s
ec
tio
n
4
.
1
.
Fig
u
r
e
6
.
E
x
p
er
im
e
n
tal
en
v
ir
o
n
m
en
t setu
p
:
th
e
in
te
g
r
atio
n
o
f
Go
o
g
le
C
o
lab
Pro
,
GPU
r
eso
u
r
ce
s
,
an
d
t
h
e
k
ey
lib
r
ar
ies u
s
ed
to
tr
ain
a
n
d
ev
al
u
ate
th
e
ViT
-
b
ased
FER m
o
d
e
l
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
E
x
perim
ent
s
env
iro
nm
e
nt
T
h
e
ex
p
er
im
en
ts
wer
e
im
p
le
m
en
ted
in
th
e
Go
o
g
le
C
o
lab
Pro
en
v
ir
o
n
m
e
n
t
with
GPU
a
cc
eler
atio
n
(
T
esla
T
4
)
.
C
o
lab
Pro
p
r
o
v
id
es
h
ig
h
-
p
er
f
o
r
m
an
ce
r
eso
u
r
ce
s
with
p
r
e
-
in
s
talled
f
r
am
ewo
r
k
s
s
u
ch
as
Py
T
o
r
ch
an
d
T
en
s
o
r
Flo
w,
alo
n
g
with
s
u
p
p
o
r
tin
g
lib
r
ar
ies
f
o
r
p
r
ep
r
o
ce
s
s
in
g
an
d
v
is
u
aliza
tio
n
.
T
h
is
s
etu
p
en
s
u
r
ed
ef
f
icien
t tr
ain
in
g
an
d
r
ep
r
o
d
u
c
ib
ilit
y
o
f
th
e
ex
p
er
im
en
tal
r
esu
lts
.
4
.
2
.
E
x
perim
ent
d
a
t
a
s
et
T
o
ad
d
r
ess
th
e
f
ea
tu
r
e
lo
s
s
f
r
o
m
f
ac
e
m
ask
s
,
th
e
o
r
ig
in
al
eig
h
t
em
o
tio
n
class
es
in
Af
f
ec
tNet
wer
e
r
estru
ctu
r
ed
in
to
f
iv
e
(
an
g
er
-
d
is
g
u
s
t,
f
ea
r
-
s
u
r
p
r
is
e,
h
ap
p
in
ess
,
s
ad
n
ess
,
an
d
n
eu
tr
al
)
,
as
s
h
o
wn
in
Fig
u
r
e
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
1
,
Feb
r
u
ar
y
20
2
6
:
3
9
5
-
403
400
T
h
is
r
ec
ateg
o
r
izatio
n
e
n
s
u
r
ed
r
ec
o
g
n
itio
n
o
f
k
ey
em
o
tio
n
s
d
esp
ite
o
cc
lu
s
io
n
.
T
h
e
d
ataset
(
≈
3
2
0
K
im
a
g
es)
was sp
lit to
8
5
% f
o
r
tr
ain
in
g
,
an
d
1
5
% f
o
r
v
alid
atio
n
.
4
.
3
.
I
m
ple
m
ent
a
t
io
n
d
et
a
ils
I
n
th
is
s
ec
tio
n
,
th
e
im
p
lem
en
tatio
n
s
ettin
g
s
u
s
ed
d
u
r
in
g
th
e
f
in
e
-
tu
n
in
g
p
r
o
ce
s
s
o
f
th
e
ViT
m
o
d
el
ar
e
d
etailed
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
was
im
p
lem
en
ted
u
s
in
g
Py
T
o
r
ch
L
ig
h
tn
in
g
an
d
f
i
n
e
-
tu
n
e
d
i
n
th
e
G
o
o
g
le
C
o
lab
Pro
en
v
ir
o
n
m
en
t w
ith
ac
ce
s
s
to
a
T
esla T
4
GPU.
W
e
u
s
ed
th
e
p
r
etr
ain
ed
ViT
-
lar
g
e
-
p
atch
1
6
-
2
2
4
-
in
2
1
k
m
o
d
el
f
r
o
m
Hu
g
g
in
g
Face
an
d
tr
ain
ed
it
o
n
th
e
m
ask
ed
Af
f
ec
tNet
d
ataset
ca
teg
o
r
ized
in
to
f
i
v
e
em
o
tio
n
class
es.
T
r
ain
in
g
was
co
n
d
u
cted
f
o
r
3
ep
o
ch
s
with
a
lear
n
i
n
g
r
at
e
o
f
2
e
-
5
,
u
s
in
g
th
e
Ad
am
o
p
tim
izer
an
d
m
ix
e
d
p
r
ec
is
io
n
(
1
6
-
b
it).
T
h
e
d
atas
et
was
s
p
lit
in
to
8
5
%
f
o
r
tr
ain
in
g
a
n
d
1
5
%
f
o
r
v
alid
ati
o
n
,
u
s
in
g
s
tr
atif
ied
s
am
p
lin
g
.
E
v
alu
atio
n
m
etr
ics
in
clu
d
ed
ac
c
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
s
co
r
e
,
co
m
p
u
ted
u
s
in
g
m
ac
r
o
-
av
er
ag
in
g
to
en
s
u
r
e
f
air
co
m
p
ar
is
o
n
ac
r
o
s
s
all
class
es.
Fig
u
r
e
7
.
T
h
e
o
r
i
g
in
al
Af
f
ec
t
Net
class
e
s
wer
e
m
er
g
ed
,
b
ased
o
n
v
is
u
al
s
im
ilar
ity
,
in
to
f
iv
e
ca
teg
o
r
ies:
an
g
er
-
d
is
g
u
s
t,
f
ea
r
-
s
u
r
p
r
is
e,
h
ap
p
i
n
e
s
s
,
s
ad
n
ess
,
an
d
n
eu
tr
al
4
.
4
.
Cho
o
s
ing
t
he
bes
t
ViT
pret
ra
ined m
o
del
T
h
e
f
ir
s
t
ex
p
er
im
en
t
ev
alu
ated
s
ev
er
al
p
r
e
-
tr
ain
ed
Vi
T
ar
ch
itectu
r
es
f
o
r
m
ask
ed
em
o
tio
n
r
ec
o
g
n
itio
n
.
E
ac
h
m
o
d
el
was
f
in
e
-
tu
n
e
d
o
n
t
h
e
d
ataset
to
a
d
ap
t
to
o
cc
l
u
s
io
n
ef
f
ec
ts
.
T
h
e
g
o
o
g
le/ViT
-
lar
g
e
-
p
atch
1
6
-
224
-
in
2
1
k
m
o
d
el
ac
h
i
ev
ed
th
e
h
ig
h
est ac
cu
r
ac
y
o
f
8
0
.
8
%.
4
.
5
.
Co
m
pa
riso
n wit
h a
s
t
a
t
e
-
of
-
t
he
-
a
rt
CNN
mo
del
T
h
is
ex
p
er
im
en
t
co
m
p
a
r
ed
th
e
b
est
ViT
m
o
d
el
id
en
tifie
d
i
n
th
e
p
r
ev
io
u
s
ex
p
e
r
im
en
t
with
R
esNet
-
5
0
.
B
o
th
m
o
d
els
wer
e
f
in
e
-
tu
n
ed
o
n
th
e
s
am
e
m
ask
ed
f
ac
i
al
d
ataset
an
d
ev
al
u
ated
u
s
in
g
p
r
ec
is
io
n
,
r
e
ca
ll,
ac
cu
r
ac
y
,
a
n
d
F1
s
co
r
e.
As
s
h
o
wn
in
T
a
b
le
2
,
th
e
ViT
m
o
d
el
o
u
tp
e
r
f
o
r
m
s
R
esNet
-
5
0
in
all
m
etr
ics.
T
h
is
r
esu
lt c
o
n
f
ir
m
s
th
e
ViT
s
u
p
e
r
i
o
r
ity
f
o
r
ac
cu
r
ately
class
if
y
in
g
em
o
tio
n
s
in
th
e
p
r
esen
ce
o
f
f
a
ce
m
ask
s
.
T
ab
le
2
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
ViT
an
d
R
esNet
-
50
M
o
d
e
l
/
M
e
tr
i
c
A
c
c
u
ra
c
y
P
r
e
c
i
si
o
n
R
e
c
a
l
l
F1
s
co
re
V
i
T
-
la
r
g
e
-
p
at
c
h
16
-
22
4
-
i
n
2
1
k
0
.
81
0
.
77
0
.
77
0
.
75
Res
N
et
-
50
0
.
6
1
0
.
4
9
0
.
5
1
0
.
4
7
4
.
6
.
Co
m
pa
riso
n wit
h
o
t
her
s
t
a
t
e
o
f
t
he
a
rt
wo
r
k
s
T
o
f
u
r
th
e
r
v
alid
ate
th
e
e
f
f
ec
t
iv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
ViT
m
o
d
el
f
o
r
FER
in
th
e
p
r
esen
ce
o
f
f
ac
e
m
ask
s
,
we
co
n
d
u
cted
a
c
o
m
p
a
r
ativ
e
an
aly
s
is
with
o
th
er
s
tate
-
of
-
th
e
-
ar
t
ap
p
r
o
ac
h
es.
T
h
e
co
m
p
ar
is
o
n
i
n
clu
d
es
C
NNs
an
d
ViT
-
b
ased
ar
ch
it
ec
tu
r
es.
As
s
h
o
wn
in
T
a
b
le
3
,
o
u
r
ViT
-
b
ased
m
o
d
el
ac
h
iev
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
(
8
1
%)
am
o
n
g
o
th
e
r
co
m
p
ar
ed
m
eth
o
d
s
.
T
h
is
u
n
d
er
s
co
r
es
th
e
ef
f
ec
tiv
en
ess
o
f
o
u
r
ap
p
r
o
ac
h
in
ad
d
r
ess
in
g
th
e
ch
allen
g
es o
f
f
ac
ial
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
ta
s
k
s
u
n
d
er
co
n
d
itio
n
o
f
f
ac
ial
o
cc
lu
s
io
n
.
T
ab
le
3
.
C
o
m
p
a
r
is
o
n
to
o
th
er
s
tate
-
of
-
th
e
-
ar
t w
o
r
k
s
W
o
r
k
M
o
d
e
l
D
a
t
a
s
e
t
Y
e
a
r
A
c
c
u
r
a
c
y
O
c
c
l
u
si
o
n
a
w
a
r
e
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
C
N
N
w
i
t
h
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm
[
2
3
]
C
N
N
F
ED
-
RO
2
0
1
8
6
6
.
5
0
%
F
a
c
e
-
ma
sk
-
a
w
a
r
e
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
se
d
o
n
f
a
c
e
p
a
r
si
n
g
a
n
d
v
i
s
i
o
n
t
r
a
n
sf
o
r
m
e
r
[
2
9
]
V
i
T
M
-
F
ER
2
0
1
3
a
n
d
M
C
K
+
2
0
2
2
6
6
.
5
3
%
M
a
s
k
e
d
f
a
c
e
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
f
a
c
i
a
l
l
a
n
d
mar
k
s a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s f
o
r
v
i
s
u
a
l
l
y
i
m
p
a
i
r
e
d
p
e
o
p
l
e
[
3
0
]
C
N
N
A
f
f
e
c
t
N
e
t
2
0
2
3
6
9
.
3
%
Emo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
i
n
t
h
e
t
i
mes
o
f
C
O
V
I
D
1
9
:
C
o
p
i
n
g
w
i
t
h
f
a
c
e
mas
k
s
[
1
1
]
R
e
sN
e
t
A
f
f
e
c
t
N
e
t
2
0
2
2
7
0
%
Th
e
p
r
o
p
o
s
e
d
V
i
T
-
b
a
s
e
d
mo
d
e
l
V
i
T
A
f
f
e
c
t
N
e
t
2
0
2
4
8
1
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
cia
l e
mo
tio
n
r
ec
o
g
n
itio
n
u
n
d
er fa
ce
ma
s
k
o
cc
lu
s
io
n
u
s
in
g
visi
o
n
…
(
A
s
h
r
a
f Yu
n
is
Ma
g
h
a
r
i
)
401
4
.
7
.
Dis
cus
s
io
n
T
h
e
ex
p
er
im
e
n
tal
r
esu
lts
r
ev
ea
led
th
at
th
e
v
is
io
n
tr
an
s
f
o
r
m
er
(
ViT
)
m
o
d
el
o
u
t
p
er
f
o
r
m
ed
s
tate
-
of
-
th
e
-
ar
t
im
ag
e
class
if
icatio
n
m
eth
o
d
s
,
v
alid
atin
g
th
e
clai
m
s
m
ad
e
b
y
Do
s
o
v
its
k
iy
et
a
l.
[
1
3
]
.
T
h
eir
g
r
o
u
n
d
b
r
ea
k
in
g
wo
r
k
d
e
m
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
Vi
T
m
o
d
els
i
n
ca
p
t
u
r
in
g
s
p
atial
r
elatio
n
s
h
ip
s
a
n
d
g
lo
b
al
co
n
te
x
t,
lead
in
g
to
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
in
task
s
s
u
ch
as
em
o
tio
n
r
ec
o
g
n
itio
n
.
M
o
r
eo
v
er
,
o
u
r
f
in
d
in
g
s
s
h
o
wed
th
at
th
e
ViT
m
o
d
el
a
ch
iev
ed
co
m
p
ar
ab
le
r
esu
lts
to
th
e
m
eth
o
d
p
r
o
p
o
s
ed
b
y
Ma
g
h
er
in
i
et
a
l.
[
1
1
]
,
d
esp
ite
o
f
u
s
in
g
o
n
ly
a
s
m
alle
r
s
u
b
s
et
o
f
Af
f
ec
tNet
d
ataset.
T
h
is
f
in
d
in
g
h
ig
h
lig
h
ts
th
e
Vi
T
m
o
d
el
ef
f
icie
n
cy
in
u
tili
zin
g
d
ata
r
eso
u
r
ce
s
to
p
r
o
d
u
ce
co
m
p
etitiv
e
p
er
f
o
r
m
a
n
ce
co
m
p
a
r
ed
to
R
esNet
-
5
0
wh
ich
r
eq
u
ir
es
v
ast
am
o
u
n
t
o
f
d
ata
a
n
d
ex
te
n
s
iv
e
co
m
p
u
tin
g
tim
e.
F
u
r
t
h
e
r
m
o
r
e
,
b
a
s
e
d
o
n
t
h
e
t
r
a
i
n
i
n
g
e
n
v
i
r
o
n
m
e
n
t
u
s
e
d
i
n
t
h
i
s
s
t
u
d
y
(
G
o
o
g
l
e
C
o
l
a
b
P
r
o
w
i
th
T
e
s
l
a
T
4
G
P
U
)
,
t
h
e
e
s
t
i
m
a
te
d
i
n
f
e
r
e
n
c
e
t
i
m
e
p
e
r
i
m
a
g
e
f
o
r
t
h
e
f
i
n
e
-
t
u
n
e
d
V
i
T
-
l
a
r
g
e
-
p
a
t
c
h
1
6
-
224
-
i
n
2
1
k
m
o
d
e
l
i
s
a
p
p
r
o
x
i
m
a
t
e
l
y
1
8
–
2
2
m
i
l
l
is
e
co
n
d
s
.
T
h
i
s
p
e
r
f
o
r
m
a
n
c
e
i
n
d
i
ca
t
e
s
t
h
a
t
t
h
e
m
o
d
el
is
c
a
p
a
b
le
o
f
n
e
a
r
r
e
a
l
-
t
i
m
e
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
,
m
a
k
i
n
g
it
s
u
it
a
b
l
e
f
o
r
d
e
p
l
o
y
m
e
n
t
i
n
p
r
a
c
t
i
c
al
a
p
p
l
i
ca
t
i
o
n
s
.
T
h
e
p
a
tch
-
b
a
s
e
d
p
r
o
c
e
s
s
i
n
g
a
n
d
s
e
l
f
-
a
tt
e
n
t
i
o
n
m
e
c
h
a
n
is
m
em
p
l
o
y
e
d
b
y
t
h
e
v
i
s
i
o
n
t
r
a
n
s
f
o
r
m
e
r
c
o
n
t
r
i
b
u
t
e
n
o
t
o
n
l
y
t
o
it
s
ac
c
u
r
a
c
y
,
b
u
t
a
l
s
o
t
o
i
t
s
c
o
m
p
u
t
a
ti
o
n
a
l
e
f
f
i
c
i
e
n
c
y
.
T
h
e
s
e
a
d
v
a
n
t
a
g
e
s
m
a
k
e
t
h
e
m
o
d
e
l
w
e
l
l
-
s
u
i
t
e
d
f
o
r
e
n
v
i
r
o
n
m
e
n
t
s
w
i
t
h
l
i
m
i
t
e
d
r
e
s
o
u
r
c
e
s
,
s
u
c
h
a
s
m
o
b
il
e
d
e
v
ic
e
s
o
r
e
m
b
e
d
d
e
d
s
y
s
t
e
m
s
,
w
h
er
e
b
o
t
h
s
p
e
e
d
a
n
d
p
e
r
f
o
r
m
a
n
c
e
a
r
e
c
r
i
t
i
ca
l
.
On
e
o
f
th
e
ad
v
an
tag
es
o
f
th
e
ViT
m
o
d
el
th
at
b
ec
am
e
ev
id
en
t
d
u
r
i
n
g
o
u
r
ex
p
er
im
e
n
ts
was
it
s
r
elativ
ely
f
aster
ex
ec
u
tio
n
ti
m
e
co
m
p
ar
e
d
to
tr
ad
itio
n
al
C
NN
ar
ch
itectu
r
es.
Mo
r
eo
v
e
r
,
th
e
s
elf
-
atten
tio
n
m
ec
h
an
is
m
allo
ws
it
t
o
ca
p
tu
r
e
l
o
n
g
-
r
an
g
e
d
e
p
en
d
e
n
c
ies
in
th
e
im
ag
e,
elim
in
ati
n
g
th
e
n
ee
d
f
o
r
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
c
o
n
v
o
lu
ti
o
n
al
o
p
er
atio
n
s
.
T
h
i
s
ad
v
an
tag
e
n
o
t
o
n
l
y
ac
ce
le
r
ates
tr
ain
in
g
a
n
d
in
f
er
en
ce
b
u
t
also
m
ak
es
Vi
T
m
o
d
els
m
o
r
e
s
ca
lab
le
t
o
l
ar
g
er
d
atasets
an
d
co
m
p
u
tati
o
n
ally
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
.
Ad
d
itio
n
ally
,
th
e
ViT
m
o
d
el'
s
ab
ilit
y
to
o
u
tp
er
f
o
r
m
ex
is
tin
g
s
tate
-
of
-
t
h
e
-
ar
t
m
eth
o
d
s
an
d
ac
h
iev
e
b
etter
p
er
f
o
r
m
a
n
ce
with
r
ed
u
ce
d
d
atab
ase
s
ize
d
em
o
n
s
tr
ates
its
p
o
ten
tial
as
a
p
o
wer
f
u
l
im
ag
e
class
if
icatio
n
to
o
l.
As
we
co
n
tin
u
e
to
e
x
p
lo
r
e
an
d
r
ef
in
e
th
e
ViT
ar
ch
itectu
r
e,
we
ca
n
an
ticip
ate
f
u
r
th
e
r
im
p
r
o
v
em
e
n
ts
in
ac
cu
r
ac
y
,
g
e
n
er
aliza
tio
n
,
an
d
ef
f
icien
c
y
,
o
p
en
in
g
u
p
n
ew
p
o
s
s
ib
ilit
ies
in
v
ar
io
u
s
co
m
p
u
te
r
v
is
io
n
task
s
.
5.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
e
r
,
we
em
p
lo
y
ed
ViT
f
o
r
f
ac
ial
e
m
o
tio
n
r
ec
o
g
n
itio
n
u
n
d
er
m
ask
o
cc
lu
s
io
n
.
T
r
ad
itio
n
al
f
ac
ial
em
o
tio
n
r
ec
o
g
n
itio
n
h
as
b
ee
n
p
r
im
ar
ily
b
ased
o
n
v
is
ib
ilit
y
o
f
th
e
f
ac
e
.
T
o
co
n
d
u
ct
o
u
r
ex
p
er
im
en
ts
,
th
e
Af
f
ec
tNet
d
ataset,
wh
ich
c
o
n
tain
s
a
lar
g
e
co
llectio
n
o
f
em
o
tio
n
al
f
ac
ial
im
ag
es,
h
as
b
ee
n
u
s
ed
.
A
n
ew
ap
p
r
o
ac
h
is
u
s
ed
to
s
im
u
late
r
ea
l
-
wo
r
ld
co
n
d
itio
n
s
o
f
wea
r
i
n
g
f
ac
e
m
ask
s
.
W
e
au
g
m
en
te
d
th
e
im
ag
es
in
th
e
Af
f
ec
tNet
d
ataset
b
y
ad
d
in
g
f
ac
e
m
ask
s
u
s
in
g
a
cu
s
to
m
s
cr
i
p
t.
T
h
is
au
g
m
e
n
tatio
n
was
ess
en
tial
to
en
s
u
r
e
th
at
o
u
r
ViT
-
b
ased
FER
m
o
d
el
w
o
u
ld
b
e
ex
p
o
s
ed
to
th
e
ch
allen
g
es
p
o
s
ed
b
y
p
a
r
tially
o
cc
lu
d
e
d
f
ac
es,
r
e
p
licatin
g
th
e
co
n
d
itio
n
s
we
en
co
u
n
ter
in
o
u
r
d
aily
liv
es.
Su
b
s
eq
u
en
tly
,
we
f
in
etu
n
e
d
an
d
ev
alu
ated
o
u
r
ViT
-
b
ased
m
o
d
el
o
n
th
is
au
g
m
en
ted
d
ataset.
T
h
e
r
esu
lts
o
f
o
u
r
ex
p
er
im
e
n
ts
wer
e
q
u
ite
p
r
o
m
is
in
g
,
as
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
ac
h
iev
e
d
an
ac
cu
r
ac
y
o
f
8
1
%.
T
h
is
f
in
d
in
g
d
em
o
n
s
tr
ates
th
e
r
e
m
ar
k
ab
le
ca
p
ab
ilit
y
o
f
ViT
m
o
d
els
to
ac
cu
r
ately
r
ec
o
g
n
ize
e
m
o
tio
n
s
ev
e
n
wh
e
n
th
e
f
ac
e
is
p
ar
tially
o
cc
lu
d
ed
b
y
a
m
ask
.
T
h
is
is
p
ar
ticu
lar
l
y
s
ig
n
if
ican
t
in
th
e
co
n
tex
t
o
f
o
u
r
c
u
r
r
en
t
tim
es,
wh
er
e
m
ask
-
wea
r
in
g
is
p
r
ev
al
en
t a
n
d
ess
en
tial f
o
r
p
u
b
lic
h
e
alth
.
I
n
o
r
d
er
to
ev
alu
ate
o
u
r
p
r
o
p
o
s
ed
m
o
d
el,
we
e
m
p
lo
y
e
d
v
ar
io
u
s
ev
alu
atio
n
m
etr
ic
s
,
s
u
ch
as
ac
cu
r
ac
y
,
F1
-
s
co
r
e,
an
d
r
ec
all
.
T
h
ese
m
etr
ics
g
av
e
u
s
m
o
r
e
q
u
alitativ
e
in
f
o
r
m
atio
n
o
n
th
e
m
o
d
el’
s
a
b
ilit
y
to
p
r
ed
ict
p
o
s
itiv
e
em
o
tio
n
s
,
as
well
a
s
o
n
th
e
d
is
tr
ib
u
tio
n
o
f
th
e
em
o
tio
n
s
in
th
e
d
ataset.
Mo
r
eo
v
er
,
we
co
m
p
ar
e
t
h
e
ef
f
icien
cy
o
f
o
u
r
p
r
o
p
o
s
ed
ViT
-
b
ased
m
o
d
el
with
o
th
er
s
tate
-
of
-
th
e
-
a
r
t
m
eth
o
d
s
f
o
r
m
ask
ed
f
ac
ial
em
o
tio
n
r
ec
o
g
n
itio
n
.
T
h
e
r
esu
lts
s
h
o
wed
th
at
th
e
ViT
-
b
ased
m
o
d
el
o
u
tp
e
r
f
o
r
m
ed
o
th
er
tec
h
n
iq
u
es
in
th
e
f
ield
o
f
FER
ap
p
licatio
n
.
Fo
r
f
u
tu
r
e
wo
r
k
,
th
e
FER
s
y
s
t
em
ca
n
b
e
im
p
r
o
v
ed
b
y
o
p
tim
izin
g
th
e
ViT
m
o
d
el
f
o
r
m
ask
ed
f
ac
es,
u
s
in
g
lar
g
e
r
an
d
m
o
r
e
d
iv
er
s
e
d
atasets
to
im
p
r
o
v
e
g
en
e
r
aliza
tio
n
,
a
n
d
ex
p
lo
r
in
g
h
o
w
th
e
tr
ain
ed
ViT
m
o
d
el
ca
n
b
e
ad
a
p
ted
to
o
t
h
er
task
s
lik
e
f
ac
ial
e
x
p
r
ess
io
n
an
aly
s
is
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
p
ap
er
is
p
a
r
tially
s
u
p
p
o
r
ted
b
y
th
e
d
ea
n
o
f
h
ig
h
er
s
t
u
d
ies
an
d
s
cien
tific
r
esear
ch
at
I
s
lam
ic
Un
iv
er
s
ity
o
f
Gaz
a.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
1
,
Feb
r
u
ar
y
20
2
6
:
3
9
5
-
403
402
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ash
r
af
Yu
n
is
Ma
g
h
ar
i
✓
✓
✓
✓
✓
✓
✓
Am
ee
r
M.
T
elb
an
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
co
n
f
lict o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
D
a
t
a
a
v
a
il
a
b
i
li
t
y
is
n
o
t
a
p
p
l
i
ca
b
l
e
t
o
t
h
is
p
a
p
e
r
a
s
n
o
n
e
w
d
at
a
w
e
r
e
c
r
e
a
t
e
d
o
r
a
n
al
y
z
e
d
i
n
t
h
is
s
t
u
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
Y
.
L.
Ti
a
n
,
T.
K
a
n
a
d
e
,
a
n
d
J
.
F
.
C
o
n
n
,
“
R
e
c
o
g
n
i
z
i
n
g
a
c
t
i
o
n
u
n
i
t
s
f
o
r
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
a
n
a
l
y
si
s
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
Pa
t
t
e
rn
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
2
3
,
n
o
.
2
,
p
p
.
9
7
–
1
1
5
,
2
0
0
1
,
d
o
i
:
1
0
.
1
1
0
9
/
3
4
.
9
0
8
9
6
2
.
[
2
]
Y
.
H
u
a
n
g
,
F
.
C
h
e
n
,
S
.
Lv
,
a
n
d
X
.
W
a
n
g
,
“
F
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
:
a
su
r
v
e
y
,
”
S
y
m
m
e
t
r
y
,
v
o
l
.
1
1
,
n
o
.
1
0
,
p
.
1
1
8
9
,
S
e
p
.
2
0
1
9
,
d
o
i
:
1
0
.
3
3
9
0
/
s
y
m
1
1
1
0
1
1
8
9
.
[
3
]
S
.
D
u
,
Y
.
Ta
o
,
a
n
d
A
.
M
.
M
a
r
t
i
n
e
z
,
“
C
o
m
p
o
u
n
d
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
s o
f
e
mo
t
i
o
n
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
N
a
t
i
o
n
a
l
A
c
a
d
e
m
y
o
f
S
c
i
e
n
c
e
s
o
f
t
h
e
U
n
i
t
e
d
S
t
a
t
e
s
o
f
Am
e
r
i
c
a
,
v
o
l
.
1
1
1
,
n
o
.
1
5
,
p
p
.
E
1
4
5
4
–
E
1
4
6
2
,
2
0
1
4
,
d
o
i
:
1
0
.
1
0
7
3
/
p
n
a
s.
1
3
2
2
3
5
5
1
1
1
.
[
4
]
S
.
Li
a
n
d
W
.
D
e
n
g
,
“
D
e
e
p
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
:
A
su
r
v
e
y
,
”
I
EE
E
T
ra
n
sa
c
t
i
o
n
s
o
n
A
f
f
e
c
t
i
v
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
3
,
n
o
.
3
,
p
p
.
1
1
9
5
–
1
2
1
5
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
A
F
F
C
.
2
0
2
0
.
2
9
8
1
4
4
6
.
[
5
]
W
.
M
e
l
l
o
u
k
a
n
d
W
.
H
a
n
d
o
u
z
i
,
“
F
a
c
i
a
l
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
:
r
e
v
i
e
w
a
n
d
i
n
s
i
g
h
t
s
,
”
Pro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
7
5
,
p
p
.
6
8
9
–
6
9
4
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
0
.
0
7
.
1
0
1
.
[
6
]
H
.
P
.
W
e
i
,
Y
.
Y
.
D
e
n
g
,
F
.
T
a
n
g
,
X
.
J.
P
a
n
,
a
n
d
W
.
M
.
D
o
n
g
,
“
A
c
o
m
p
a
r
a
t
i
v
e
s
t
u
d
y
o
f
C
N
N
-
a
n
d
t
r
a
n
sf
o
r
mer
-
b
a
s
e
d
v
i
s
u
a
l
st
y
l
e
t
r
a
n
sf
e
r
,
”
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
3
7
,
n
o
.
3
,
p
p
.
6
0
1
–
6
1
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
3
9
0
-
0
2
2
-
2
1
4
0
-
7.
[
7
]
K
.
F
e
n
g
a
n
d
T.
C
h
a
s
p
a
r
i
,
“
A
r
e
v
i
e
w
o
f
g
e
n
e
r
a
l
i
z
a
b
l
e
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
i
n
a
u
t
o
ma
t
i
c
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
,
”
Fro
n
t
i
e
rs
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
,
n
o
.
9
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
3
3
8
9
/
f
c
o
mp
.
2
0
2
0
.
0
0
0
0
9
.
[
8
]
L.
Zh
a
n
g
,
B
.
V
e
r
ma
,
D
.
Tj
o
n
d
r
o
n
e
g
o
r
o
,
a
n
d
V
.
C
h
a
n
d
r
a
n
,
“
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
a
n
a
l
y
s
i
s
u
n
d
e
r
p
a
r
t
i
a
l
o
c
c
l
u
si
o
n
:
A
su
r
v
e
y
,
”
AC
M
C
o
m
p
u
t
i
n
g
S
u
r
v
e
y
s
,
v
o
l
.
5
1
,
n
o
.
2
,
p
p
.
1
–
4
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
4
5
/
3
1
5
8
3
6
9
.
[
9
]
B
.
Y
a
n
g
,
J.
W
u
,
a
n
d
G
.
H
a
t
t
o
r
i
,
“
F
a
c
i
a
l
e
x
p
r
e
s
si
o
n
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
t
h
e
a
d
v
e
n
t
o
f
f
a
c
e
mas
k
s,
”
i
n
A
C
M
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
Pro
c
e
e
d
i
n
g
S
e
r
i
e
s
,
2
0
2
0
,
p
p
.
3
3
5
–
3
3
7
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
2
8
3
6
1
.
3
4
3
2
0
7
5
.
[
1
0
]
G
.
C
a
st
e
l
l
a
n
o
,
B
.
D
e
C
a
r
o
l
i
s,
a
n
d
N
.
M
a
c
c
h
i
a
r
u
l
o
,
“
A
u
t
o
ma
t
i
c
f
a
c
i
a
l
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
a
t
t
h
e
C
O
V
I
D
-
1
9
p
a
n
d
e
mi
c
t
i
m
e
,
”
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
.
8
2
,
n
o
.
9
,
p
p
.
1
2
7
5
1
–
1
2
7
6
9
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
022
-
1
4
0
5
0
-
0.
[
1
1
]
R
.
M
a
g
h
e
r
i
n
i
,
E.
M
u
ssi
,
M
.
S
e
r
v
i
,
a
n
d
Y
.
V
o
l
p
e
,
“
Em
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
i
n
t
h
e
t
i
m
e
s
o
f
C
O
V
I
D
1
9
:
C
o
p
i
n
g
w
i
t
h
f
a
c
e
mas
k
s,”
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
5
,
p
.
2
0
0
0
9
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
sw
a
.
2
0
2
2
.
2
0
0
0
9
4
.
[
1
2
]
Q
.
Y
a
n
g
,
Y
.
Zh
a
n
g
,
W
.
D
a
i
,
a
n
d
S
.
J
.
P
a
n
,
T
ra
n
sf
e
r
l
e
a
r
n
i
n
g
.
C
a
m
b
r
i
d
g
e
U
n
i
v
e
r
s
i
t
y
P
r
e
ss
,
2
0
2
0
.
[
1
3
]
A
.
D
o
so
v
i
t
s
k
i
y
e
t
a
l
.
,
“
A
n
i
m
a
g
e
i
s
w
o
r
t
h
1
6
x
1
6
w
o
r
d
s:
t
r
a
n
sf
o
r
m
e
r
s
f
o
r
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
a
t
sc
a
l
e
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
2
0
1
0
.
1
1
9
2
9
,
2
0
2
0
.
[
1
4
]
A
.
F
.
A
l
n
a
b
i
h
a
n
d
A
.
Y
.
M
a
g
h
a
r
i
,
“
A
r
a
b
i
c
s
i
g
n
l
a
n
g
u
a
g
e
l
e
t
t
e
r
s
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
v
i
si
o
n
t
r
a
n
sf
o
r
mer,
”
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
.
8
3
,
n
o
.
3
4
,
p
p
.
8
1
7
2
5
–
8
1
7
3
9
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
4
-
1
8
6
8
1
-
3.
[
1
5
]
Z.
H
u
a
n
g
,
Y
.
Y
u
,
a
n
d
C
.
G
o
u
,
“
D
r
i
v
e
r
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
V
i
T
a
n
d
S
t
a
r
G
A
N
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
2
0
2
1
I
EEE
1
s
t
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
D
i
g
i
t
a
l
T
w
i
n
s
a
n
d
P
a
r
a
l
l
e
l
I
n
t
e
l
l
i
g
e
n
c
e
,
D
T
PI
2
0
2
1
,
2
0
2
1
,
p
p
.
2
5
4
–
2
5
7
,
d
o
i
:
1
0
.
1
1
0
9
/
D
TPI
5
2
9
6
7
.
2
0
2
1
.
9
5
4
0
0
7
1
.
[
1
6
]
X
.
F
u
,
“
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
se
d
o
n
s
q
u
e
e
z
e
v
i
s
i
o
n
t
r
a
n
sf
o
r
mer
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
-
2
0
2
2
I
n
t
e
rn
a
t
i
o
n
a
l
S
y
m
p
o
s
i
u
m
o
n
Ad
v
a
n
c
e
s
i
n
I
n
f
o
rm
a
t
i
c
s
,
E
l
e
c
t
r
o
n
i
c
s
a
n
d
E
d
u
c
a
t
i
o
n
,
I
S
AI
EE
2
0
2
2
,
2
0
2
2
,
v
o
l
.
2
2
,
n
o
.
1
0
,
p
p
.
1
6
4
–
1
6
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
A
I
EE5
7
4
2
0
.
2
0
2
2
.
0
0
0
4
2
.
[
1
7
]
N
.
S
.
F
a
t
i
ma
e
t
a
l
.
,
“
E
n
h
a
n
c
e
d
f
a
c
i
a
l
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
v
i
si
o
n
t
r
a
n
sf
o
r
m
e
r
m
o
d
e
l
s
,
”
J
o
u
r
n
a
l
o
f
El
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
2
0
,
n
o
.
2
,
p
p
.
1
1
4
3
–
1
1
5
2
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
8
3
5
-
0
2
4
-
0
2
1
1
8
-
w.
[
1
8
]
I
.
K
o
t
s
i
a
,
I
.
B
u
c
i
u
,
a
n
d
I
.
P
i
t
a
s
,
“
A
n
a
n
a
l
y
si
s
o
f
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
u
n
d
e
r
p
a
r
t
i
a
l
f
a
c
i
a
l
i
ma
g
e
o
c
c
l
u
si
o
n
,
”
I
m
a
g
e
a
n
d
Vi
si
o
n
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
6
,
n
o
.
7
,
p
p
.
1
0
5
2
–
1
0
6
7
,
2
0
0
8
.
[
1
9
]
H
.
A
l
a
b
si
,
A
.
M
.
A
l
a
s
h
q
a
r
,
a
n
d
A
.
M
a
g
h
a
r
i
,
“
W
o
ma
n
h
i
j
a
b
d
e
t
e
c
t
i
o
n
u
s
i
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s a
n
d
D
i
g
i
t
a
l
T
e
c
h
n
o
l
o
g
i
e
s
,
v
o
l
.
7
,
n
o
.
1
,
p
p
.
1
4
5
–
1
5
6
,
2
0
2
5
.
[
2
0
]
H
.
F
e
n
g
a
n
d
J.
S
h
a
o
,
“
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
l
o
c
a
l
f
e
a
t
u
r
e
s
o
f
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
2
0
2
0
I
EEE
4
t
h
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
N
e
t
w
o
rk
i
n
g
,
El
e
c
t
r
o
n
i
c
a
n
d
A
u
t
o
m
a
t
i
o
n
C
o
n
t
ro
l
C
o
n
f
e
r
e
n
c
e
,
I
T
N
E
C
2
0
2
0
,
2
0
2
0
,
v
o
l
.
1
,
p
p
.
7
1
–
7
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
TN
EC
4
8
6
2
3
.
2
0
2
0
.
9
0
8
4
7
9
4
.
[
2
1
]
G
.
W
a
n
g
a
n
d
J.
G
o
n
g
,
“
F
a
c
i
a
l
e
x
p
r
e
s
si
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
se
d
o
n
i
mp
r
o
v
e
d
Le
N
e
t
-
5
C
N
N
,
”
i
n
2
0
1
9
C
h
i
n
e
se
C
o
n
t
ro
l
A
n
d
D
e
c
i
si
o
n
C
o
n
f
e
re
n
c
e
(
C
C
D
C
)
,
Ju
n
.
2
0
1
9
,
p
p
.
5
6
5
5
–
5
6
6
0
,
d
o
i
:
1
0
.
1
1
0
9
/
C
C
D
C
.
2
0
1
9
.
8
8
3
2
5
3
5
.
[
2
2
]
M
.
C
h
e
n
,
J.
C
h
e
n
g
,
Z
.
Z
h
a
n
g
,
Y
.
Li
,
a
n
d
Y
.
Zh
a
n
g
,
“
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
me
t
h
o
d
c
o
m
b
i
n
e
d
w
i
t
h
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm,
”
Mo
b
i
l
e
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
,
v
o
l
.
2
0
2
1
,
p
p
.
1
–
1
0
,
S
e
p
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
1
/
5
6
0
8
3
4
0
.
[
2
3
]
Y
.
Li
,
J.
Z
e
n
g
,
S
.
S
h
a
n
,
a
n
d
X
.
C
h
e
n
,
“
O
c
c
l
u
si
o
n
a
w
a
r
e
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
C
N
N
w
i
t
h
a
t
t
e
n
t
i
o
n
m
e
c
h
a
n
i
s
m,”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
m
a
g
e
Pro
c
e
ss
i
n
g
,
v
o
l
.
2
8
,
n
o
.
5
,
p
p
.
2
4
3
9
–
2
4
5
0
,
M
a
y
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
P
.
2
0
1
8
.
2
8
8
6
7
6
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
cia
l e
mo
tio
n
r
ec
o
g
n
itio
n
u
n
d
er fa
ce
ma
s
k
o
cc
lu
s
io
n
u
s
in
g
visi
o
n
…
(
A
s
h
r
a
f Yu
n
is
Ma
g
h
a
r
i
)
403
[
2
4
]
B
.
Y
a
n
g
e
t
a
l
.
,
“
F
a
c
e
-
mas
k
-
a
w
a
r
e
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
f
a
c
e
p
a
r
si
n
g
a
n
d
v
i
si
o
n
t
r
a
n
s
f
o
r
mer,”
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
L
e
t
t
e
rs
,
v
o
l
.
1
6
4
,
p
p
.
1
7
3
–
1
8
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
a
t
r
e
c
.
2
0
2
2
.
1
1
.
0
0
4
.
[
2
5
]
A
.
M
o
l
l
a
h
o
sse
i
n
i
,
B
.
H
a
sa
n
i
,
a
n
d
M
.
H
.
M
a
h
o
o
r
,
“
A
f
f
e
c
t
N
e
t
:
a
d
a
t
a
b
a
s
e
f
o
r
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
,
v
a
l
e
n
c
e
,
a
n
d
a
r
o
u
sal
c
o
mp
u
t
i
n
g
i
n
t
h
e
w
i
l
d
,
”
I
E
EE
T
r
a
n
sa
c
t
i
o
n
s
o
n
A
f
f
e
c
t
i
v
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
1
8
–
3
1
,
J
a
n
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
TA
F
F
C
.
2
0
1
7
.
2
7
4
0
9
2
3
.
[
2
6
]
A
.
A
n
w
a
r
a
n
d
A
.
R
a
y
c
h
o
w
d
h
u
r
y
,
“
M
a
sk
e
d
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
f
o
r
s
e
c
u
r
e
a
u
t
h
e
n
t
i
c
a
t
i
o
n
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rX
i
v
:
2
0
0
8
.
1
1
1
0
4
,
2
0
2
0
.
[
2
7
]
A
.
V
a
sw
a
n
i
e
t
a
l
.
,
“
A
t
t
e
n
t
i
o
n
i
s
a
l
l
y
o
u
n
e
e
d
,
”
A
d
v
a
n
c
e
s
i
n
n
e
u
r
a
l
i
n
f
o
rm
a
t
i
o
n
p
r
o
c
e
ssi
n
g
s
y
st
e
m
s
,
2
0
1
7
.
[
2
8
]
J.
D
e
n
g
,
W
.
D
o
n
g
,
R
.
S
o
c
h
e
r
,
L.
-
J.
Li
,
K
a
i
Li
,
a
n
d
Li
F
e
i
-
F
e
i
,
“
I
mag
e
N
e
t
:
A
l
a
r
g
e
-
s
c
a
l
e
h
i
e
r
a
r
c
h
i
c
a
l
i
ma
g
e
d
a
t
a
b
a
s
e
,
”
i
n
2
0
0
9
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
Ju
n
.
2
0
0
9
,
p
p
.
2
4
8
–
2
5
5
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
0
9
.
5
2
0
6
8
4
8
.
[
2
9
]
B
.
Y
a
n
g
,
W
.
J
i
a
n
mi
n
g
,
a
n
d
G
.
H
a
t
t
o
r
i
,
“
F
a
c
e
ma
sk
a
w
a
r
e
r
o
b
u
s
t
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
d
u
r
i
n
g
t
h
e
C
O
V
I
D
-
1
9
p
a
n
d
e
m
i
c
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
-
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
m
a
g
e
P
ro
c
e
ss
i
n
g
,
I
C
I
P
,
2
0
2
1
,
v
o
l
.
2
0
2
1
-
S
e
p
t
e
,
p
p
.
2
4
0
–
2
4
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
P
4
2
9
2
8
.
2
0
2
1
.
9
5
0
6
0
4
7
.
[
3
0
]
M
.
M
u
k
h
i
d
d
i
n
o
v
,
O
.
D
j
u
r
a
e
v
,
F
.
A
k
h
me
d
o
v
,
A
.
M
u
k
h
a
ma
d
i
y
e
v
,
a
n
d
J.
C
h
o
,
“
M
a
s
k
e
d
f
a
c
e
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
se
d
o
n
f
a
c
i
a
l
l
a
n
d
m
a
r
k
s
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
f
o
r
v
i
s
u
a
l
l
y
i
mp
a
i
r
e
d
p
e
o
p
l
e
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
3
,
n
o
.
3
,
p
.
1
0
8
0
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
3
0
3
1
0
8
0
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
As
h
r
a
f
Yun
is
M
a
g
h
a
r
i
is
a
n
a
ss
o
c
iate
p
ro
fe
ss
o
r
o
f
c
o
m
p
u
ter
sc
ien
c
e
a
t
th
e
Isla
m
ic
Un
iv
e
rsity
o
f
G
a
z
a
(IUG
)
.
He
h
o
ld
s
a
P
h
.
D.
in
c
o
m
p
u
ter
v
isio
n
a
n
d
ima
g
e
p
r
o
c
e
ss
in
g
fro
m
Un
i
v
e
rsiti
S
a
i
n
s
M
a
lay
sia
(USM
).
He
h
a
s
e
x
ten
si
v
e
re
se
a
rc
h
e
x
p
e
rien
c
e
i
n
c
o
m
p
u
ter
sc
ien
c
e
field
s
su
c
h
a
s
d
a
ta
m
in
in
g
,
ima
g
e
p
ro
c
e
ss
in
g
,
c
o
m
p
u
ter
v
i
sio
n
,
a
n
d
d
e
e
p
lea
rn
i
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
m
a
g
h
a
ri@i
u
g
a
z
a
.
e
d
u
.
p
s
.
Am
e
e
r
M.
Te
lb
a
n
i
wa
s
a
lec
tu
re
r
in
th
e
M
u
lt
ime
d
ia
De
p
a
rtme
n
t
a
t
th
e
Isla
m
ic
Un
iv
e
rsity
o
f
G
a
z
a
(IUG
).
He
h
e
ld
a
M
a
ste
r’s
d
e
g
re
e
in
i
n
fo
rm
a
ti
o
n
tec
h
n
o
l
o
g
y
fro
m
IUG
.
He
h
a
d
a
c
a
d
e
m
ic
e
x
p
e
rien
c
e
in
m
u
lt
ime
d
ia
tec
h
n
o
l
o
g
y
d
isc
ip
li
n
e
s
su
c
h
a
s
ima
g
e
p
ro
c
e
ss
in
g
,
3
D
m
o
d
e
li
n
g
,
a
n
d
a
n
ima
ti
o
n
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
c
o
m
p
u
ter
v
isi
o
n
,
m
u
lt
ime
d
ia
a
p
p
li
c
a
ti
o
n
s,
a
n
d
d
e
e
p
lea
rn
i
n
g
.
T
h
is
a
rti
c
le
is
b
a
se
d
o
n
h
is
m
a
ste
r’
s
th
e
sis
wo
rk
.
He
trag
ica
ll
y
p
a
ss
e
d
a
wa
y
d
u
rin
g
th
e
wa
r
in
Ga
z
a
b
e
fo
re
th
e
c
o
m
p
letio
n
o
f
th
is
re
se
a
rc
h
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
m
y
a
sh
ra
f2
@
g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.