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.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
4
7
6
2
~
4
7
7
3
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
4
7
6
2
-
4
7
7
3
4762
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Dete
c
ting a
utism
with
Vietnam
ese
child f
a
cia
l ima
g
e
s usin
g
deep learning
T
ra
n Va
n T
ha
n
h
1
,
L
a
m
T
ha
nh
H
ien
2
,
Do
Na
ng
K
ho
a
3
,
L
e
Anh T
u
4
,
H
a
M
a
nh
T
o
a
n
5
,
Do
Na
ng
T
o
a
n
5
1
F
a
c
u
l
t
y
o
f
M
e
c
h
a
t
r
o
n
i
c
s
a
n
d
E
l
e
c
t
r
o
n
i
c
s,
La
c
H
o
n
g
U
n
i
v
e
r
si
t
y
,
B
i
e
n
H
o
a
,
V
i
e
t
n
a
m
2
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
La
c
H
o
n
g
U
n
i
v
e
r
si
t
y
,
B
i
e
n
H
o
a
,
V
i
e
t
n
a
m
3
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
H
a
n
o
i
A
r
c
h
i
t
e
c
t
u
r
a
l
U
n
i
v
e
r
si
t
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
4
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
H
a
L
o
n
g
U
n
i
v
e
r
s
i
t
y
,
U
o
n
g
B
i
,
V
i
e
t
n
a
m
5
I
n
st
i
t
u
t
e
o
f
I
n
f
o
r
ma
t
i
o
n
Te
c
h
n
o
l
o
g
y
,
V
i
e
t
n
a
m
A
c
a
d
e
my
o
f
S
c
i
e
n
c
e
a
n
d
Te
c
h
n
o
l
o
g
y
,
H
a
n
o
i
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
1
2
,
2
0
2
5
R
ev
is
ed
J
u
n
5
,
2
0
2
5
Acc
ep
ted
J
u
l 3
,
2
0
2
5
De
e
p
lea
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
c
re
a
te
d
a
s
i
g
n
i
fic
a
n
t
in
c
re
a
s
e
i
n
in
te
ll
i
g
e
n
t
s
y
st
e
m
s
,
e
sp
e
c
ia
ll
y
i
n
t
h
e
m
e
d
ic
a
l
f
iel
d
.
Am
o
n
g
m
e
n
ta
l
p
r
o
b
lem
s,
a
u
t
ism
is
a
d
a
n
g
e
r
o
u
s
n
e
u
r
o
d
e
v
e
l
o
p
m
e
n
tal
d
is
o
r
d
e
r
a
n
d
it
n
e
e
d
s
t
o
b
e
d
ia
g
n
o
se
d
e
a
rl
y
b
e
c
a
u
s
e
o
f
t
h
e
m
a
ll
e
a
b
il
i
t
y
o
f
c
h
i
l
d
b
ra
i
n
d
e
v
e
l
o
p
m
e
n
t.
I
n
o
u
r
st
u
d
y
,
we
fo
c
u
se
d
o
n
a
u
t
ism
d
e
tec
t
i
o
n
b
y
u
si
n
g
t
h
e
Vi
e
t
n
a
m
e
se
fa
c
ial
c
h
il
d
i
m
a
g
e
a
n
d
st
u
d
ie
d
t
h
e
r
o
le
o
f
in
te
r
n
a
t
i
o
n
a
l
d
a
ta
a
n
d
V
ie
tn
a
m
e
se
d
a
ta
wh
e
n
a
p
p
l
y
i
n
g
d
e
e
p
lea
r
n
i
n
g
a
p
p
r
o
a
c
h
t
o
d
ia
g
n
o
se
a
u
t
ism
.
T
o
d
o
t
h
a
t
,
we
p
r
o
p
o
se
d
d
if
fe
re
n
t
stra
te
g
ies
b
a
se
d
o
n
o
u
r
h
y
p
o
t
h
e
si
s
a
b
o
u
t
fa
c
t
o
r
s
o
f
th
e
t
ra
n
sfe
r
lea
rn
i
n
g
a
n
d
tra
i
n
i
n
g
se
t
t
y
p
e
s
.
T
o
c
o
n
d
u
c
t
t
h
e
e
x
p
e
r
ime
n
t,
we
p
re
p
a
re
d
a
Vie
tn
a
m
e
se
fa
c
ia
l
c
h
i
l
d
ima
g
e
se
t
fr
o
m
se
v
e
ra
l
k
i
n
d
e
r
g
a
r
te
n
s
i
n
H
o
C
h
i
M
i
n
h
C
it
y
,
Vie
tn
a
m
a
n
d
w
e
a
p
p
li
e
d
d
if
fe
re
n
t
d
e
e
p
a
rc
h
it
e
c
t
u
re
s
s
u
c
h
a
s
R
e
sN
e
t,
De
n
se
Ne
t,
a
n
d
A
le
x
Ne
t
in
t
h
e
a
u
ti
sm
c
la
ss
i
fic
a
t
i
o
n
e
x
p
e
r
ime
n
t
w
it
h
b
o
t
h
Vie
tn
a
m
e
se
a
n
d
i
n
t
e
r
n
a
ti
o
n
a
l
fa
c
ial
c
h
il
d
im
a
g
e
s.
We
a
n
a
l
y
z
e
d
im
p
o
rta
n
t
fa
c
t
o
rs
f
r
o
m
t
h
e
e
x
p
e
r
ime
n
t
re
su
lt
s
wi
t
h
a
re
a
u
n
d
e
r
t
h
e
c
u
r
v
e
(
AUC
)
,
acc
u
ra
c
y
,
se
n
s
i
ti
v
it
y
,
a
n
d
s
p
e
c
i
fic
i
ty
,
i
n
c
l
u
d
i
n
g
a
p
p
l
y
i
n
g
tr
a
n
sfe
r
lea
rn
i
n
g
a
n
d
th
e
a
p
p
e
a
ra
n
c
e
o
f
V
ie
tn
a
m
e
se
d
a
ta
i
n
th
e
tra
i
n
i
n
g
se
t
.
Be
s
i
d
e
s,
we
a
ls
o
d
i
sc
u
ss
e
d
t
h
e
d
if
fe
re
n
c
e
o
f
i
n
te
r
n
a
ti
o
n
a
l
a
n
d
V
ie
t
n
a
m
e
se
d
a
ta
d
o
m
a
i
n
s
.
T
h
e
e
x
p
o
s
u
r
e
o
f
d
a
ta
d
ist
ri
b
u
ti
o
n
d
i
ffe
r
e
n
c
e
s
i
n
t
h
e
p
r
o
p
o
se
d
s
tra
te
g
i
e
s
a
ls
o
h
i
g
h
li
g
h
ts
t
h
e
im
p
o
r
ta
n
c
e
o
f
c
o
l
le
c
ti
n
g
fa
c
ia
l
d
a
ta
o
f
V
iet
n
a
m
e
se
c
h
il
d
re
n
.
K
ey
w
o
r
d
s
:
Au
tis
tic
d
etec
tio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Gen
er
aliza
tio
n
T
r
an
s
f
er
lear
n
i
n
g
Vietn
am
ese
f
ac
ial
ch
ild
im
ag
e
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
Nan
g
T
o
a
n
I
n
s
titu
te
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
Vietn
a
m
Aca
d
em
y
o
f
S
cien
ce
an
d
T
ec
h
n
o
lo
g
y
18
Ho
an
g
Qu
o
c
Viet,
C
au
Gia
y
Dis
tr
ict,
Han
o
i1
0
0
7
2
,
Vietn
a
m
E
m
ail: d
n
to
an
@
io
it.a
c.
v
n
1.
I
NT
RO
D
UCT
I
O
N
Au
tis
m
is
a
n
eu
r
o
d
e
v
elo
p
m
e
n
tal
d
is
o
r
d
er
ch
ar
ac
ter
ized
b
y
a
lack
o
f
s
o
cial
co
m
m
u
n
i
ca
tio
n
an
d
r
ep
etitiv
e
b
eh
av
io
r
s
,
b
o
th
v
er
b
al
an
d
p
h
y
s
ical
[
1
]
.
Au
tis
tic
s
y
n
d
r
o
m
e
h
as
d
if
f
icu
lt
y
in
ter
ac
tin
g
an
d
co
m
m
u
n
icatin
g
with
o
th
er
s
.
S
y
m
p
to
m
s
o
f
a
u
tis
m
ca
n
o
cc
u
r
in
m
an
y
s
itu
atio
n
s
an
d
i
n
m
an
y
p
lace
s
’
d
if
f
er
e
n
t
p
lace
s
.
T
h
is
d
is
ea
s
e
in
v
o
lv
es
ab
n
o
r
m
al
b
r
ain
d
ev
elo
p
m
en
t
t
h
at
af
f
ec
ts
f
ac
ial
ex
p
r
ess
io
n
s
an
d
p
h
y
s
ical
s
tates
[
2
]
.
C
h
ild
r
en
with
au
tis
m
h
a
v
e
u
n
u
s
u
al
f
a
cial
f
ea
tu
r
es,
w
h
ich
s
ets
th
em
ap
ar
t
f
r
o
m
ty
p
ically
d
ev
el
o
p
in
g
ch
ild
r
en
.
T
h
is
d
is
ea
s
e
n
o
t
o
n
l
y
ca
u
s
es
d
if
f
icu
lties
f
o
r
th
e
p
a
tien
t
b
u
t
also
f
o
r
th
eir
lo
v
e
d
o
n
es
in
d
aily
life
[
3
]
.
Fo
r
p
ar
en
ts
wh
o
h
av
e
h
ad
a
ch
ild
with
au
tis
m
,
th
e
r
is
k
o
f
h
av
in
g
an
o
th
e
r
ch
ild
with
au
tis
m
is
ap
p
r
o
x
im
ately
3
to
1
0
%
[
4
]
.
T
h
e
r
e
h
av
e
b
ee
n
r
ec
en
t
co
n
ce
r
n
s
ab
o
u
t
r
is
in
g
au
tis
m
r
ates.
T
h
e
W
o
r
ld
Hea
lth
Or
g
an
izatio
n
esti
m
ates
th
at
th
e
p
r
ev
alen
ce
o
f
au
tis
m
is
0
.
6
7
%
wo
r
ld
w
id
e,
b
u
t
th
is
f
ig
u
r
e
is
o
n
ly
r
ep
o
r
ted
in
1
6
%
o
f
ch
ild
r
en
g
lo
b
ally
[
5
]
.
T
h
e
ce
n
ter
s
f
o
r
d
is
ea
s
e
co
n
tr
o
l
an
d
p
r
ev
en
tio
n
esti
m
ates
th
at
ab
o
u
t
1
.
6
8
%
o
f
US
ch
ild
r
en
ag
e
d
8
y
ea
r
s
an
d
o
ld
e
r
ar
e
d
iag
n
o
s
ed
with
au
tis
m
,
w
h
ich
is
1
in
5
9
ch
ild
r
en
[
6
]
.
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
Dete
ctin
g
a
u
tis
m
w
ith
V
ietn
a
mese
ch
ild
fa
cia
l ima
g
es
u
s
in
g
…
(
Tr
a
n
V
a
n
Th
a
n
h
)
4763
E
ar
ly
d
iag
n
o
s
is
o
f
a
u
tis
m
is
ess
en
tial.
C
h
ild
r
en
s
h
o
win
g
s
ig
n
s
o
f
a
u
tis
m
ca
n
b
en
ef
it
f
r
o
m
ea
r
ly
d
iag
n
o
s
is
d
u
e
to
th
e
m
allea
b
ilit
y
o
f
th
eir
b
r
ain
d
e
v
elo
p
m
e
n
t,
wh
ich
ca
n
h
el
p
th
em
im
p
r
o
v
e
th
eir
s
o
cial
life
.
L
ate
d
iag
n
o
s
is
ca
n
r
ed
u
ce
th
e
ef
f
ec
tiv
e
n
ess
o
f
t
r
ea
tm
en
ts
f
o
r
p
atien
ts
with
au
tis
m
[
7
]
.
Ac
co
r
d
in
g
to
a
r
ec
en
t
s
tu
d
y
,
ch
ild
r
e
n
wh
o
r
ec
ei
v
ed
m
ed
ical
ca
r
e
b
ef
o
r
e
th
e
a
g
e
o
f
two
h
ad
a
h
ig
h
e
r
in
tellig
en
ce
q
u
o
tien
t
t
h
an
ch
ild
r
en
wh
o
d
id
n
o
t r
ec
ei
v
e
m
ed
ical
ca
r
e
u
n
til later
in
life
[
8
]
.
T
h
er
e
h
av
e
b
ee
n
s
ev
er
al
ar
tific
ial
in
tellig
en
ce
s
tu
d
ies
r
elatin
g
to
au
tis
m
d
ia
g
n
o
s
is
.
I
n
2
0
1
8
,
Hein
s
f
eld
et
a
l.
[
9
]
s
tu
d
ied
a
u
tis
m
class
if
icatio
n
with
b
r
ai
n
m
ag
n
etic
r
eso
n
a
n
ce
im
ag
in
g
d
ata.
I
n
t
h
e
s
tu
d
y
,
th
e
au
th
o
r
s
u
s
ed
a
cu
s
to
m
d
e
ep
n
eu
r
al
n
etwo
r
k
m
o
d
el
c
o
m
b
in
ed
with
two
s
tack
ed
d
en
o
is
in
g
au
to
en
co
d
er
s
an
d
ex
p
er
im
en
ts
u
s
in
g
1
0
-
f
o
l
d
cr
o
s
s
-
v
alid
atio
n
.
Ah
m
ed
et
a
l
.
[
1
0
]
h
ad
r
esear
ch
aim
in
g
a
t
d
iag
n
o
s
in
g
a
u
tis
m
with
v
ar
io
u
s
task
s
s
u
ch
as
r
e
m
o
v
in
g
all
n
o
is
e
f
r
o
m
th
e
ey
e
p
ath
ar
ea
,
ex
tr
ac
tin
g
th
e
p
ath
o
f
ey
e
p
o
in
ts
o
n
th
e
im
ag
e,
b
u
ild
in
g
a
class
if
icatio
n
m
o
d
el,
an
d
ev
alu
atin
g
.
T
o
p
er
f
o
r
m
th
ese
task
s
,
t
h
e
a
u
th
o
r
s
u
s
ed
s
ev
er
al
tech
n
iq
u
es
s
u
ch
as
lo
ca
l
b
in
a
r
y
s
am
p
les,
an
d
g
r
ay
-
lev
el
c
o
-
o
cc
u
r
r
en
ce
m
atr
ices.
B
esid
es,
th
e
au
th
o
r
s
also
co
n
d
u
cte
d
ex
p
er
im
e
n
ts
o
n
s
o
m
e
d
ee
p
lear
n
in
g
m
o
d
els
s
u
ch
as
Go
o
g
le
-
N
et
an
d
R
esNet
-
1
8
.
Far
o
o
q
et
a
l.
[
1
1
]
p
r
esen
ted
a
s
tu
d
y
u
s
in
g
f
ed
er
ated
lear
n
in
g
to
d
iag
n
o
s
e
au
tis
m
.
I
n
th
eir
wo
r
k
,
th
e
a
u
th
o
r
s
ap
p
lied
two
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
m
o
d
els,
lo
g
is
tic
r
e
g
r
ess
io
n
a
n
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e,
t
o
d
ata
f
r
o
m
v
a
r
io
u
s
s
o
u
r
ce
s
.
I
n
th
e
ex
p
er
im
en
t,
th
e
au
th
o
r
s
u
s
ed
m
o
r
e
th
an
6
0
0
tab
u
lar
d
ata
r
ec
o
r
d
s
an
d
ac
h
ie
v
ed
an
ac
cu
r
ac
y
o
f
9
8
% a
n
d
8
1
% f
o
r
two
g
r
o
u
p
s
o
f
ch
ild
r
e
n
an
d
ad
u
lts
,
r
esp
ec
t
iv
ely
.
I
n
Vietn
am
,
th
er
e
h
av
e
b
ee
n
s
ev
er
al
attem
p
ts
b
y
th
e
g
o
v
er
n
m
en
t
an
d
co
m
m
u
n
ity
r
elatin
g
to
au
tis
m
.
I
n
2
0
2
2
,
th
e
Min
is
tr
y
o
f
L
ab
o
r
-
I
n
v
alid
s
an
d
So
cial
Af
f
a
ir
s
an
n
o
u
n
ce
d
n
ews
r
esp
o
n
d
i
n
g
to
wo
r
l
d
au
tis
m
awa
r
en
ess
d
ay
[
1
2
]
.
T
h
at
n
ew
s
p
r
esen
ted
a
p
r
o
g
r
am
in
Ho
C
h
i
Min
h
C
ity
th
at
allo
ws
th
e
co
m
m
u
n
ity
to
g
et
a
b
etter
u
n
d
er
s
tan
d
i
n
g
o
f
au
tis
m
an
d
s
u
p
p
o
r
t
au
tis
tic
ch
ild
r
e
n
.
I
n
th
is
p
ap
er
,
o
u
r
s
tu
d
y
is
p
u
t
in
th
e
co
n
tex
t
o
f
au
tis
m
in
Vietn
am
ese
ch
ild
r
en
.
W
e
aim
to
au
tis
m
d
etec
tio
n
in
Vietn
am
ese
ch
ild
r
en
with
m
o
d
er
n
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es
b
y
u
s
in
g
f
ac
ial
im
a
g
e
d
ata
th
at
ca
n
b
e
ca
p
tu
r
e
d
b
y
u
s
in
g
a
n
o
r
m
al
ca
m
er
a
i
n
s
m
ar
tp
h
o
n
es.
Sev
er
al
s
tu
d
ies
wer
e
u
s
in
g
d
e
ep
lear
n
in
g
m
eth
o
d
s
f
o
r
a
u
tis
m
d
etec
tio
n
.
Kar
r
i
et
a
l.
[
1
3
]
p
u
b
lis
h
ed
a
s
tu
d
y
u
s
in
g
a
d
ee
p
-
lear
n
in
g
ap
p
r
o
ac
h
to
class
if
y
au
tis
m
b
ased
o
n
f
ac
ial
im
ag
es.
I
n
th
e
ex
p
er
im
en
t,
th
e
au
th
o
r
s
u
s
ed
th
e
d
ee
p
lear
n
i
n
g
m
o
d
el
Den
s
e
-
N
et
alo
n
g
with
a
d
ata
s
et
o
f
f
ac
e
im
ag
es
d
o
wn
lo
ad
ed
f
r
o
m
th
e
Kag
g
le
p
latf
o
r
m
.
R
ab
b
i
et
a
l
.
[
1
4
]
p
r
esen
ted
th
e
wo
r
k
f
o
r
a
u
tis
m
class
if
icatio
n
test
ed
wit
h
a
d
ataset
o
f
2
,
9
3
6
im
ag
es
p
r
o
v
id
ed
b
y
th
e
Ka
g
g
l
e
p
latf
o
r
m
.
I
n
th
at
s
tu
d
y
,
th
e
au
th
o
r
s
ap
p
lied
t
h
e
tr
an
s
f
e
r
le
ar
n
in
g
ap
p
r
o
ac
h
to
VGG
1
9
,
I
n
ce
p
tio
n
V3
,
an
d
Den
s
e
-
N
et
2
0
1
m
o
d
els
an
d
u
s
ed
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
AUC
(
Ar
ea
u
n
d
e
r
th
e
R
OC
cu
r
v
e)
to
ev
alu
ate
th
eir
r
esu
lts
.
L
i
et
a
l.
[
1
5
]
p
r
esen
ted
a
s
tu
d
y
u
s
in
g
th
e
tr
an
s
f
e
r
lear
n
in
g
ap
p
r
o
ac
h
with
Mo
b
ileNetV2
an
d
Mo
b
ileNetv
3
-
L
ar
g
e
n
etwo
r
k
s
to
id
en
tify
au
tis
tic
ch
ild
r
e
n
b
ased
o
n
f
ac
ial
im
ag
es.
T
h
e
au
th
o
r
s
b
u
ilt
a
f
r
am
ewo
r
k
o
f
f
ac
ial
im
ag
e
class
if
icatio
n
co
m
b
in
in
g
th
e
two
-
p
h
ase
tr
an
s
f
er
lear
n
in
g
an
d
th
e
m
u
lti
-
class
if
ier
in
teg
r
atio
n
.
I
n
th
e
ex
p
er
im
en
t,
th
eir
wo
r
k
r
ea
c
h
ed
0
.
8
8
3
3
ac
c
u
r
ac
y
f
o
r
Mo
b
ileNetV2
an
d
0
.
8
7
6
7
ac
cu
r
ac
y
f
o
r
Mo
b
ileNetV3
-
L
ar
g
e.
An
o
th
er
s
tu
d
y
p
u
b
lis
h
ed
in
2
0
2
3
b
y
Gh
az
al
et
a
l.
[
1
6
]
.
T
h
e
au
th
o
r
s
p
r
o
p
o
s
ed
a
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
et
wo
r
k
b
ased
o
n
Alex
-
N
et
with
f
ac
ial
im
ag
es
as
th
e
in
p
u
t.
T
h
ei
r
s
tu
d
y
aim
e
d
at
th
e
r
o
b
u
s
t
ex
tr
ac
tio
n
o
f
n
u
m
er
o
u
s
f
ac
ial
f
ea
tu
r
es,
wh
ich
is
a
d
if
f
icu
lt
task
b
ec
au
s
e
o
f
th
e
s
u
b
tlety
r
e
q
u
ir
em
e
n
ts
.
I
n
th
e
ex
p
e
r
im
en
t,
th
at
wo
r
k
r
e
ac
h
ed
a
v
alid
atio
n
ac
cu
r
ac
y
o
f
0
.
8
7
7
,
a
v
alid
atio
n
s
en
s
itiv
ity
o
f
0
.
8
7
6
,
a
n
d
a
v
alid
atio
n
s
p
ec
if
icity
o
f
0
.
8
7
6
.
R
ed
d
y
a
n
d
An
d
r
ew
[
1
7
]
p
u
b
lis
h
e
d
a
p
ap
e
r
o
n
u
s
in
g
th
e
d
ee
p
lea
r
n
in
g
ap
p
r
o
ac
h
t
o
id
en
tify
au
tis
tic
ch
ild
r
e
n
.
T
h
e
au
th
o
r
s
u
s
ed
t
h
r
ee
m
o
d
els
VGG1
6
,
VGG1
9
,
a
n
d
E
f
f
icien
t
N
etB
0
with
p
r
e
-
tr
ai
n
ed
weig
h
ts
in
th
e
f
r
am
ew
o
r
k
o
f
a
p
p
ly
in
g
tr
an
s
f
er
le
ar
n
in
g
.
Fro
m
t
h
e
ex
p
er
im
en
ts
,
th
e
a
u
th
o
r
s
ac
h
iev
ed
an
ac
c
u
r
ac
y
o
f
8
4
.
6
6
%
f
o
r
th
e
VGG1
6
m
o
d
el,
8
0
.
0
5
%
f
o
r
th
e
VGG1
9
m
o
d
el,
a
n
d
8
7
.
9
%
f
o
r
th
e
E
f
f
icien
t
N
etB
0
m
o
d
el.
Als
o
,
Ah
m
ad
et
a
l.
[
1
8
]
p
u
b
lis
h
ed
a
p
ap
er
u
s
in
g
d
if
f
er
e
n
t
m
o
d
els
s
u
ch
as
Alex
Net,
Mo
b
ileNetV2
,
R
esNet3
4
,
R
esNe
t5
0
,
VGG1
6
,
an
d
VGG1
9
to
d
etec
t
au
tis
m
f
r
o
m
f
ac
ial
im
ag
es
an
d
an
aly
ze
th
eir
r
esu
lts
.
I
n
th
e
ex
p
er
im
en
t,
th
e
tr
ain
in
g
tim
e
was
ap
p
r
o
x
i
m
ately
2
h
o
u
r
s
,
an
d
th
e
test
in
g
tim
e
was
n
ea
r
ly
3
m
in
u
tes.
I
n
r
esu
lts
,
R
esNet3
4
r
ea
ch
ed
an
ac
cu
r
ac
y
s
co
r
e
s
u
r
p
ass
in
g
0
.
8
6
with
2
4
8
x
2
4
8
r
eso
lu
tio
n
a
n
d
0
.
8
3
with
1
2
4
×
1
2
4
r
eso
lu
tio
n
.
An
im
p
o
r
tan
t
is
s
u
e
we
wan
t
to
ex
p
lo
r
e
is
th
e
d
if
f
er
e
n
ce
in
th
e
d
ata
d
o
m
ain
o
f
Vietn
am
ese
an
d
in
ter
n
atio
n
al
ch
ild
f
ac
ial
im
ag
es.
Ou
r
h
y
p
o
th
esis
is
th
at
t
h
is
d
if
f
er
en
ce
will
h
av
e
a
s
ig
n
if
ican
t
im
p
ac
t
an
d
th
er
ef
o
r
e
t
h
e
p
r
esen
ce
o
f
Vietn
am
ese
ch
ild
f
ac
ial
im
ag
e
d
at
a
in
th
e
s
tep
s
o
f
d
ev
elo
p
in
g
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
th
is
p
r
o
b
lem
.
I
n
f
ac
t,
m
a
n
y
s
tu
d
ies
o
n
f
ac
ial
m
o
r
p
h
o
lo
g
y
an
d
an
at
o
m
y
s
h
o
w
th
at
Vietn
am
ese
f
ac
es
h
a
v
e
d
is
tin
ct
ch
ar
ac
ter
is
tics
.
T
h
e
s
t
u
d
y
[
1
9
]
co
m
p
ar
ed
Vietn
am
e
s
e
p
eo
p
le
with
No
r
th
Am
er
ican
wh
ite
p
eo
p
le
an
d
s
h
o
wed
th
at
Vietn
am
ese
p
eo
p
le
h
av
e
a
wid
e
r
d
is
tan
ce
b
e
twee
n
th
e
two
ey
e
c
o
r
n
er
s
,
a
lar
g
er
n
o
s
e,
b
u
t
a
s
m
aller
m
o
u
th
wid
th
.
C
o
m
p
a
r
ed
with
o
th
er
Asi
an
g
r
o
u
p
s
,
r
esear
ch
[
2
0
]
s
h
o
ws
th
at
C
h
in
ese
p
eo
p
le
ten
d
to
h
av
e
s
m
aller
f
o
r
eh
ea
d
s
,
h
ig
h
er
n
o
s
es,
an
d
a
s
m
aller
r
a
tio
b
etwe
en
n
o
s
e
len
g
th
an
d
f
ac
e
h
eig
h
t
th
a
n
Vietn
am
ese
p
eo
p
le.
I
n
o
u
r
r
esear
ch
,
we
aim
to
ap
p
ly
a
d
ee
p
lear
n
i
n
g
a
p
p
r
o
ac
h
to
e
f
f
ec
tiv
ely
class
if
y
au
tis
m
f
r
o
m
Vietn
am
ese
ch
ild
r
en
'
s
d
ata.
On
th
e
o
n
e
h
a
n
d
,
we
o
f
ten
n
ee
d
a
lar
g
e
en
o
u
g
h
d
ata
s
et
f
o
r
ef
f
ec
tiv
e
tr
ain
in
g
in
d
ee
p
lear
n
in
g
m
eth
o
d
s
.
On
th
e
o
th
er
h
an
d
,
co
llectin
g
a
lar
g
e
en
o
u
g
h
Vietn
am
ese
ch
ild
r
en
d
ataset
is
n
o
t
ea
s
y
in
r
ea
l
-
life
c
o
n
d
itio
n
s
.
T
h
e
ac
ce
s
s
r
eq
u
ir
es
th
e
p
er
m
is
s
io
n
a
n
d
s
u
p
p
o
r
t
o
f
th
e
c
h
ild
r
en
'
s
p
a
r
en
ts
as
well
as
th
e
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
6
2
-
4
7
7
3
4764
s
u
p
p
o
r
t
o
f
th
e
teac
h
e
r
s
at
th
e
k
in
d
er
g
ar
te
n
.
T
h
er
ef
o
r
e,
we
d
esig
n
e
d
s
tr
ateg
ies
r
el
atin
g
to
ap
p
ly
i
n
g
in
ter
n
atio
n
al
d
ata
in
g
en
er
alizi
n
g
Vietn
am
ese
f
ac
ial
im
ag
es.
B
esid
es,
b
ec
au
s
e
o
u
r
s
tu
d
y
ai
m
s
to
b
u
ild
a
s
tate
-
of
-
th
e
-
a
r
t
s
o
lu
tio
n
f
o
r
th
e
au
ti
s
m
d
etec
tio
n
p
r
o
b
lem
in
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
im
ag
es
,
we
also
co
llected
d
ata
f
r
o
m
s
ev
er
al
lo
ca
l
Viet
n
am
ese
k
in
d
er
g
ar
ten
s
th
at
h
av
e
b
o
th
au
tis
tic
ch
ild
r
en
a
n
d
n
o
r
m
al
ch
ild
r
e
n
lear
n
in
g
.
T
h
e
Vietn
am
ese
d
a
taset
wo
u
ld
p
lay
an
im
p
o
r
ta
n
t
r
o
le
in
e
v
alu
atin
g
o
u
r
d
if
f
er
en
t
s
tr
ateg
ies.
I
n
d
etail,
we
d
esig
n
d
i
f
f
er
en
t
s
tr
ateg
ies
r
eg
ar
d
in
g
h
o
w
to
ap
p
ly
in
ter
n
atio
n
al
d
ata
t
o
th
e
p
r
o
b
lem
o
f
au
tis
m
d
iag
n
o
s
is
o
n
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
im
ag
es.
B
ased
o
n
s
ep
ar
ate
ev
alu
atio
n
e
v
id
en
ce
o
n
in
ter
n
atio
n
al
an
d
Vietn
am
ese
d
ata,
we
d
elv
e
in
to
th
e
im
p
o
r
tan
t
f
ac
to
r
s
to
b
u
ild
an
ef
f
e
ctiv
e
s
o
lu
tio
n
s
u
ch
as
h
o
w
to
u
s
e
p
r
e
-
tr
ain
ed
weig
h
ts
as
well
as
th
e
r
o
le
o
f
Vietn
am
ese
d
ata
in
t
h
e
tr
ain
in
g
p
h
ase.
T
h
e
ev
i
d
en
ce
also
r
ev
ea
ls
th
e
lim
itatio
n
s
o
f
ex
is
tin
g
in
ter
n
at
io
n
al
d
ata
in
g
en
er
alizin
g
Vietn
am
ese
f
ac
ial
im
ag
es.
I
n
s
h
o
r
t,
th
e
m
ain
c
o
n
tr
ib
u
tio
n
s
wer
e
d
escr
ib
e
d
as
f
o
llo
ws.
Firstl
y
,
we
s
et
u
p
a
r
e
ad
y
-
to
-
u
s
e
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
im
ag
e
d
ataset
f
o
r
th
e
au
ti
s
m
d
etec
tio
n
.
Seco
n
d
ly
,
we
p
r
o
p
o
s
ed
d
i
f
f
er
en
t
s
tr
ateg
ies
r
elatin
g
to
ap
p
ly
in
g
in
ter
n
atio
n
al
d
ata
in
g
en
e
r
alizin
g
Vietn
am
ese
f
ac
ial
im
ag
es.
L
astl
y
,
we
an
aly
ze
d
th
e
ex
p
er
im
e
n
t
r
esu
lts
o
f
th
e
s
tr
ateg
ie
s
to
d
ee
p
ly
d
elv
e
in
to
im
p
o
r
tan
t
f
ac
to
r
s
s
u
ch
as
o
p
tio
n
s
o
f
u
s
in
g
p
r
e
-
tr
ain
ed
weig
h
ts
an
d
th
e
d
if
f
er
e
n
ce
o
f
in
ter
n
atio
n
al
an
d
Vietn
am
ese
d
ata
d
o
m
ain
s
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
p
a
p
er
is
p
r
esen
ted
as
f
o
llo
ws:
Sectio
n
2
is
th
e
b
ac
k
g
r
o
u
n
d
o
f
d
ee
p
-
lear
n
in
g
ar
ch
itectu
r
e
m
o
d
els
f
o
r
th
e
p
r
o
p
o
s
ed
s
y
s
tem
.
Sectio
n
3
s
h
o
ws
th
e
p
r
o
p
o
s
ed
s
tu
d
y
.
Secti
o
n
4
d
escr
ib
es
an
d
an
aly
ze
s
th
e
r
esu
lts
o
f
th
e
e
x
p
er
im
en
t.
I
n
th
e
en
d
,
s
ec
tio
n
5
co
n
clu
d
es o
u
r
s
tu
d
y
.
2.
DE
E
P
-
L
E
A
RNING
AR
CH
I
T
E
C
T
UR
E
S
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
n
ee
d
s
s
o
m
e
d
ee
p
-
lear
n
in
g
ar
ch
itectu
r
e
m
o
d
els.
T
o
co
n
d
u
ct
ex
p
er
i
m
en
ts
,
we
u
s
ed
s
o
m
e
p
o
p
u
lar
d
ee
p
-
lear
n
in
g
m
o
d
els
s
u
ch
as
R
es
-
Net
3
4
,
R
es
-
Net
5
0
,
Alex
-
Net,
an
d
Den
s
e
-
Net.
T
h
ese
m
o
d
els we
r
e
ap
p
lied
s
u
cc
ess
f
u
lly
in
a
lo
t
o
f
ar
tific
ial
in
tellig
en
ce
r
esear
ch
.
Kr
izh
ev
s
k
y
et
a
l.
[
2
1
]
p
r
esen
ted
a
n
ew
d
ee
p
ar
ch
itectu
r
e
an
d
th
is
m
o
d
el
wo
n
th
e
im
a
g
e
n
et
lar
g
e
s
ca
le
v
is
u
al
r
ec
o
g
n
itio
n
ch
allen
g
e
2
0
1
2
.
T
h
e
au
th
o
r
s
r
ea
ch
e
d
a
to
p
-
5
er
r
o
r
r
esu
lt o
f
1
5
.
3
%
an
d
th
is
f
ig
u
r
e
was
f
ar
h
ig
h
er
th
a
n
th
e
s
ec
o
n
d
g
r
o
u
p
,
wh
ich
r
ea
ch
e
d
o
n
l
y
2
6
.
2
%.
I
n
ar
ch
itectu
r
e
,
th
e
Alex
-
N
et
m
o
d
el
co
n
s
is
ts
o
f
f
iv
e
co
n
v
o
lu
tio
n
lay
e
r
s
f
o
llo
w
ed
b
y
th
r
ee
f
u
lly
co
n
n
ec
ted
la
y
er
s
an
d
th
e
en
d
is
th
e
So
f
tM
ax
f
u
n
ctio
n
.
I
n
th
e
m
ed
ical
f
ield
,
Alex
-
N
et
was
also
u
s
ed
f
o
r
s
o
m
e
r
esear
ch
.
Fo
r
ex
am
p
le,
Mo
h
i
u
d
Din
an
d
J
ay
an
th
y
[
2
2
]
p
u
b
lis
h
ed
a
s
tu
d
y
o
f
class
if
y
in
g
au
tis
m
u
s
in
g
E
lectr
o
e
n
ce
p
h
alo
g
r
ap
h
y
s
ig
n
als
in
2
0
2
2
.
Stu
d
ies
[
1
6
]
,
[
1
8
]
also
u
s
ed
Alex
-
Net.
He
et
a
l.
[
2
3
]
h
ad
a
s
tu
d
y
a
b
o
u
t
tr
ain
in
g
d
ee
p
n
etwo
r
k
s
to
s
o
lv
e
th
e
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
.
I
n
th
eir
p
ap
er
,
th
eir
id
ea
was
i
m
p
lem
en
ted
in
th
e
r
esid
u
al
b
lo
ck
m
o
d
el
o
f
R
es
-
Net
ar
ch
itectu
r
e.
W
ith
th
at
ar
ch
itectu
r
e,
th
e
au
th
o
r
s
wo
n
th
e
im
ag
e
n
et
lar
g
e
s
ca
le
v
is
u
al
r
ec
o
g
n
itio
n
ch
allen
g
e
2
0
1
5
a
n
d
r
ea
c
h
ed
a
class
if
icatio
n
er
r
o
r
o
f
3
.
5
7
%.
Sin
ce
th
en
,
th
is
ar
ch
itectu
r
e
h
as
b
ec
o
m
e
f
am
o
u
s
in
th
e
d
ee
p
-
lear
n
in
g
f
ield
an
d
h
as
h
ad
m
an
y
v
ar
ian
ts
with
d
if
f
er
en
t
d
ep
th
s
.
I
n
m
e
d
ica
l
r
esear
ch
,
R
es
-
Net
is
ap
p
li
ed
to
m
an
y
im
ag
e
class
if
icatio
n
p
r
o
b
lem
s
,
s
u
ch
a
s
[
1
0
]
,
[
1
8
]
.
Hu
an
g
et
a
l.
[
2
4
]
p
r
o
p
o
s
ed
Den
s
e
-
Net
with
th
e
id
ea
o
f
c
o
n
n
ec
tin
g
d
e
n
s
ely
b
etwe
en
lay
er
s
.
T
h
eir
id
ea
was
im
p
lem
en
ted
in
th
e
Den
s
e
b
lo
ck
m
o
d
el
an
d
was
r
ath
er
s
im
ilar
to
R
es
-
Net
in
th
e
co
n
tex
t
o
f
th
e
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
.
I
n
m
ed
ical
r
esear
ch
,
Den
s
e
-
N
e
t
is
ap
p
lied
to
im
ag
e
class
if
icatio
n
p
r
o
b
lem
s
,
s
u
ch
as c
lass
if
y
in
g
au
tis
m
[
1
3
]
,
[
1
4
]
.
3.
P
RO
P
O
SE
D
S
T
UDY
3
.
1
.
Da
t
a
s
et
prepa
ra
t
io
n
Ou
r
s
tu
d
y
is
d
esig
n
e
d
to
f
o
cu
s
o
n
th
e
Vietn
am
ese
co
n
tex
t.
On
e
im
p
o
r
tan
t
task
is
th
e
p
r
ep
ar
atio
n
o
f
a
Vietn
am
ese
im
ag
e
d
ataset
f
o
r
th
e
ex
p
er
im
en
t.
I
n
t
h
is
s
tu
d
y
,
all
f
ac
ial
im
ag
es
wer
e
ca
p
tu
r
ed
with
th
e
s
u
p
p
o
r
t
o
f
th
e
ca
m
e
r
a
o
n
th
e
s
m
ar
tp
h
o
n
e.
W
e
co
llected
d
ata
at
s
ev
er
al
k
in
d
er
g
ar
te
n
s
in
Ho
C
h
i
Min
h
C
ity
,
Vietn
am
.
Data
we
r
e
c
o
llected
f
r
o
m
s
o
m
e
ch
ild
r
en
with
au
tis
m
an
d
s
o
m
e
n
o
r
m
al
ch
ild
r
en
.
T
h
e
co
llectio
n
p
r
o
ce
s
s
was c
ar
r
ied
o
u
t w
ith
th
e
s
u
p
p
o
r
t
o
f
p
ar
en
ts
.
Acc
o
r
d
in
g
ly
,
p
ar
e
n
ts
will
d
ir
ec
tly
r
ec
o
r
d
im
ag
e
s
o
f
th
eir
ch
ild
r
en
'
s
f
ac
es
u
s
i
n
g
s
m
ar
tp
h
o
n
es
o
r
p
ar
en
ts
will
b
ab
y
s
it
th
e
ch
ild
r
en
wh
ile
s
o
m
eo
n
e
else
r
ec
o
r
d
s
th
e
im
ag
es.
C
o
llected
im
ag
e
s
will
b
e
s
elec
ted
to
g
et
q
u
alif
ied
im
ag
es.
Acc
o
r
d
in
g
ly
,
we
will
r
em
o
v
e
im
a
g
e
s
in
wh
ich
th
e
ch
ild
'
s
f
ac
e
is
b
lu
r
r
ed
,
p
ar
tially
o
b
s
cu
r
ed
d
u
e
to
t
h
e
ch
ild
'
s
f
ac
ial
p
o
s
e,
o
r
o
b
s
cu
r
ed
b
y
o
th
er
o
b
jects o
r
b
o
d
y
p
ar
ts
s
u
ch
as a
h
an
d
.
T
o
p
er
f
o
r
m
d
ata
lab
elin
g
,
we
cr
o
p
p
e
d
th
e
q
u
alif
ied
f
ac
e
im
ag
e
ar
ea
s
an
d
ar
r
an
g
ed
th
e
m
in
to
tw
o
g
r
o
u
p
s
co
r
r
esp
o
n
d
in
g
to
th
e
s
tatu
s
o
f
ch
ild
r
en
with
au
tis
m
o
r
n
o
r
m
al.
T
o
d
o
th
is
,
we
b
u
ilt
a
p
r
o
g
r
a
m
th
at
au
to
m
atica
lly
lo
ca
lizes
th
e
f
a
ce
in
th
e
im
ag
e,
an
d
b
ased
o
n
th
at
r
esu
lt,
we
p
er
f
o
r
m
ed
o
p
er
atio
n
s
o
n
th
e
p
r
o
g
r
a
m
in
ter
f
ac
e
to
b
e
ab
le
to
c
o
r
r
ec
t
t
h
e
f
ac
e
ar
ea
if
n
ec
ess
ar
y
.
Fin
ally
,
we
wo
u
ld
an
n
o
tate
th
e
s
tates
co
r
r
esp
o
n
d
in
g
t
o
th
e
two
g
r
o
u
p
s
an
d
th
e
p
r
o
g
r
am
will sh
o
w
th
e
f
in
al
r
esu
ltin
g
im
a
g
es.
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
Dete
ctin
g
a
u
tis
m
w
ith
V
ietn
a
mese
ch
ild
fa
cia
l ima
g
es
u
s
in
g
…
(
Tr
a
n
V
a
n
Th
a
n
h
)
4765
I
n
s
u
m
m
ar
y
,
we
h
a
v
e
p
r
ep
ar
ed
a
d
ata
s
et
o
f
8
9
2
im
ag
es
o
f
Vietn
am
ese
ch
ild
r
e
n
'
s
f
ac
es.
Am
o
n
g
th
em
,
th
er
e
ar
e
4
4
4
p
h
o
to
s
o
f
ch
ild
r
en
with
au
tis
m
a
n
d
4
4
8
p
h
o
to
s
o
f
n
o
r
m
al
c
h
ild
r
en
.
All
o
f
th
ese
im
a
g
es
wer
e
u
s
ed
in
th
e
e
x
p
er
im
e
n
tal
s
tep
o
f
th
is
s
tu
d
y
.
B
esid
es
th
e
d
ata
o
f
Vietn
am
e
s
e
ch
ild
r
en
,
we
u
s
e
an
in
ter
n
atio
n
al
d
ataset
p
u
b
lis
h
ed
o
n
th
e
Kag
g
le
p
latf
o
r
m
[
2
5
]
.
T
h
is
d
ataset
in
clu
d
es
2
,
9
3
6
im
ag
es
o
f
ch
ild
r
en
'
s
f
ac
es
th
at
ar
e
also
d
iv
id
ed
in
to
two
g
r
o
u
p
s
ac
co
r
d
in
g
to
au
tis
m
cr
iter
ia.
I
n
th
e
d
ataset,
th
er
e
ar
e
1
,
4
6
8
im
ag
es
m
ar
k
ed
as
b
ei
n
g
o
f
ch
ild
r
en
with
au
tis
m
an
d
1
,
4
6
8
i
m
ag
es o
f
t
h
e
o
p
p
o
s
ite
ca
s
e.
T
o
ex
p
er
im
en
t
i
n
th
is
s
tu
d
y
,
we
wo
u
ld
p
r
ep
a
r
e
tr
ain
in
g
,
v
alid
atio
n
,
an
d
test
d
atasets
b
a
s
ed
o
n
t
h
e
Vietn
am
ese
d
ataset
an
d
th
e
in
ter
n
atio
n
al
d
ataset.
So
,
b
o
th
th
e
Vietn
am
ese
d
ataset
an
d
th
e
i
n
ter
n
atio
n
al
d
ataset
wer
e
s
p
lit.
T
h
e
in
ter
n
atio
n
al
d
ataset
was
alr
ea
d
y
s
p
li
t.
I
n
d
etail,
f
r
o
m
t
h
e
in
ter
n
atio
n
al
d
ataset,
th
e
test
s
et
h
as
3
0
0
s
am
p
les,
th
e
v
alid
atio
n
s
et
h
as
1
0
0
s
am
p
les,
an
d
th
e
tr
ain
in
g
s
et
h
as
2
,
5
3
6
s
am
p
les.
Fo
r
th
e
Vietn
am
ese
d
ataset,
we
s
et
u
p
th
e
test
s
et
with
9
0
s
am
p
l
es,
th
e
v
alid
atio
n
s
et
with
8
9
s
am
p
les,
an
d
th
e
tr
ain
in
g
s
et
with
7
1
3
s
am
p
les.
3
.
2
.
P
r
o
po
s
ed
s
t
ra
t
eg
ies re
la
t
ing
t
o
a
pp
ly
inte
rna
t
io
na
l d
a
t
a
in g
ener
a
lizing
t
o
Viet
na
m
ese
da
t
a
T
o
a
p
p
ly
i
n
t
er
n
a
tio
n
al
d
at
a
to
th
e
p
r
o
b
le
m
o
f
d
iag
n
o
s
in
g
au
ti
s
m
in
V
iet
n
am
es
e
ch
il
d
r
en
,
w
e
an
al
y
ze
th
e
u
s
e
o
f
th
e
tr
an
s
f
er
lea
r
n
in
g
ap
p
r
o
ac
h
.
T
r
an
s
f
er
l
ea
r
n
in
g
i
s
a
p
o
p
u
lar
a
p
p
r
o
a
ch
to
p
r
o
b
lem
s
u
s
in
g
d
ee
p
le
ar
n
in
g
m
o
d
el
s
.
I
n
wh
i
ch
,
t
h
e
ap
p
l
ic
at
io
n
o
f
tr
an
s
f
er
le
ar
n
in
g
w
il
l
b
e
ef
f
ec
t
iv
e
if
th
e
d
o
m
a
in
o
f
th
e
d
ata
u
s
ed
f
o
r
p
r
e
tr
a
in
in
g
an
d
th
e
d
o
m
ain
o
f
th
e
d
a
ta
in
th
e
t
ar
g
e
t
p
r
o
b
l
em
ar
e
s
im
il
ar
an
d
t
h
i
s
ap
p
l
ic
at
io
n
wi
ll
b
e
in
ef
f
ec
ti
v
e
if
th
e
two
d
ata
d
o
m
a
in
s
ar
e
s
u
f
f
ic
ie
n
t
ly
d
if
f
e
r
en
t.
Fo
r
th
e
p
r
o
b
l
em
o
f
d
iag
n
o
s
in
g
a
u
t
i
s
m
f
r
o
m
f
ac
ia
l
i
m
ag
e
s
o
f
Vi
etn
a
m
es
e
ch
ild
r
en
,
we
h
av
e
s
o
m
e
o
b
s
e
r
v
a
tio
n
s
a
s
f
o
ll
o
w
s
.
On
t
h
e
o
n
e
h
an
d
,
au
t
i
s
m
is
a
s
p
ec
if
ic
p
r
o
b
lem
wi
th
d
i
s
t
in
c
t
s
em
an
t
ic
s
.
T
h
at
m
ak
e
s
it
p
o
s
s
i
b
l
e
to
u
s
e
a
m
o
d
e
l
tr
ain
e
d
f
r
o
m
s
cr
a
tch
.
On
th
e
o
th
er
h
an
d
,
th
e
d
at
a
u
s
e
d
i
s
f
ac
ia
l
im
ag
e
d
a
ta.
T
h
i
s
i
s
a
ty
p
e
o
f
d
a
ta
th
at
h
a
s
b
ee
n
u
s
ed
in
m
an
y
o
th
er
p
r
o
b
lem
s
s
u
ch
as
f
a
c
ia
l
r
e
co
g
n
it
io
n
,
em
o
tio
n
r
e
c
o
g
n
it
io
n
,
an
d
g
en
d
er
r
ec
o
g
n
it
io
n
.
Fr
o
m
th
i
s
p
er
s
p
ec
t
iv
e,
th
e
ap
p
l
ic
at
io
n
o
f
th
e
tr
an
s
f
er
l
ea
r
n
in
g
a
p
p
r
o
ac
h
i
s
f
ea
s
ib
l
e.
Fro
m
o
u
r
p
er
s
p
ec
ti
v
e,
it
m
ak
e
s
s
en
s
e
to
ap
p
ly
a
t
r
an
s
f
er
le
ar
n
in
g
ap
p
r
o
ac
h
.
T
o
f
u
r
t
h
er
c
l
ar
if
y
th
i
s
,
in
o
u
r
ex
p
er
i
m
en
t
s
,
we
w
i
ll
t
r
a
in
th
e
m
o
d
el
f
r
o
m
s
cr
a
tch
an
d
tr
a
in
th
e
m
o
d
el
w
i
th
p
r
e
-
tr
a
in
ed
w
eig
h
t
s
.
T
h
e
ch
o
i
ce
u
s
ed
i
s
th
e
p
r
e
-
tr
ain
ed
m
o
d
e
l
we
ig
h
t
s
p
r
o
v
id
e
d
b
y
th
e
Py
T
o
r
ch
p
la
tf
o
r
m
[
2
6
]
.
T
h
e
s
e
ar
e
m
o
d
el
s
th
a
t
h
av
e
b
ee
n
p
r
e
-
tr
a
in
ed
o
n
t
h
e
p
o
p
u
lar
I
m
ag
e
Ne
t d
at
as
et
.
Dig
g
in
g
d
ee
p
er
in
to
t
h
e
p
r
o
b
lem
,
we
f
o
u
n
d
th
at
with
au
ti
s
m
d
iag
n
o
s
is
b
ased
o
n
c
h
ild
r
en
'
s
f
ac
ial
im
ag
es,
th
e
s
ig
n
s
o
f
au
tis
m
w
ill
b
e
ex
p
r
ess
ed
th
r
o
u
g
h
ch
ild
r
en
'
s
f
ac
ial
ex
p
r
ess
io
n
s
.
T
h
er
ef
o
r
e,
it
ca
n
b
e
s
aid
th
at
th
e
d
ata
d
o
m
ain
o
f
th
is
p
r
o
b
lem
will
h
av
e
a
ce
r
tain
s
i
m
ilar
ity
with
th
e
d
ata
d
o
m
ain
o
f
th
e
p
r
o
b
lem
o
f
r
ec
o
g
n
izin
g
ch
ild
r
en
'
s
f
ac
ial
ex
p
r
ess
io
n
s
.
On
th
is
b
asis
,
we
f
u
r
th
e
r
d
esig
n
e
d
th
e
ap
p
licatio
n
o
f
tr
an
s
f
e
r
lear
n
in
g
u
s
in
g
a
p
r
e
-
tr
ain
e
d
m
o
d
el
o
n
th
e
c
h
ild
r
en
'
s
f
ac
ial
ex
p
r
ess
io
n
d
ataset.
Ou
r
e
x
p
e
r
im
en
tal
ch
o
ice
is
to
u
s
e
p
r
e
-
tr
ain
ed
m
o
d
els
o
n
th
e
f
ac
ial
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
(
FER)
ch
ild
r
en
'
s
f
ac
ial
ex
p
r
ess
io
n
d
ataset
[
2
7
]
.
W
e
h
y
p
o
th
esize
th
at
th
e
m
o
d
e
l
p
r
e
-
tr
ain
e
d
o
n
th
e
FER
ch
ild
r
en
'
s
f
ac
ial
ex
p
r
ess
io
n
d
ataset
will
h
av
e
a
h
ig
h
er
f
it th
an
th
e
m
o
d
el
p
r
e
-
tr
ain
ed
o
n
th
e
I
m
ag
eNe
t d
ataset.
T
h
u
s
,
i
n
d
e
s
i
g
n
i
n
g
s
t
r
at
e
g
i
es
to
t
e
s
t
t
h
e
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
a
p
p
l
i
c
a
t
i
o
n
,
w
e
u
s
e
t
h
r
e
e
d
i
f
f
e
r
en
t
o
p
t
i
o
n
s
.
S
p
e
c
i
f
i
ca
l
l
y
,
t
h
e
o
p
t
i
o
n
w
it
h
m
o
d
e
l
s
t
r
a
i
n
e
d
f
r
o
m
s
c
r
a
t
c
h
,
t
h
e
o
p
t
i
o
n
w
it
h
m
o
d
e
ls
p
r
e
-
t
r
ai
n
ed
o
n
I
m
a
g
e
N
e
t
a
n
d
a
v
a
i
l
a
b
l
e
o
n
t
h
e
P
y
T
o
r
c
h
p
la
t
f
o
r
m
,
a
n
d
t
h
e
f
i
n
al
o
p
ti
o
n
is
m
o
d
e
l
s
p
r
e
-
t
r
a
i
n
e
d
o
n
t
h
e
F
E
R
c
h
i
l
d
f
a
c
ia
l
e
x
p
r
e
s
s
i
o
n
d
a
t
a
s
et
.
T
h
e
t
r
a
i
n
e
d
m
o
d
e
l
s
wil
l
b
e
e
v
a
l
u
a
t
e
d
o
n
b
o
t
h
t
h
e
i
n
te
r
n
a
t
i
o
n
a
l
d
a
t
as
e
t
a
n
d
t
h
e
V
ie
tn
a
m
e
s
e
d
a
t
as
e
t
f
o
r
t
h
e
a
u
t
is
m
d
i
a
g
n
o
s
i
s
f
r
o
m
c
h
il
d
f
a
c
i
a
l
i
m
a
g
es
.
Ac
c
o
r
d
i
n
g
t
o
o
u
r
h
y
p
o
t
h
e
s
i
s
,
t
h
e
o
r
d
e
r
o
f
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
c
a
s
es
w
i
l
l
b
e
s
i
m
i
l
a
r
o
n
b
o
t
h
t
e
s
t
s
et
s
w
it
h
t
h
e
b
e
s
t
r
e
s
u
l
ts
b
e
lo
n
g
i
n
g
t
o
t
h
e
m
o
d
e
l
s
p
r
e
-
t
r
a
i
n
e
d
o
n
t
h
e
c
h
i
l
d
f
a
c
i
a
l
e
x
p
r
e
s
s
i
o
n
d
a
ta
s
et
a
n
d
t
h
e
w
o
r
s
t
r
es
u
l
ts
b
e
l
o
n
g
i
n
g
t
o
t
h
e
m
o
d
e
l
s
t
r
ai
n
e
d
f
r
o
m
s
c
r
a
tc
h
.
An
o
th
er
o
p
tio
n
we
h
av
e
s
et
o
u
t
to
ap
p
ly
in
ter
n
atio
n
al
d
ata
to
th
e
p
r
o
b
lem
o
f
d
iag
n
o
s
in
g
au
tis
m
in
Vietn
am
ese
ch
ild
r
en
is
to
co
m
b
in
e
in
ter
n
atio
n
al
d
ata
an
d
Vietn
am
ese
d
ata
in
to
th
e
tr
ain
in
g
s
et
as d
escr
ib
ed
in
T
ab
le
1
.
Usi
n
g
th
is
o
p
tio
n
wi
ll
g
iv
e
u
s
a
clea
r
ass
es
s
m
en
t
o
f
th
e
d
if
f
e
r
en
ce
b
etwe
en
th
e
two
d
ata
d
o
m
ain
s
,
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
i
m
ag
es
an
d
in
ter
n
atio
n
al
c
h
ild
r
en
'
s
f
ac
ial
im
ag
es,
in
th
e
p
r
o
b
lem
o
f
d
ia
g
n
o
s
in
g
au
tis
m
f
r
o
m
ch
ild
r
en
'
s
f
ac
ial
i
m
ag
es.
Acc
o
r
d
in
g
ly
,
we
d
esig
n
s
tr
ateg
ies
with
tr
ain
in
g
d
ata
o
p
tio
n
s
in
clu
d
in
g
th
e
f
ir
s
t
o
p
tio
n
o
f
tr
ain
in
g
o
n
l
y
with
in
ter
n
atio
n
al
d
ata
an
d
t
h
e
s
ec
o
n
d
o
p
tio
n
o
f
tr
ain
in
g
with
d
ata
co
m
b
in
in
g
in
ter
n
atio
n
al
d
ata
with
Vietn
a
m
ese
d
ata.
Similar
ly
,
th
e
tr
ai
n
ed
m
o
d
els
will
also
b
e
ev
al
u
ated
s
ep
ar
ately
o
n
th
e
in
ter
n
atio
n
al
d
ata
s
et
an
d
t
h
e
Vietn
am
ese
d
ata
s
et.
Acc
o
r
d
in
g
to
o
u
r
h
y
p
o
th
esis
,
th
e
e
v
alu
atio
n
r
esu
lts
o
n
th
e
in
ter
n
atio
n
al
d
ata
s
et
will
n
o
t
c
h
an
g
e
m
u
ch
d
u
e
to
t
h
e
p
ar
ticip
atio
n
o
f
in
ter
n
atio
n
al
d
ata
in
th
e
tr
ain
in
g
s
et.
Ho
wev
er
,
we
b
eliev
e
th
at
th
e
ev
alu
atio
n
r
esu
lts
o
n
th
e
Vietn
am
ese
d
ataset
will
h
av
e
a
b
ig
d
if
f
er
e
n
ce
.
I
n
p
ar
ticu
lar
,
th
e
r
esu
lts
o
f
m
o
d
els
tr
ain
ed
o
n
th
e
c
o
m
b
in
e
d
d
ataset
o
f
in
ter
n
atio
n
al
d
ata
a
n
d
Vietn
am
ese
d
ata
will
b
e
h
ig
h
er
th
an
th
e
r
esu
lts
o
f
m
o
d
els
tr
ain
ed
o
n
ly
o
n
in
ter
n
atio
n
al
d
ata
d
u
e
to
th
e
d
if
f
er
en
ce
in
d
ata
d
o
m
ain
.
T
h
e
s
tr
ateg
ies ar
e
d
es
cr
ib
ed
in
d
etail
as f
o
llo
ws:
I
n
Fig
u
r
e
1
,
Stra
teg
y
1
A
tr
ain
s
d
ee
p
lear
n
in
g
m
o
d
els
f
r
o
m
s
cr
atch
o
n
in
ter
n
atio
n
al
d
ata,
a
n
d
s
tr
ateg
y
1
B
also
tr
ain
s
d
ee
p
lear
n
in
g
m
o
d
els
f
r
o
m
s
cr
atch
b
u
t
o
n
c
o
m
b
in
ed
d
ata.
Acc
o
r
d
in
g
l
y
,
th
e
m
o
d
el
weig
h
ts
will
b
e
r
an
d
o
m
ly
in
itialized
.
T
r
ai
n
in
g
f
r
o
m
s
cr
atch
ca
n
b
e
co
n
s
id
er
ed
b
ec
a
u
s
e
au
tis
m
is
a
r
ath
er
s
p
ec
ialized
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
6
2
-
4
7
7
3
4766
co
n
ten
t.
H
o
wev
er
,
ac
co
r
d
in
g
to
o
u
r
h
y
p
o
th
esis
,
th
e
t
r
ain
in
g
r
esu
lts
f
r
o
m
s
cr
atch
will
li
k
ely
b
e
wo
r
s
e
th
an
u
s
in
g
tr
an
s
f
er
lea
r
n
in
g
with
in
th
e
ex
p
er
im
en
tal
s
co
p
e
s
et.
T
h
e
ev
alu
ati
o
n
r
esu
lts
o
n
in
ter
n
atio
n
al
a
n
d
Vietn
am
ese
test
d
ata
s
ets
o
f
t
h
e
two
s
tr
ateg
ies
ar
e
ex
p
ec
te
d
to
also
r
ef
lect
th
e
d
if
f
e
r
en
ce
in
d
ata
d
o
m
ain
s
o
n
th
e
two
s
ets.
Nex
t
in
Fig
u
r
e
2
,
s
tr
ateg
y
2
A
tr
ain
s
d
ee
p
lear
n
in
g
m
o
d
el
s
o
n
in
ter
n
atio
n
al
d
ata
with
p
ar
am
eter
s
lear
n
ed
b
y
tr
an
s
f
er
r
i
n
g
f
r
o
m
p
r
etr
ain
e
d
m
o
d
els
with
I
m
a
g
eNe
t
an
d
s
tr
ateg
y
2
B
is
s
im
ilar
b
u
t
p
er
f
o
r
m
s
tr
ain
in
g
o
n
co
m
b
in
ed
d
ata.
Acc
o
r
d
in
g
ly
,
th
e
m
o
d
el
weig
h
ts
will
b
e
tak
en
f
r
o
m
m
o
d
e
ls
p
r
o
v
id
ed
b
y
th
e
Py
T
o
r
ch
p
latf
o
r
m
an
d
tr
ain
e
d
o
n
th
e
I
m
a
g
eNe
t
d
ataset.
T
r
a
n
s
f
er
lear
n
in
g
is
p
e
r
f
o
r
m
ed
o
n
t
h
e
ass
u
m
p
tio
n
th
at
th
e
au
tis
m
d
ata
u
s
ed
h
er
e
is
f
ac
ial
im
ag
e
d
ata
-
a
f
air
ly
co
m
m
o
n
ty
p
e
o
f
im
a
g
e
d
ata.
Acc
o
r
d
in
g
to
o
u
r
h
y
p
o
th
esis
,
th
e
ev
alu
atio
n
r
es
u
lts
o
f
th
e
two
s
tr
ateg
ies
will
b
e
b
etter
th
an
th
o
s
e
o
f
s
tr
ate
g
ies
1
A
an
d
1
B
in
ca
s
e
o
f
th
e
s
am
e
tr
ain
in
g
s
et.
Similar
ly
,
th
e
ev
alu
atio
n
r
esu
l
ts
o
n
th
e
in
ter
n
atio
n
al
an
d
Vie
tn
am
ese
test
s
e
ts
o
f
th
ese
two
s
tr
ateg
ies ar
e
also
ex
p
ec
ted
to
r
ef
lect
th
e
d
i
f
f
er
en
c
e
in
d
ata
d
o
m
ain
s
o
n
t
h
e
two
s
ets.
Fin
ally
in
Fig
u
r
e
3
,
s
tr
ateg
y
3
A
tr
ain
s
d
ee
p
lear
n
i
n
g
m
o
d
els
o
n
th
e
in
ter
n
atio
n
al
d
ataset
with
p
ar
am
eter
s
lear
n
e
d
b
y
tr
a
n
s
f
er
r
in
g
f
r
o
m
m
o
d
els p
r
e
-
tr
ain
e
d
o
n
th
e
FER
d
ataset,
an
d
s
tr
ate
g
y
3
B
is
s
im
ilar
b
u
t
tr
ain
s
o
n
th
e
co
m
b
in
e
d
d
ataset.
Acc
o
r
d
in
g
ly
,
we
will
tr
ain
th
e
p
r
e
-
m
o
d
els
o
n
th
e
FER
d
ataset
with
f
ac
ial
ex
p
r
ess
io
n
lab
els.
T
r
an
s
f
er
lear
n
in
g
h
er
e
is
p
er
f
o
r
m
ed
o
n
th
e
ass
u
m
p
tio
n
th
at
au
tis
m
f
ea
tu
r
es
o
n
ch
ild
r
en
'
s
f
ac
ial
im
ag
es
will
also
b
e
ex
p
r
ess
ed
th
r
o
u
g
h
f
ac
ial
ex
p
r
ess
io
n
s
.
Acc
o
r
d
in
g
to
o
u
r
h
y
p
o
th
esis
,
th
e
ev
alu
atio
n
r
esu
lts
o
f
th
ese
two
s
tr
ateg
ie
s
will
b
e
b
est
wh
en
co
n
s
id
er
ed
in
th
e
ca
s
e
o
f
th
e
s
am
e
tr
ain
in
g
s
et.
Similar
to
th
e
p
r
ev
io
u
s
ca
s
es,
th
e
ev
alu
atio
n
r
esu
lts
o
n
th
e
in
ter
n
atio
n
al
an
d
Vietn
am
ese
test
s
e
ts
o
f
th
es
e
two
s
tr
ateg
ies
ar
e
also
ex
p
ec
ted
to
r
ef
lect
th
e
d
if
f
er
en
ce
in
t
h
e
d
ata
d
o
m
ain
s
o
n
th
e
two
s
ets.
T
ab
le
1
.
I
n
ter
n
atio
n
al
d
ata
an
d
co
m
b
in
e
d
d
ata
in
tr
ai
n
in
g
p
h
a
s
e
C
a
se
I
n
t
e
r
n
a
t
i
o
n
a
l
d
a
t
a
V
i
e
t
n
a
m
e
se
d
a
t
a
I
n
t
e
r
n
a
t
i
o
n
a
l
t
r
a
i
n
i
n
g
se
t
2
5
3
6
0
I
n
t
e
r
n
a
t
i
o
n
a
l
v
a
l
i
d
a
t
i
o
n
s
e
t
1
0
0
0
C
o
m
b
i
n
e
d
t
r
a
i
n
i
n
g
set
2
5
3
6
7
1
3
C
o
m
b
i
n
e
d
v
a
l
i
d
a
t
i
o
n
s
e
t
1
0
0
89
Fig
u
r
e
1
.
Stra
teg
y
1
A
(
lef
t)
an
d
s
tr
ateg
y
1
B
(
r
ig
h
t)
Fig
u
r
e
2
.
Stra
teg
y
2
A
(
lef
t)
an
d
s
tr
ateg
y
2
B
(
r
ig
h
t)
Fig
u
r
e
3
.
Stra
teg
y
3
A
(
lef
t)
an
d
s
tr
ateg
y
3
B
(
r
ig
h
t)
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
Dete
ctin
g
a
u
tis
m
w
ith
V
ietn
a
mese
ch
ild
fa
cia
l ima
g
es
u
s
in
g
…
(
Tr
a
n
V
a
n
Th
a
n
h
)
4767
4.
RE
SU
L
T
S AN
D
E
VA
L
UA
T
I
O
N
4
.
1
.
P
er
f
o
rma
nce
m
ea
s
ures
I
n
th
is
s
tu
d
y
,
we
u
s
ed
p
h
o
t
o
s
o
f
ch
ild
r
en
'
s
f
ac
es
to
ass
es
s
au
tis
m
s
tatu
s
.
T
h
eo
r
etica
lly
,
th
is
is
a
2
-
class
class
if
icat
io
n
p
r
o
b
le
m
.
W
ith
a
p
h
o
t
o
,
th
e
p
r
o
g
r
am
wo
u
ld
c
o
n
clu
d
e
"No
r
m
a
l"
o
r
"Au
tis
m
".
Fo
r
ev
alu
atio
n
,
we
u
s
ed
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
ac
cu
r
ac
y
,
a
n
d
AUC.
AU
C
is
th
e
ab
b
r
ev
iatio
n
f
o
r
th
e
ar
ea
u
n
d
er
th
e
r
ec
eiv
er
o
p
er
atin
g
ch
ar
a
cter
is
tic
cu
r
v
e,
wh
ich
is
a
g
r
ap
h
th
at
s
h
o
ws
th
e
r
elatio
n
s
h
ip
b
etwe
en
T
r
u
e
Po
s
itiv
e
R
ate
an
d
Fal
s
e
Po
s
it
iv
e
R
ate
o
v
er
a
s
et
th
r
esh
o
ld
.
AUC
is
a
m
ea
s
u
r
e
o
f
th
e
o
v
er
all
q
u
ality
o
f
a
b
in
ar
y
class
if
ier
.
Fo
r
t
h
e
o
th
er
s
,
s
en
s
itiv
ity
an
d
s
p
ec
if
icity
ar
e
two
c
o
m
m
o
n
ly
u
s
ed
m
ea
s
u
r
es
in
m
ed
ical
r
esear
ch
.
T
h
ese
m
ea
s
u
r
es
al
o
n
g
with
t
h
e
ac
cu
r
ac
y
ar
e
c
alcu
lated
b
y
c
o
m
b
in
i
n
g
p
r
ed
icted
an
d
r
e
f
er
en
ce
v
alu
es
th
r
o
u
g
h
th
e
v
alu
es:
tr
u
e
p
o
s
itiv
e
(
T
P),
tr
u
e
n
eg
ativ
e
(
T
N)
,
f
alse
p
o
s
itiv
e
(
FP
)
,
an
d
f
alse
n
eg
ativ
e
(
FN)
.
T
h
e
ca
lcu
latio
n
is
p
er
f
o
r
m
e
d
a
s
f
o
llo
ws:
=
+
(
1
)
=
+
(
2
)
=
+
+
+
+
(
3
)
4
.
2
.
E
x
perim
ent
s
a
nd
re
s
ults
T
h
is
s
ec
tio
n
wo
u
ld
clar
if
y
o
u
r
h
y
p
o
th
eses
r
elatin
g
to
ap
p
ly
in
ter
n
atio
n
al
d
ata
to
th
e
p
r
o
b
lem
o
f
d
iag
n
o
s
in
g
au
tis
m
in
Vietn
a
m
ese
ch
ild
r
en
b
y
d
is
cu
s
s
in
g
th
e
r
esu
lts
o
b
tain
ed
f
r
o
m
th
e
ex
p
er
im
e
n
ts
.
I
n
th
e
tr
ain
in
g
p
h
ase,
th
e
m
o
d
els
ar
e
lear
n
ed
u
s
in
g
th
e
tr
ain
in
g
an
d
v
alid
atio
n
s
ets
as
d
escr
ib
ed
ab
o
v
e.
T
h
e
lear
n
in
g
p
r
o
ce
s
s
will
b
e
s
to
p
p
e
d
b
ased
o
n
th
e
ev
alu
atio
n
o
f
th
e
m
ea
s
u
r
ed
s
co
r
e
f
r
o
m
th
e
v
alid
atio
n
s
et.
Mo
r
e
s
p
ec
if
ically
,
th
e
ac
cu
r
ac
y
s
co
r
e
will
b
e
u
s
ed
as
an
in
d
icato
r
.
Af
ter
b
ein
g
tr
ain
ed
,
th
e
m
o
d
els
will
b
e
p
u
t
in
to
th
e
ev
alu
atio
n
s
tep
with
test
s
ets.
T
h
r
o
u
g
h
th
ese
m
etr
ics,
we
will
b
e
ab
le
to
an
aly
z
e
an
d
d
is
cu
s
s
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
d
esig
n
ed
s
tr
ateg
ies
an
d
h
ig
h
lig
h
t
im
p
o
r
tan
t
ch
ar
ac
ter
is
tics
.
T
o
ex
p
er
im
en
t,
we
u
s
e
th
e
Go
o
g
le
C
o
lab
d
ee
p
lear
n
in
g
s
er
v
er
s
u
p
p
o
r
ted
b
y
th
e
Nv
id
ia
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
co
m
p
u
tatio
n
p
o
wer
.
T
h
e
test
p
r
o
g
r
am
is
b
u
ilt
o
n
th
e
Py
th
o
n
p
r
o
g
r
am
m
in
g
lan
g
u
a
g
e
an
d
is
s
u
p
p
o
r
ted
b
y
th
e
Py
T
o
r
ch
d
ee
p
lear
n
in
g
lib
r
ar
y
.
Fo
u
r
m
o
d
el
s
R
esNet3
4
,
R
es
Net5
0
,
Alex
-
Net,
an
d
Den
s
eNe
t1
2
1
wer
e
ch
o
s
en
f
o
r
ea
c
h
s
tr
ateg
y
.
Af
ter
th
at,
th
e
tr
ain
e
d
m
o
d
els
o
f
s
tr
ateg
ies
wer
e
test
ed
o
n
th
e
test
s
et
s
o
f
th
e
Vie
tn
am
ese
d
ataset
an
d
th
e
in
ter
n
atio
n
al
d
ataset.
T
o
h
av
e
a
f
air
a
n
d
o
v
er
all
as
s
ess
m
en
t
o
f
th
e
q
u
ality
o
f
th
e
s
tr
ateg
ies,
we
ev
alu
ate
th
e
av
er
ag
e
o
f
ea
ch
m
ea
s
u
r
e
f
r
o
m
th
e
f
o
u
r
m
o
d
els.
Fro
m
Fig
u
r
e
4
,
it
is
c
lear
th
at
th
e
av
er
ag
e
s
co
r
es
f
o
r
all
f
o
u
r
m
ea
s
u
r
es,
in
clu
d
in
g
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
an
d
AUC,
ar
e
i
n
th
e
s
am
e
o
r
d
er
.
Sp
ec
if
ically
,
s
tr
ateg
y
1
A
wit
h
tr
ain
in
g
d
ee
p
lear
n
i
n
g
m
o
d
els
f
r
o
m
s
cr
atch
h
as
th
e
wo
r
s
t
av
er
ag
e
s
co
r
es,
an
d
s
tr
ateg
y
3
A
with
tr
an
s
f
er
lear
n
in
g
f
r
o
m
p
r
e
-
tr
ai
n
ed
weig
h
ts
o
n
th
e
FER
f
ac
ial
ex
p
r
ess
io
n
d
ataset
ac
h
iev
es
th
e
b
est
r
esu
lts
.
Sp
ec
if
ically
,
th
e
s
tr
ateg
y
3
A
wh
en
e
v
alu
ate
d
with
th
e
in
ter
n
atio
n
al
test
d
ataset
ac
h
iev
ed
an
av
er
a
g
e
ac
c
u
r
ac
y
o
f
0
.
8
7
0
8
3
3
,
an
av
e
r
ag
e
s
en
s
itiv
ity
o
f
0
.
8
5
6
6
6
7
,
an
av
e
r
ag
e
s
p
ec
if
icity
o
f
0
.
8
8
5
a
n
d
a
n
a
v
er
ag
e
AU
C
o
f
0
.
9
4
3
1
7
8
.
T
h
e
r
esu
lts
o
f
s
u
ch
s
tr
ateg
ies
also
ac
cu
r
ately
r
ef
lect
o
u
r
h
y
p
o
t
h
e
s
is
ab
o
u
t
in
itializin
g
th
e
weig
h
ts
o
f
d
ee
p
lear
n
in
g
m
o
d
els
b
ef
o
r
e
tr
ain
i
n
g
.
B
esid
es,
s
tr
ateg
y
3
A
also
o
u
tp
e
r
f
o
r
m
s
with
an
a
v
er
ag
e
tr
u
e
p
o
s
itiv
e
o
f
1
2
8
.
5
,
an
av
er
ag
e
tr
u
e
n
eg
ativ
e
o
f
1
3
2
.
7
5
,
an
a
v
er
ag
e
f
alse p
o
s
itiv
e
o
f
1
7
.
2
5
,
an
d
an
av
er
ag
e
f
alse n
eg
ativ
e
o
f
2
1
.
5
.
Fig
u
r
e
5
s
h
o
ws
th
e
av
er
a
g
e
s
c
o
r
es
o
n
t
h
e
test
s
et
o
f
th
e
in
ter
n
atio
n
al
d
ata
s
et
with
s
tr
ateg
ie
s
tr
ain
in
g
o
n
th
e
co
m
b
in
ed
d
ata.
W
h
en
l
o
o
k
in
g
at
th
e
ev
alu
atio
n
r
esu
lt
s
o
f
th
ese
s
tr
ateg
ies,
we
also
s
ee
a
clea
r
s
im
ilar
ity
to
th
e
r
esu
lts
in
Fig
u
r
e
4
.
T
r
ain
in
g
th
e
d
ee
p
lear
n
i
n
g
m
o
d
els
f
r
o
m
s
cr
atch
g
iv
es
th
e
wo
r
s
t
r
esu
lts
wh
ile
tr
ain
in
g
with
tr
a
n
s
f
er
lear
n
in
g
f
r
o
m
th
e
tr
ai
n
in
g
weig
h
ts
with
FER
g
iv
es
th
e
b
est
r
esu
lts
.
Sp
ec
if
ically
in
th
is
ca
s
e,
s
tr
ateg
y
3
B
ac
h
ie
v
ed
a
n
av
er
ag
e
ac
cu
r
ac
y
o
f
0
.
8
7
0
8
3
3
,
an
av
er
a
g
e
s
en
s
itiv
ity
o
f
0
.
8
6
6
6
6
7
,
an
a
v
er
ag
e
s
p
ec
if
icity
o
f
0
.
8
7
5
a
n
d
an
av
er
ag
e
AUC
o
f
0
.
9
3
9
2
4
4
.
Sim
ilar
ly
,
s
tr
ateg
y
3
B
also
s
h
o
ws
ef
f
ec
tiv
en
ess
with
av
er
ag
e
tr
u
e
p
o
s
itiv
e
an
d
av
e
r
ag
e
tr
u
e
n
eg
ativ
e
b
ei
n
g
h
ig
h
er
th
an
th
e
o
th
er
s
tr
ateg
ies
wh
ile
av
er
ag
e
Fals
e
Po
s
itiv
e
an
d
av
er
ag
e
f
alse
n
eg
ativ
e
b
ein
g
th
e
lo
west
am
o
n
g
th
e
s
tr
ateg
ies.
Sp
ec
if
ically
i
n
th
is
ca
s
e,
s
tr
ateg
y
3
B
ac
h
iev
ed
a
n
a
v
er
ag
e
tr
u
e
p
o
s
itiv
e
o
f
1
3
0
,
an
a
v
er
ag
e
tr
u
e
n
eg
ativ
e
o
f
1
3
1
.
2
5
,
an
av
er
a
g
e
f
alse
p
o
s
itiv
e
o
f
1
8
.
7
5
,
an
d
an
av
er
a
g
e
f
alse
n
eg
ativ
e
o
f
2
0
.
T
h
e
r
ef
o
r
e,
t
h
e
h
y
p
o
th
esis
o
f
th
e
f
ea
s
ib
il
ity
o
f
u
s
in
g
f
ac
ial
ex
p
r
ess
io
n
d
ata
was
clea
r
ly
d
em
o
n
s
tr
ated
th
r
o
u
g
h
th
e
ev
alu
atio
n
r
esu
lts
o
f
s
tr
ateg
ies
3
A
an
d
3
B
o
n
t
h
e
in
ter
n
atio
n
al
test
s
et.
An
o
th
er
p
o
in
t
to
n
o
te
is
t
h
at
a
lth
o
u
g
h
th
e
s
tr
ateg
ies
u
s
in
g
p
r
e
-
tr
ain
ed
weig
h
ts
with
FER
h
av
e
b
etter
r
esu
lts
th
an
th
e
s
tr
ateg
ies
u
s
in
g
p
r
e
-
tr
ain
e
d
weig
h
ts
with
I
m
ag
eNe
t
in
ea
ch
r
esp
ec
tiv
e
ca
s
e,
th
e
d
if
f
er
en
ce
in
r
esu
lts
is
n
o
t
lar
g
e
.
T
h
is
also
r
ef
lects
th
at
b
o
th
o
p
tio
n
s
a
r
e
f
ea
s
ib
le.
W
e
o
b
s
er
v
e
m
o
r
e
s
p
ec
if
ically
ab
o
u
t
th
e
d
if
f
er
en
ce
in
r
esu
lts
in
T
ab
le
2
.
T
ab
le
2
s
h
o
ws
th
e
d
if
f
er
en
ce
in
th
e
ev
alu
atio
n
d
ata
o
n
t
h
e
in
ter
n
atio
n
al
test
s
et
b
etwe
en
th
e
s
tr
ateg
ies
u
s
in
g
p
r
e
-
tr
ain
ed
weig
h
ts
o
n
t
h
e
f
ac
ial
ex
p
r
ess
io
n
d
ataset
FER
an
d
p
r
e
-
t
r
ain
ed
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
6
2
-
4
7
7
3
4768
weig
h
ts
o
n
th
e
g
e
n
er
al
d
ataset
I
m
ag
eNe
t
in
ea
ch
ca
s
e
o
f
th
e
tr
ain
in
g
d
ata.
O
b
v
io
u
s
ly
,
th
e
d
if
f
er
en
ce
in
r
esu
lts
is
n
o
t
lar
g
e.
Sp
ec
if
ically
,
wh
e
n
co
m
p
ar
in
g
s
tr
ateg
y
3
A
with
s
tr
ateg
y
2
A,
th
e
r
esu
lts
o
f
3
A
ar
e
s
lig
h
tly
h
ig
h
e
r
th
an
s
tr
ateg
y
2
A
with
an
i
n
c
r
ea
s
e
in
av
er
ag
e
ac
c
u
r
ac
y
o
f
0
.
0
1
1
6
6
6
,
an
in
c
r
ea
s
e
in
av
er
ag
e
s
en
s
itiv
ity
o
f
0
.
0
0
3
3
3
4
,
an
in
cr
ea
s
e
in
a
v
er
ag
e
s
p
ec
if
icity
o
f
0
.
0
2
,
a
n
d
a
n
in
cr
ea
s
e
in
av
er
a
g
e
AUC
o
f
0
.
0
1
2
1
.
Similar
ly
,
wh
en
co
m
p
ar
in
g
s
tr
ateg
y
3
B
with
s
tr
ateg
y
2
B
,
t
h
e
r
esu
lts
o
f
3
B
also
h
av
e
a
s
lig
h
t
in
c
r
ea
s
e
co
m
p
a
r
ed
to
s
tr
ateg
y
2
B
with
an
in
cr
ea
s
e
i
n
av
e
r
ag
e
ac
cu
r
ac
y
o
f
0
.
0
1
6
6
6
6
,
a
n
i
n
cr
ea
s
e
in
av
e
r
ag
e
s
en
s
itiv
ity
o
f
0
.
0
2
3
3
3
4
,
an
in
cr
ea
s
e
in
av
er
a
g
e
s
p
ec
if
i
city
o
f
0
.
0
1
,
an
d
an
in
cr
ea
s
e
in
av
er
ag
e
AUC
o
f
0
.
0
0
7
9
1
1
.
T
h
er
ef
o
r
e,
i
n
o
u
r
o
p
in
io
n
,
wh
en
c
o
n
d
u
ctin
g
f
u
tu
r
e
s
tu
d
ies,
b
o
th
o
f
th
e
a
b
o
v
e
c
ases
ar
e
ca
n
d
id
ates.
Fig
u
r
e
4
.
Av
e
r
ag
e
s
co
r
es o
f
s
tr
ateg
ies 1
A,
2
A,
an
d
3
A
o
n
th
e
in
ter
n
atio
n
al
test
s
et:
ac
cu
r
a
cy
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
AUC (
lef
t)
an
d
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse p
o
s
itiv
es,
an
d
f
alse n
e
g
ativ
e
s
(
r
ig
h
t)
Fig
u
r
e
5
.
Av
e
r
ag
e
s
co
r
es o
f
s
tr
ateg
ies 1
B
,
2
B
,
an
d
3
B
o
n
th
e
in
ter
n
atio
n
al
test
s
et:
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
AUC (
lef
t)
an
d
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse p
o
s
itiv
es,
an
d
f
alse n
e
g
ativ
e
s
(
r
ig
h
t)
T
ab
le
2
.
C
o
m
p
a
r
e
th
e
av
e
r
ag
e
s
co
r
es o
f
s
tr
ateg
ies 2
A
with
3
A
an
d
2
B
with
3
B
o
n
t
h
e
in
ter
n
atio
n
al
test
s
et
C
a
se
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
A
v
e
r
a
g
e
se
n
si
t
i
v
i
t
y
A
v
e
r
a
g
e
sp
e
c
i
f
i
c
i
t
y
A
v
e
r
a
g
e
A
U
C
S
t
r
a
t
e
g
y
2
A
0
.
8
5
9
1
6
7
0
.
8
5
3
3
3
3
0
.
8
6
5
0
.
9
3
1
0
7
8
S
t
r
a
t
e
g
y
3
A
0
.
8
7
0
8
3
3
0
.
8
5
6
6
6
7
0
.
8
8
5
0
.
9
4
3
1
7
8
C
o
m
p
a
r
e
3
A
t
o
2
A
+
0
.
0
1
1
6
6
6
+
0
.
0
0
3
3
3
4
+
0
.
0
2
+
0
.
0
1
2
1
S
t
r
a
t
e
g
y
2
B
0
.
8
5
4
1
6
7
0
.
8
4
3
3
3
3
0
.
8
6
5
0
.
9
3
1
3
3
3
S
t
r
a
t
e
g
y
3
B
0
.
8
7
0
8
3
3
0
.
8
6
6
6
6
7
0
.
8
7
5
0
.
9
3
9
2
4
4
C
o
m
p
a
r
e
3
B
t
o
2
B
+
0
.
0
1
6
6
6
6
+
0
.
0
2
3
3
3
4
+
0
.
0
1
+
0
.
0
0
7
9
1
1
Nex
t,
we
s
ee
th
e
av
er
ag
e
s
co
r
es
o
n
th
e
test
s
et
o
f
Vietn
am
e
s
e
d
ata
s
et.
Fig
u
r
e
6
,
we
ca
n
clea
r
ly
s
ee
th
at
th
e
test
r
esu
lts
o
n
th
e
Vietn
am
ese
test
s
et
ar
e
clea
r
ly
lo
w
wh
en
th
e
m
o
d
els
ar
e
tr
ain
ed
p
u
r
ely
o
n
th
e
in
ter
n
atio
n
al
d
ata
s
et.
Sp
ec
if
ically
,
th
e
h
i
g
h
est
av
er
a
g
e
a
cc
u
r
ac
y
v
alu
e
is
0
.
4
1
3
8
8
9
,
t
h
e
h
ig
h
est
av
er
ag
e
s
p
ec
if
icity
v
alu
e
is
0
.
0
5
,
an
d
th
e
h
ig
h
est
av
er
ag
e
AUC
v
alu
e
is
0
.
2
6
4
9
3
8
.
Alth
o
u
g
h
th
e
h
ig
h
est
av
er
ag
e
s
en
s
itiv
ity
v
alu
e
is
0
.
8
,
b
ec
au
s
e
th
e
av
er
ag
e
s
p
ec
if
icity
is
to
o
lo
w,
th
e
m
o
d
els
af
ter
tr
ain
in
g
ar
e
b
iase
d
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
Dete
ctin
g
a
u
tis
m
w
ith
V
ietn
a
mese
ch
ild
fa
cia
l ima
g
es
u
s
in
g
…
(
Tr
a
n
V
a
n
Th
a
n
h
)
4769
to
war
d
s
o
n
e
class
lab
el
an
d
ig
n
o
r
e
th
e
o
th
e
r
class
lab
el,
s
o
t
h
e
s
ig
n
if
ican
ce
o
f
th
e
a
v
er
ag
e
s
en
s
itiv
ity
v
alu
e
is
n
o
t
lar
g
e
.
Su
ch
lo
w
r
esu
lts
d
o
n
o
t
r
e
f
lect
th
e
ad
v
a
n
tag
e
s
an
d
d
is
ad
v
an
tag
es
o
f
t
h
e
weig
h
t
in
itializatio
n
m
eth
o
d
s
f
o
r
d
ee
p
lear
n
in
g
m
o
d
els.
Ho
wev
er
,
th
ese
r
esu
lts
ac
cu
r
ately
r
e
f
lect
o
u
r
h
y
p
o
th
esis
ab
o
u
t
th
e
d
if
f
er
en
ce
s
in
d
ata
d
o
m
ain
s
an
d
clea
r
ly
r
ev
e
al
th
e
lim
itatio
n
s
o
f
th
e
ex
is
tin
g
in
ter
n
atio
n
al
d
ataset
in
g
en
er
alizin
g
Vietn
am
ese
f
ac
ia
l
im
ag
es.
W
ith
s
tr
ateg
ies
1
B
,
2
B
an
d
3
B
,
d
u
e
to
th
e
i
n
f
lu
e
n
ce
o
f
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
im
ag
e
d
ata
s
am
p
les
in
th
e
co
m
b
in
ed
tr
ain
i
n
g
d
ataset,
th
e
av
er
ag
e
s
co
r
es
h
av
e
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
co
m
p
ar
ed
to
t
h
e
t
r
ain
in
g
ca
s
es
with
p
u
r
e
i
n
ter
n
atio
n
al
d
ata.
Sp
ec
if
ically
,
t
h
e
h
ig
h
est
av
er
ag
e
ac
cu
r
ac
y
v
al
u
e
is
0
.
7
7
5
,
th
e
h
ig
h
est
av
er
ag
e
s
en
s
itiv
ity
v
al
u
e
is
0
.
9
,
th
e
h
i
g
h
est
av
er
ag
e
s
p
ec
if
icity
v
alu
e
is
0
.
6
6
1
1
1
1
,
an
d
th
e
h
ig
h
est
av
e
r
ag
e
AUC
v
alu
e
is
0
.
8
6
7
5
3
1
.
I
t
is
wo
r
th
n
o
tin
g
th
at
in
th
ese
ca
s
es,
th
e
o
r
d
er
o
f
r
esu
lts
o
f
th
e
s
tr
ateg
ies
h
as
c
h
an
g
ed
co
m
p
ar
ed
to
th
e
ca
s
e
o
f
e
v
alu
atin
g
o
n
p
u
r
ely
in
te
r
n
atio
n
al
test
d
ata
.
Alth
o
u
g
h
s
tr
ateg
y
1
B
with
d
ee
p
lear
n
in
g
m
o
d
els
tr
ain
ed
f
r
o
m
s
cr
atch
s
till
g
iv
es
th
e
lo
west
s
co
r
es,
th
e
b
es
t
p
o
s
itio
n
am
o
n
g
th
e
s
co
r
es
is
n
o
t
f
ix
e
d
ly
b
elo
n
g
in
g
to
s
tr
ateg
y
2
B
o
r
s
tr
ateg
y
3
B
.
Str
ateg
y
3
B
win
s
with
av
er
ag
e
s
en
s
itiv
ity
o
f
0
.
9
a
n
d
av
er
ag
e
AUC
o
f
0
.
8
6
7
5
3
1
wh
ile
s
tr
ateg
y
2
B
win
s
with
av
er
ag
e
ac
c
u
r
ac
y
o
f
0
.
7
7
5
a
n
d
av
e
r
ag
e
s
p
ec
if
icity
o
f
0
.
6
6
1
1
1
1
.
W
e
also
o
b
s
er
v
e
m
o
r
e
s
p
ec
if
ically
ab
o
u
t
th
e
d
if
f
er
en
ce
b
etwe
en
s
tr
ateg
y
2
B
an
d
3
B
in
r
esu
lts
in
T
ab
le
3
.
Fig
u
r
e
6
.
Av
e
r
ag
e
ac
c
u
r
ac
y
,
a
v
er
ag
e
s
en
s
itiv
ity
,
av
er
a
g
e
s
p
e
cif
icity
,
an
d
av
e
r
ag
e
AUC o
n
Vietn
am
ese
test
s
et
o
f
s
tr
ateg
ies 1
A,
2
A,
3
A
(
lef
t)
an
d
1
B
,
2
B
,
3
B
(
r
i
g
h
t)
T
ab
le
3
.
C
o
m
p
a
r
e
th
e
av
e
r
ag
e
s
co
r
es o
f
s
tr
ateg
ies 2
B
with
3
B
o
n
Vietn
am
ese
test
s
et
C
a
se
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
A
v
e
r
a
g
e
se
n
si
t
i
v
i
t
y
A
v
e
r
a
g
e
sp
e
c
i
f
i
c
i
t
y
A
v
e
r
a
g
e
A
U
C
S
t
r
a
t
e
g
y
2
B
0
.
7
7
5
0
.
8
8
8
8
8
9
0
.
6
6
1
1
1
1
0
.
8
6
6
2
9
6
S
t
r
a
t
e
g
y
3
B
0
.
7
3
6
1
1
1
0
.
9
0
.
5
7
2
2
2
2
0
.
8
6
7
5
3
1
C
o
m
p
a
r
e
3
B
t
o
2
B
-
0
.
0
3
8
8
8
9
+
0
.
0
1
1
1
1
1
-
0
.
0
8
8
8
8
9
+
0
.
0
0
1
2
3
5
T
h
u
s
,
alth
o
u
g
h
th
e
r
esu
lts
h
av
e
s
h
o
wn
th
e
ad
v
an
tag
e
o
f
u
s
in
g
tr
an
s
f
er
lear
n
in
g
,
th
e
d
if
f
er
en
ce
b
etwe
en
u
s
in
g
p
r
e
-
tr
ain
ed
weig
h
ts
o
n
th
e
f
ac
ial
e
x
p
r
ess
io
n
d
ataset
FER
o
r
p
r
e
-
tr
ain
ed
wei
g
h
ts
o
n
th
e
g
en
er
al
d
ataset
I
m
ag
eNe
t
is
n
o
lo
n
g
er
clea
r
ly
e
v
id
en
t.
I
n
o
u
r
o
p
in
io
n
,
th
is
is
d
u
e
to
th
e
d
if
f
e
r
en
ce
in
th
e
Vietn
am
ese
an
d
th
e
in
ter
n
atio
n
al
d
ataset
th
at
wer
e
p
o
in
ted
o
u
t
in
o
u
r
h
y
p
o
th
esis
.
Alth
o
u
g
h
s
tr
ateg
y
3
B
u
s
es
p
r
e
-
tr
ain
ed
weig
h
ts
f
r
o
m
f
ac
ial
ex
p
r
ess
io
n
d
ata,
th
e
d
ataset
FER
is
n
o
t
Vietn
am
ese
d
ata
wh
ile
t
h
e
te
s
t
d
ata
co
n
s
id
er
e
d
h
er
e
is
p
u
r
el
y
Vietn
am
ese
d
ata.
T
h
er
ef
o
r
e,
s
tr
ateg
y
3
B
n
o
lo
n
g
e
r
h
as
a
clea
r
ad
v
a
n
ta
g
e
as
in
th
e
a
b
o
v
e
ex
p
er
im
en
t w
h
e
n
ev
alu
ati
n
g
o
n
th
e
in
ter
n
atio
n
al
test
s
et.
Fro
m
th
e
ab
o
v
e
ev
id
e
n
ce
,
we
o
n
ce
ag
ain
s
ee
th
at
th
er
e
is
v
alu
e
in
in
itializin
g
th
e
weig
h
ts
o
f
d
ee
p
lear
n
in
g
m
o
d
els
f
r
o
m
m
o
d
els
p
r
etr
ain
ed
o
n
f
ac
ial
ex
p
r
ess
io
n
d
ata
s
u
ch
as
FER
o
r
o
n
g
en
er
al
d
ata
s
u
ch
as
I
m
ag
eNe
t.
W
h
en
co
n
d
u
ctin
g
f
u
r
th
er
r
esear
ch
,
esp
ec
ially
wh
en
wo
r
k
in
g
with
lo
ca
l
d
a
ta
s
o
u
r
ce
s
s
u
ch
as
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
ial
d
ata,
b
o
th
o
f
th
ese
o
p
tio
n
s
s
h
o
u
ld
b
e
co
n
s
id
er
ed
.
T
o
g
et
a
m
o
r
e
d
etailed
v
iew,
we
will lo
o
k
at
th
e
ev
alu
atio
n
r
esu
lts
o
n
ea
ch
in
d
i
v
id
u
al
d
ee
p
lear
n
in
g
m
o
d
el
i
n
d
etail.
T
ab
le
4
d
escr
ib
es
th
e
d
etai
led
co
m
p
a
r
is
o
n
o
f
th
e
r
esu
lts
o
f
d
ee
p
lea
r
n
in
g
m
o
d
els
b
etwe
en
s
tr
ateg
y
3
A
an
d
s
tr
ateg
y
2
A
wh
en
ev
alu
atin
g
o
n
th
e
in
ter
n
atio
n
al
test
s
et.
W
h
en
co
m
p
ar
in
g
th
e
s
co
r
es
o
f
s
tr
ateg
ies
3
A
with
s
tr
ateg
y
2
A,
we
s
ee
s
o
m
e
ch
an
g
es
as
f
o
llo
ws.
Fo
r
ac
cu
r
ac
y
s
co
r
e,
th
e
i
n
cr
ea
s
e
o
cc
u
r
s
in
3
m
o
d
els
with
th
e
lar
g
est
d
if
f
er
en
ce
b
ein
g
0
.
0
2
3
3
3
3
o
f
Den
s
eNe
t1
2
1
m
o
d
el
an
d
th
e
o
n
ly
d
eg
r
ad
atio
n
ca
s
e
is
R
esNet5
0
with
th
e
lev
el
o
f
0
.
0
0
6
6
6
7
.
Fo
r
s
en
s
itiv
ity
,
th
e
i
n
cr
ea
s
e
also
o
cc
u
r
s
in
3
m
o
d
els
with
th
e
lar
g
est
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.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
4
7
6
2
-
4
7
7
3
4770
d
if
f
er
en
ce
b
ein
g
0
.
0
1
3
3
3
4
o
f
Den
s
eNe
t1
2
1
m
o
d
el
an
d
th
e
o
n
ly
d
eg
r
a
d
atio
n
ca
s
e
is
Alex
Net
with
th
e
lev
el
o
f
0
.
0
1
3
3
3
3
.
Fo
r
s
p
ec
if
icity
s
co
r
e,
th
e
in
cr
ea
s
e
also
o
cc
u
r
s
i
n
3
m
o
d
els
with
th
e
lar
g
est
d
if
f
er
en
ce
b
ein
g
0
.
0
4
o
f
Alex
n
Net
m
o
d
el
a
n
d
t
h
e
o
n
ly
d
eg
r
ad
atio
n
ca
s
e
is
R
esNet
5
0
with
th
e
lev
el
o
f
0
.
0
2
.
Fin
ally
,
th
e
AUC
also
in
cr
ea
s
es
ac
r
o
s
s
th
e
t
h
r
ee
m
o
d
els
with
th
e
lar
g
est
d
if
f
er
en
c
e
b
ein
g
0
.
0
2
2
2
2
2
f
o
r
th
e
Den
s
eNe
t1
2
1
m
o
d
el
an
d
th
e
o
n
l
y
d
ec
lin
e
b
ei
n
g
Alex
Net
with
0
.
0
0
2
0
4
4
.
I
n
g
en
er
al,
t
h
e
s
co
r
e
d
if
f
e
r
en
ce
is
n
o
t
lar
g
e
with
th
e
ab
s
o
lu
te
v
alu
e
n
o
t
e
x
ce
ed
in
g
0
.
0
4
an
d
u
s
u
ally
f
lu
ctu
atin
g
ar
o
u
n
d
0
.
0
1
.
Ov
er
all,
t
h
e
im
p
ac
t
o
f
th
e
two
s
tr
ateg
ies
o
n
ea
ch
d
ee
p
lear
n
in
g
m
o
d
el
wh
e
n
ev
alu
ated
o
n
two
i
n
ter
n
atio
n
al
test
s
et
s
d
o
es n
o
t h
av
e
a
cle
ar
d
if
f
er
e
n
ce
.
T
ab
le
4
.
Deta
iled
s
co
r
e
c
o
m
p
a
r
is
o
n
o
f
s
tr
ateg
ies 2
A
with
3
A
o
n
in
ter
n
atio
n
al
test
s
et
C
a
se
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
AUC
S
t
r
a
t
e
g
y
2
A
R
e
sN
e
t
3
4
0
.
8
7
3
3
3
3
0
.
8
8
6
6
6
7
0
.
8
6
0
.
9
4
4
7
5
6
S
t
r
a
t
e
g
y
3
A
R
e
sN
e
t
3
4
0
.
8
9
0
.
8
9
3
3
3
3
0
.
8
8
6
6
6
7
0
.
9
5
3
7
7
8
C
o
m
p
a
r
e
3
A
t
o
2
A
R
e
sN
e
t
3
4
+
0
.
0
1
6
6
6
7
+
0
.
0
0
6
6
6
6
+
0
.
0
2
6
6
6
7
+
0
.
0
0
9
0
2
2
S
t
r
a
t
e
g
y
2
A
R
e
sN
e
t
50
0
.
8
6
3
3
3
3
0
.
8
3
3
3
3
3
0
.
8
9
3
3
3
3
0
.
9
2
3
9
1
1
S
t
r
a
t
e
g
y
3
A
R
e
sN
e
t
5
0
0
.
8
5
6
6
6
7
0
.
8
4
0
.
8
7
3
3
3
3
0
.
9
4
3
1
1
1
C
o
m
p
a
r
e
3
A
t
o
2
A
R
e
sN
e
t
5
0
-
0
.
0
0
6
6
6
7
+
0
.
0
0
6
6
6
7
-
0
.
0
2
+
0
.
0
1
9
2
S
t
r
a
t
e
g
y
2
A
A
l
e
x
N
e
t
0
.
8
4
0
.
8
4
0
.
8
4
0
.
9
2
6
1
3
3
S
t
r
a
t
e
g
y
3
A
A
l
e
x
N
e
t
0
.
8
5
3
3
3
3
0
.
8
2
6
6
6
7
0
.
8
8
0
.
9
2
4
0
8
9
C
o
m
p
a
r
e
3
A
t
o
2
A
A
l
e
x
N
e
t
+
0
.
0
1
3
3
3
3
-
0
.
0
1
3
3
3
3
+
0
.
0
4
-
0
.
0
0
2
0
4
4
S
t
r
a
t
e
g
y
2
A
D
e
n
seN
e
t
1
2
1
0
.
8
6
0
.
8
5
3
3
3
3
0
.
8
6
6
6
6
7
0
.
9
2
9
5
1
1
S
t
r
a
t
e
g
y
3
A
D
e
n
seN
e
t
1
2
1
0
.
8
8
3
3
3
3
0
.
8
6
6
6
6
7
0
.
9
0
.
9
5
1
7
3
3
C
o
m
p
a
r
e
3
A
t
o
2
A
D
e
n
seN
e
t
1
2
1
+
0
.
0
2
3
3
3
3
+
0
.
0
1
3
3
3
4
+
0
.
0
3
3
3
3
3
+
0
.
0
2
2
2
2
2
Nex
t,
T
ab
le
5
d
escr
ib
es
th
e
d
etailed
co
m
p
ar
is
o
n
o
f
th
e
r
esu
lts
o
f
d
ee
p
lear
n
in
g
m
o
d
e
ls
b
etwe
en
s
tr
ateg
y
3
B
an
d
s
tr
ateg
y
2
B
wh
en
ev
alu
atin
g
o
n
th
e
in
ter
n
atio
n
al
test
s
et.
W
h
en
co
m
p
ar
in
g
th
e
s
co
r
es
o
f
s
tr
ateg
y
3
B
with
s
tr
ateg
y
2
B
,
we
s
ee
s
o
m
e
c
h
an
g
es
as
f
o
llo
ws.
Fo
r
ac
cu
r
ac
y
s
co
r
e,
t
h
e
in
cr
ea
s
e
o
cc
u
r
s
in
2
m
o
d
els
with
th
e
lar
g
est
d
if
f
er
en
ce
b
ein
g
0
.
0
5
3
3
3
3
o
f
A
lex
N
et
m
o
d
el
an
d
th
e
o
n
l
y
d
eg
r
ad
atio
n
ca
s
e
is
R
es
N
et3
4
with
th
e
lev
el
o
f
0
.
0
0
6
6
6
6
.
Fo
r
s
en
s
itiv
ity
,
th
e
i
n
cr
ea
s
e
also
o
cc
u
r
s
in
2
m
o
d
els
with
th
e
lar
g
est
d
if
f
er
en
ce
b
ei
n
g
0
.
0
8
6
6
6
6
o
f
Den
s
e
N
et1
2
1
m
o
d
el
an
d
t
h
e
o
n
ly
d
eg
r
ad
atio
n
ca
s
e
is
R
es
N
et3
4
with
th
e
lev
el
o
f
0
.
0
7
3
3
3
3
.
T
ab
le
5
.
Deta
iled
s
co
r
e
c
o
m
p
a
r
is
o
n
o
f
s
tr
ateg
ies 2
B
with
3
B
o
n
in
ter
n
atio
n
al
test
s
et
C
a
se
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
AUC
S
t
r
a
t
e
g
y
2
B
R
e
sN
e
t
3
4
0
.
8
8
3
3
3
3
0
.
9
3
3
3
3
3
0
.
8
3
3
3
3
3
0
.
9
3
4
6
2
2
S
t
r
a
t
e
g
y
3
B
R
e
sN
e
t
3
4
0
.
8
7
6
6
6
7
0
.
8
6
0
.
8
9
3
3
3
3
0
.
9
4
0
0
4
4
C
o
m
p
a
r
e
3
B
t
o
2
B
R
e
sN
e
t
3
4
-
0
.
0
0
6
6
6
6
-
0
.
0
7
3
3
3
3
+
0
.
0
6
+
0
.
0
0
5
4
2
2
S
t
r
a
t
e
g
y
2
B
R
e
sN
e
t
5
0
0
.
8
6
3
3
3
3
0
.
8
6
0
.
8
6
6
6
6
7
0
.
9
4
3
3
7
8
S
t
r
a
t
e
g
y
3
B
R
e
sN
e
t
5
0
0
.
8
6
3
3
3
3
0
.
8
6
0
.
8
6
6
6
6
7
0
.
9
3
8
2
2
2
C
o
m
p
a
r
e
3
B
t
o
2
B
R
e
sN
e
t
5
0
0
0
0
-
0
.
0
0
5
1
5
6
S
t
r
a
t
e
g
y
2
B
A
l
e
x
N
e
t
0
.
8
3
0
.
7
7
3
3
3
3
0
.
8
8
6
6
6
7
0
.
9
1
6
3
1
1
S
t
r
a
t
e
g
y
3
B
A
l
e
x
N
e
t
0
.
8
8
3
3
3
3
0
.
8
5
3
3
3
3
0
.
9
1
3
3
3
3
0
.
9
3
1
7
3
3
C
o
m
p
a
r
e
3
B
t
o
2
B
A
l
e
x
N
e
t
+
0
.
0
5
3
3
3
3
+
0
.
0
8
+
0
.
0
2
6
6
6
7
+
0
.
0
1
5
4
2
2
S
t
r
a
t
e
g
y
2
B
D
e
n
seN
e
t
1
2
1
0
.
8
4
0
.
8
0
6
6
6
7
0
.
8
7
3
3
3
3
0
.
9
3
1
0
2
2
S
t
r
a
t
e
g
y
3
B
D
e
n
seN
e
t
1
2
1
0
.
8
6
0
.
8
9
3
3
3
3
0
.
8
2
6
6
6
7
0
.
9
4
6
9
7
8
C
o
m
p
a
r
e
3
B
t
o
2
B
D
e
n
seN
e
t
1
2
1
+
0
.
0
2
+
0
.
0
8
6
6
6
6
-
0
.
0
4
6
6
6
6
+
0
.
0
1
5
9
5
6
Fo
r
s
p
e
cif
ic
i
ty
s
co
r
e,
th
e
in
cr
ea
s
e
a
l
s
o
o
c
cu
r
s
in
2
m
o
d
e
l
s
wi
th
th
e
l
ar
g
e
s
t
d
i
f
f
er
en
c
e
b
e
i
n
g
0
.
0
6
o
f
R
e
s
N
e
t3
4
m
o
d
e
l
a
n
d
th
e
o
n
l
y
d
eg
r
ad
a
tio
n
ca
s
e
is
D
en
s
e
N
e
t1
2
1
wi
th
th
e
l
ev
e
l
o
f
0
.
0
4
6
6
6
6
.
Fin
al
ly
,
th
e
AU
C
al
s
o
in
c
r
ea
s
es
ac
r
o
s
s
th
e
th
r
ee
m
o
d
e
ls
w
ith
th
e
l
ar
g
e
s
t
d
if
f
er
en
c
e
b
e
in
g
0
.
0
1
5
9
5
6
f
o
r
th
e
D
en
s
e
N
et
1
2
1
m
o
d
el
an
d
th
e
o
n
ly
d
ec
l
in
e
b
ein
g
R
es
N
e
t5
0
wi
th
0
.
0
0
5
1
5
6
.
W
e
s
ee
th
a
t
t
h
e
ab
s
o
lu
te
v
alu
e
s
o
f
th
e
s
co
r
e
d
ev
i
at
io
n
s
h
av
e
in
c
r
ea
s
ed
s
i
g
n
if
i
ca
n
t
ly
wh
en
co
m
p
ar
ed
t
o
th
e
ab
o
v
e
ev
alu
at
io
n
c
a
s
e
o
f
s
tr
a
teg
y
3
A
an
d
s
t
r
a
teg
y
2
A.
I
n
wh
i
ch
,
th
e
l
a
r
g
e
s
t
d
if
f
e
r
en
c
e
i
s
0
.
0
8
6
6
6
6
an
d
i
t
i
s
m
o
r
e
th
an
t
wi
ce
as
h
ig
h
a
s
t
h
e
l
ar
g
e
s
t
d
if
f
e
r
en
c
e
o
f
a
d
e
ep
l
ea
r
n
in
g
ar
ch
it
ec
tu
r
e
b
et
we
en
s
tr
a
teg
y
3
A
an
d
s
tr
at
eg
y
2
A
.
T
h
is
al
s
o
cle
ar
ly
r
ef
lec
t
s
th
e
d
if
f
e
r
en
c
e
in
d
a
ta
d
o
m
a
in
b
e
tw
ee
n
V
ie
tn
am
d
a
ta
an
d
in
t
e
r
n
at
io
n
a
l
d
a
ta.
H
er
e
w
e
ca
n
s
e
e
th
e
i
m
p
ac
t
o
f
Vi
etn
am
e
s
e
ch
i
ld
r
en
's
f
ac
ia
l
i
m
ag
e
d
a
ta
wh
e
n
p
a
r
t
ic
ip
a
tin
g
in
th
e
t
r
a
in
in
g
d
a
ta
an
d
i
t
cle
ar
ly
h
a
s
an
im
p
ac
t
o
n
ch
an
g
in
g
th
e
d
i
s
tr
ib
u
t
io
n
o
f
d
a
ta
in
th
e
tr
ain
in
g
s
et.
N
ex
t,
w
e
co
n
s
id
e
r
th
e
d
et
ai
led
co
m
p
a
r
i
s
o
n
o
f
th
e
r
es
u
l
t
s
o
f
d
e
ep
lea
r
n
in
g
m
o
d
el
s
b
et
we
en
s
tr
a
teg
y
3
B
an
d
s
t
r
a
teg
y
2
B
wh
en
ev
alu
at
in
g
o
n
th
e
Vi
etn
am
e
s
e
te
s
t
s
et
in
T
ab
le
6
.
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
Dete
ctin
g
a
u
tis
m
w
ith
V
ietn
a
mese
ch
ild
fa
cia
l ima
g
es
u
s
in
g
…
(
Tr
a
n
V
a
n
Th
a
n
h
)
4771
T
ab
le
6
.
Deta
iled
s
co
r
e
c
o
m
p
a
r
is
o
n
o
f
s
tr
ateg
ies 2
B
with
3
B
o
n
Vietn
am
ese
test
s
et
C
a
se
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
S
e
n
s
i
t
i
v
i
t
y
S
p
e
c
i
f
i
c
i
t
y
AUC
S
t
r
a
t
e
g
y
2
B
R
e
sN
e
t
3
4
0
.
7
4
4
4
4
4
0
.
7
7
7
7
7
8
0
.
7
1
1
1
1
1
0
.
7
8
8
1
4
8
S
t
r
a
t
e
g
y
3
B
R
e
sN
e
t
3
4
0
.
8
2
2
2
2
2
0
.
8
8
8
8
8
9
0
.
7
5
5
5
5
6
0
.
9
1
3
0
8
6
C
o
m
p
a
r
e
3
B
t
o
2
B
R
e
sN
e
t
3
4
+
0
.
0
7
7
7
7
8
+
0
.
1
1
1
1
1
1
+
0
.
0
4
4
4
4
4
+
0
.
1
2
4
9
3
8
S
t
r
a
t
e
g
y
2
B
R
e
sN
e
t
5
0
0
.
7
7
7
7
7
8
0
.
9
1
1
1
1
1
0
.
6
4
4
4
4
4
0
.
9
0
1
7
2
8
S
t
r
a
t
e
g
y
3
B
R
e
sN
e
t
5
0
0
.
7
0
.
8
8
8
8
8
9
0
.
5
1
1
1
1
1
0
.
8
8
0
4
9
4
C
o
m
p
a
r
e
3
B
t
o
2
B
R
e
sN
e
t
5
0
-
0
.
0
7
7
7
7
8
-
0
.
0
2
2
2
2
2
-
0
.
1
3
3
3
3
3
-
0
.
0
2
1
2
3
4
S
t
r
a
t
e
g
y
2
B
A
l
e
x
N
e
t
0
.
7
7
7
7
7
8
0
.
9
3
3
3
3
3
0
.
6
2
2
2
2
2
0
.
8
8
8
3
9
5
S
t
r
a
t
e
g
y
3
B
A
l
e
x
N
e
t
0
.
7
1
1
1
1
1
0
.
9
1
1
1
1
1
0
.
5
1
1
1
1
1
0
.
8
2
3
2
1
0
C
o
m
p
a
r
e
3
B
t
o
2
B
A
l
e
x
N
e
t
-
0
.
0
6
6
6
6
7
-
0
.
0
2
2
2
2
2
-
0
.
1
1
1
1
1
1
-
0
.
0
6
5
1
8
5
S
t
r
a
t
e
g
y
2
B
D
e
n
seN
e
t
1
2
1
0
.
8
0
.
9
3
3
3
3
3
0
.
6
6
6
6
6
7
0
.
8
8
6
9
1
4
S
t
r
a
t
e
g
y
3
B
D
e
n
seN
e
t
1
2
1
0
.
7
1
1
1
1
1
0
.
9
1
1
1
1
1
0
.
5
1
1
1
1
1
0
.
8
5
3
3
3
3
C
o
m
p
a
r
e
3
B
t
o
2
B
D
e
n
seN
e
t
1
2
1
-
0
.
0
8
8
8
8
9
-
0
.
0
2
2
2
2
2
-
0
.
1
5
5
5
5
6
-
0
.
0
3
3
5
8
1
C
lear
ly
,
we
h
av
e
s
ee
n
s
tr
o
n
g
d
if
f
er
en
ce
s
b
etwe
en
th
e
r
esu
lt
in
g
s
co
r
es.
Fo
r
th
e
ac
cu
r
ac
y
s
co
r
e,
th
er
e
is
o
n
ly
o
n
e
in
cr
ea
s
e
i
n
R
es
N
et3
4
with
0
.
0
7
7
7
7
8
an
d
th
e
r
e
m
ain
in
g
t
h
r
ee
d
ec
r
ea
s
e
with
t
h
e
lar
g
est
d
ec
r
ea
s
e
b
ein
g
0
.
0
8
8
8
8
9
f
o
r
Den
s
e
N
et1
2
1
.
Fo
r
t
h
e
s
en
s
itiv
ity
s
co
r
e,
we
s
ee
th
e
s
am
e
with
o
n
e
in
cr
ea
s
e
in
R
es
N
et3
4
with
0
.
1
1
1
1
1
1
an
d
th
r
ee
d
ec
r
e
ase
with
th
e
o
v
er
all
d
ec
r
ea
s
e
b
ein
g
0
.
0
2
2
2
2
2
.
Fo
r
t
h
e
s
p
ec
if
icity
s
co
r
e,
we
also
s
ee
o
n
e
in
cr
ea
s
e
in
R
es
N
et3
4
with
0
.
0
4
4
4
4
4
an
d
th
r
ee
d
ec
r
e
ase
with
th
e
lar
g
est
d
ec
r
ea
s
e
b
ein
g
0
.
1
5
5
5
5
6
f
o
r
Den
s
e
N
et1
2
1
.
Fo
r
th
e
AUC
s
co
r
e,
th
er
e
is
also
o
n
ly
o
n
e
in
cr
ea
s
e
in
R
es
N
et3
4
with
0
.
1
2
4
9
3
8
an
d
th
e
r
em
ain
in
g
th
r
ee
d
ec
r
ea
s
e
with
th
e
lar
g
est
d
ec
r
ea
s
e
b
ein
g
0
.
0
6
5
1
8
5
f
o
r
Alex
N
et.
C
lear
l
y
,
th
e
r
esu
lts
h
av
e
a
s
tr
o
n
g
er
d
is
cr
ep
an
cy
with
th
e
h
ig
h
est
d
if
f
er
en
ce
b
ei
n
g
0
.
1
5
5
5
5
6
an
d
it
is
s
ig
n
i
f
ican
tly
h
ig
h
er
th
an
th
e
tw
o
esti
m
ates
g
iv
en
ab
o
v
e
wh
en
u
s
in
g
in
ter
n
atio
n
al
test
d
ata
.
T
h
is
h
as
s
tr
o
n
g
ly
d
em
o
n
s
tr
ated
th
e
d
i
f
f
er
en
ce
b
etwe
en
th
e
Vietn
am
ese
ch
ild
f
ac
ial
d
ata
d
o
m
ai
n
an
d
th
e
in
ter
n
atio
n
al
o
n
e.
T
h
e
r
esu
lts
wer
e
g
en
e
r
ated
with
th
e
Vietn
am
ese
test
s
et
an
d
th
e
tr
ain
in
g
d
ata
with
Vietn
am
ese
d
ata
in
th
e
m
in
o
r
ity
.
T
h
er
e
f
o
r
e,
th
e
co
llectio
n
o
f
lo
ca
l
d
ata
s
u
ch
as
Vietn
a
m
ese
ch
ild
f
ac
ial
d
ata
f
o
r
t
h
e
au
tis
m
d
iag
n
o
s
is
s
y
s
tem
f
r
o
m
f
ac
ial
im
a
g
es
in
Vietn
am
h
as c
lear
ly
d
em
o
n
s
tr
ated
its
r
o
le
an
d
im
p
o
r
tan
ce
th
r
o
u
g
h
th
e
ev
i
d
en
ce
o
f
th
e
r
esu
lts
.
5.
CO
NCLU
SI
O
N
Ou
r
ar
ticle
f
o
cu
s
es
o
n
a
n
aly
zi
n
g
a
n
d
d
is
cu
s
s
in
g
th
e
r
o
le
o
f
in
ter
n
atio
n
al
a
n
d
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
e
d
ata
an
d
th
e
in
f
lu
en
ce
o
f
d
if
f
er
en
t
p
r
etr
ai
n
ed
weig
h
ts
o
f
d
ee
p
lear
n
in
g
m
o
d
els
in
th
e
au
ti
s
m
class
if
icatio
n
p
r
o
b
lem
.
T
h
e
d
ata
f
o
r
th
e
ex
p
e
r
im
en
ts
in
th
is
s
tu
d
y
wer
e
c
o
llected
f
r
o
m
s
e
v
er
al
k
i
n
d
er
g
ar
ten
s
in
Ho
C
h
i
Min
h
C
ity
,
Vietn
a
m
alo
n
g
with
an
in
ter
n
ati
o
n
al
d
ataset
d
o
wn
lo
a
d
ed
f
r
o
m
th
e
Kag
g
le
p
latf
o
r
m
.
T
h
e
p
r
o
p
o
s
ed
r
esear
ch
f
o
cu
s
es
o
n
d
esig
n
in
g
s
tr
ateg
ies
an
d
an
aly
zin
g
a
n
d
ev
alu
atin
g
r
esu
lts
to
b
u
ild
an
au
tis
m
class
if
ica
tio
n
ap
p
licatio
n
with
th
e
f
ac
ial
d
ata
o
f
Vietn
am
ese
ch
ild
r
en
.
T
h
e
e
x
p
er
im
en
tal
r
esu
lts
ac
h
iev
ed
o
n
d
if
f
e
r
en
t
m
etr
ics
s
u
ch
as
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
an
d
AUC.
T
h
e
f
ig
u
r
es
p
o
in
ted
o
u
t
th
e
n
ec
ess
ar
y
o
f
p
r
etr
ain
e
d
weig
h
ts
o
f
d
ee
p
lear
n
in
g
m
o
d
els
an
d
th
e
r
o
le
o
f
in
ter
n
atio
n
al
an
d
Vietn
am
ese
ch
ild
r
en
'
s
f
ac
e
d
ata
in
th
e
tr
ai
n
in
g
p
h
ase.
W
e
also
d
ee
p
ly
d
i
s
cu
s
s
th
e
ex
p
o
s
u
r
e
o
f
d
ata
d
is
t
r
ib
u
tio
n
d
if
f
er
e
n
ce
s
in
th
e
p
r
o
p
o
s
ed
s
tr
ateg
ies to
h
i
g
h
lig
h
t th
e
im
p
o
r
tan
ce
o
f
c
o
llectin
g
f
ac
ial
d
ata
o
f
Vietn
am
e
s
e
ch
ild
r
en
f
o
r
n
ex
t
r
esear
ch
es.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
th
o
r
s
wis
h
to
th
an
k
L
ac
Ho
n
g
Un
iv
e
r
s
ity
f
o
r
t
h
e
f
in
a
n
cial
s
u
p
p
o
r
t.
RE
F
E
R
E
NC
E
S
[
1
]
H
.
H
o
d
g
e
,
C
.
F
e
a
l
k
o
,
a
n
d
N
.
S
o
a
r
e
s,
“
A
u
t
i
sm
s
p
e
c
t
r
u
m
d
i
s
o
r
d
e
r
:
d
e
f
i
n
i
t
i
o
n
,
e
p
i
d
e
m
i
o
l
o
g
y
,
c
a
u
ses,
a
n
d
c
l
i
n
i
c
a
l
e
v
a
l
u
a
t
i
o
n
,
”
T
ra
n
s
l
a
t
i
o
n
a
l
Pe
d
i
a
t
r
i
c
s
,
v
o
l
.
9
,
p
p
.
5
5
–
6
5
,
2
0
2
0
.
[
2
]
W
H
O
,
“
A
u
t
i
sm,
”
W
o
rl
d
H
e
a
l
t
h
O
r
g
a
n
i
z
a
t
i
o
n
(
WHO
)
,
2
0
2
3
.
h
t
t
p
s
:
/
/
w
w
w
.
w
h
o
.
i
n
t
/
n
e
w
s
-
r
o
o
m/
f
a
c
t
-
s
h
e
e
t
s
/
d
e
t
a
i
l
/
a
u
t
i
sm
-
s
p
e
c
t
r
u
m
-
d
i
s
o
r
d
e
r
s (a
c
c
e
ss
e
d
F
e
b
.
1
2
,
2
0
2
5
)
.
[
3
]
C
.
L
o
r
d
,
M
.
El
s
a
b
b
a
g
h
,
G
.
B
a
i
r
d
,
a
n
d
J.
V
e
e
n
s
t
r
a
-
V
a
n
d
e
r
w
e
e
l
e
,
“
A
u
t
i
sm
sp
e
c
t
r
u
m
d
i
so
r
d
e
r
,
”
T
h
e
L
a
n
c
e
t
,
v
o
l
.
3
9
2
,
n
o
.
1
0
1
4
6
,
p
p
.
5
0
8
–
5
2
0
,
A
u
g
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
S
0
1
4
0
-
6
7
3
6
(
1
8
)
3
1
1
2
9
-
2.
[
4
]
S
.
B
.
S
u
l
k
e
s,
“
A
u
t
i
sm
s
p
e
c
t
r
u
m
d
i
so
r
d
e
r
,
”
M
S
D
,
2
0
2
5
.
h
t
t
p
s
:
/
/
w
w
w
.
msd
ma
n
u
a
l
s.c
o
m
/
p
r
o
f
e
ss
i
o
n
a
l
/
p
e
d
i
a
t
r
i
c
s/
l
e
a
r
n
i
n
g
-
a
n
d
-
d
e
v
e
l
o
p
me
n
t
a
l
-
d
i
so
r
d
e
r
s/
a
u
t
i
s
m
-
sp
e
c
t
r
u
m
-
d
i
so
r
d
e
r
s (
a
c
c
e
sse
d
F
e
b
.
1
2
,
2
0
2
5
)
.
[
5
]
A
.
J.
B
a
x
t
e
r
,
T
.
S
.
B
r
u
g
h
a
,
H
.
E
.
Er
s
k
i
n
e
,
R
.
W
.
S
c
h
e
u
r
e
r
,
T
.
V
o
s,
a
n
d
J
.
G
.
S
c
o
t
t
,
“
T
h
e
e
p
i
d
e
m
i
o
l
o
g
y
a
n
d
g
l
o
b
a
l
b
u
r
d
e
n
o
f
a
u
t
i
s
m
sp
e
c
t
r
u
m
d
i
s
o
r
d
e
r
s,”
Psy
c
h
o
l
o
g
i
c
a
l
M
e
d
i
c
i
n
e
,
v
o
l
.
4
5
,
n
o
.
3
,
p
p
.
6
0
1
–
6
1
3
,
F
e
b
.
2
0
1
5
,
d
o
i
:
1
0
.
1
0
1
7
/
S
0
0
3
3
2
9
1
7
1
4
0
0
1
7
2
X
.
[
6
]
J.
B
a
i
o
e
t
a
l
.
,
“
P
r
e
v
a
l
e
n
c
e
o
f
a
u
t
i
sm
sp
e
c
t
r
u
m
d
i
so
r
d
e
r
a
m
o
n
g
c
h
i
l
d
r
e
n
a
g
e
d
8
y
e
a
r
s
—
a
u
t
i
sm
a
n
d
d
e
v
e
l
o
p
m
e
n
t
a
l
d
i
sa
b
i
l
i
t
i
e
s
mo
n
i
t
o
r
i
n
g
n
e
t
w
o
r
k
,
1
1
S
i
t
e
s
,
U
n
i
t
e
d
S
t
a
t
e
s,
2
0
1
4
,
”
M
MW
R
.
S
u
r
v
e
i
l
l
a
n
c
e
S
u
m
m
a
r
i
e
s
,
v
o
l
.
6
7
,
n
o
.
6
,
p
p
.
1
–
2
3
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
1
5
5
8
5
/
mm
w
r
.
ss6
7
0
6
a
1
.
[
7
]
R
.
J.
L
a
n
d
a
,
“
Ef
f
i
c
a
c
y
o
f
e
a
r
l
y
i
n
t
e
r
v
e
n
t
i
o
n
s
f
o
r
i
n
f
a
n
t
s
a
n
d
y
o
u
n
g
c
h
i
l
d
r
e
n
w
i
t
h
,
a
n
d
a
t
r
i
s
k
f
o
r
,
a
u
t
i
sm
s
p
e
c
t
r
u
m
d
i
so
r
d
e
r
s
,”
I
n
t
e
r
n
a
t
i
o
n
a
l
Re
v
i
e
w
o
f
Ps
y
c
h
i
a
t
r
y
,
v
o
l
.
3
0
,
n
o
.
1
,
p
p
.
2
5
–
3
9
,
J
a
n
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
8
0
/
0
9
5
4
0
2
6
1
.
2
0
1
8
.
1
4
3
2
5
7
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.