T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
1
8
,
No
.
5
,
Octo
b
er
2
0
2
0
,
p
p
.
2
4
6
3
~
2
4
7
1
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
: 1
0
.
1
2
9
2
8
/TE
L
KOM
NI
K
A.
v
1
8
i
5
.
1
6
7
1
7
2463
J
o
ur
na
l
ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Dev
elo
pment of
video
-
ba
sed emo
ti
o
n recog
nition us
i
ng
deep learning
wit
h G
o
o
g
le Cola
b
T
eddy
Su
ry
a
G
un
a
wa
n
1
,
Ar
s
ela
n Ash
ra
f
2
,
B
o
b S
ub
ha
n
Riza
3
,
E
dy
Vict
o
r
H
a
ry
a
nto
4
,
Rik
a
Ro
s
nelly
5
,
M
ira
K
a
rt
iwi
6
,
Z
uria
t
i J
a
nin
7
1,
2
De
p
a
rtme
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
C
o
m
p
u
ter E
n
g
in
e
e
rin
g
,
I
n
tern
a
ti
o
n
a
l
Isla
m
ic Un
iv
e
rsity
M
a
lay
sia
,
M
a
lay
sia
1,
3,
4,
5
Fa
c
u
lt
y
o
f
En
g
in
e
e
rin
g
a
n
d
Co
m
p
u
ter
S
c
ien
c
e
,
Un
i
v
e
rsitas
P
o
ten
si Ut
a
m
a
,
In
d
o
n
e
sia
6
De
p
a
rtem
e
n
t
o
f
In
f
o
rm
a
ti
o
n
S
y
st
e
m
s,
In
tern
a
ti
o
n
a
l
Isla
m
ic Un
iv
e
r
sity
M
a
lay
sia
,
M
a
lay
sia
7
F
a
c
u
lt
y
o
f
El
e
c
tri
c
a
l
E
n
g
i
n
e
e
rin
g
,
Un
iv
e
rsit
i
Tek
n
o
l
o
g
i
M
ARA
,
M
a
lay
sia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
17
,
2
0
2
0
R
ev
is
ed
Ap
r
29
,
2
0
2
0
Acc
ep
ted
May
11
,
2
0
2
0
Emo
ti
o
n
re
c
o
g
n
it
io
n
u
sin
g
ima
g
e
s,
v
i
d
e
o
s,
o
r
sp
e
e
c
h
a
s i
n
p
u
t
is co
n
sid
e
re
d
a
s
a
h
o
t
t
o
p
ic
in
t
h
e
field
o
f
re
se
a
rc
h
o
v
e
r
s
o
m
e
y
e
a
rs.
W
it
h
t
h
e
i
n
tr
o
d
u
c
t
io
n
o
f
d
e
e
p
lea
rn
i
n
g
tec
h
n
iq
u
e
s,
e
.
g
.
,
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CN
N),
a
p
p
li
e
d
in
e
m
o
ti
o
n
re
c
o
g
n
i
ti
o
n
,
h
a
s
p
r
o
d
u
c
e
d
p
r
o
m
isin
g
re
su
lt
s.
H
u
m
a
n
fa
c
ial
e
x
p
re
ss
io
n
s
a
re
c
o
n
sid
e
re
d
a
s
c
rit
ica
l
c
o
m
p
o
n
e
n
ts
in
u
n
d
e
rsta
n
d
in
g
o
n
e
'
s
e
m
o
ti
o
n
s.
T
h
is
p
a
p
e
r
sh
e
d
s
li
g
h
t
o
n
re
c
o
g
n
izi
n
g
t
h
e
e
m
o
ti
o
n
s
u
sin
g
d
e
e
p
lea
rn
in
g
tec
h
n
i
q
u
e
s
fro
m
t
h
e
v
i
d
e
o
s.
T
h
e
m
e
th
o
d
o
l
o
g
y
o
f
t
h
e
r
e
c
o
g
n
it
i
o
n
p
ro
c
e
ss
,
a
lo
n
g
with
it
s
d
e
sc
rip
t
io
n
,
is
p
ro
v
id
e
d
in
t
h
is
p
a
p
e
r.
S
o
m
e
o
f
th
e
v
i
d
e
o
-
b
a
se
d
d
a
tas
e
ts
u
se
d
i
n
m
a
n
y
sc
h
o
larly
w
o
rk
s
a
re
a
lso
e
x
a
m
in
e
d
.
Re
su
lt
s
o
b
tain
e
d
fro
m
d
iffere
n
t
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
m
o
d
e
ls
a
re
p
re
se
n
ted
a
lo
n
g
wit
h
th
e
ir
p
e
rfo
rm
a
n
c
e
p
a
r
a
m
e
ters
.
An
e
x
p
e
rime
n
t
wa
s
c
a
rr
ied
o
u
t
o
n
th
e
fe
r2
0
1
3
d
a
tas
e
t
i
n
G
o
o
g
le C
o
l
a
b
fo
r
d
e
p
re
ss
io
n
d
e
tec
ti
o
n
,
wh
ich
c
a
m
e
o
u
t
to
b
e
9
7
%
a
c
c
u
ra
te o
n
th
e
train
i
n
g
se
t
a
n
d
5
7
.
4
%
a
c
c
u
ra
te o
n
t
h
e
t
e
stin
g
se
t.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
lear
n
in
g
E
m
o
tio
n
r
ec
o
g
n
itio
n
Go
o
g
le
C
o
lab
Ma
ch
in
e
lear
n
in
g
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
:
T
ed
d
y
Su
r
y
a
Gu
n
awa
n
,
Dep
ar
tm
en
t o
f
E
lectr
ical
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
in
g
,
I
n
ter
n
atio
n
al
I
s
lam
ic
Un
iv
er
s
ity
Ma
lay
s
ia,
Ma
lay
s
ia.
E
m
ail:
ts
g
u
n
awa
n
@
iiu
m
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
p
ast
f
ew
y
ea
r
s
,
em
o
tio
n
r
ec
o
g
n
itio
n
h
as
b
ec
o
m
e
o
n
e
o
f
th
e
lead
in
g
to
p
ics
in
th
e
f
ield
o
f
m
ac
h
in
e
lear
n
in
g
an
d
a
r
tific
ial
in
tellig
en
ce
.
T
h
e
tr
em
en
d
o
u
s
i
n
cr
ea
s
e
in
th
e
d
ev
elo
p
m
e
n
t
o
f
s
o
p
h
is
ticated
h
u
m
an
-
c
o
m
p
u
ter
i
n
ter
ac
tio
n
tech
n
o
lo
g
ies
h
as
f
u
r
th
er
b
o
o
s
ted
th
e
p
ac
e
o
f
p
r
o
g
r
ess
in
th
is
f
ield
.
Facial
ac
tio
n
s
co
n
v
ey
th
e
em
o
tio
n
s
,
wh
ich
,
i
n
tu
r
n
,
co
n
v
ey
a
p
er
s
o
n
’
s
p
er
s
o
n
ality
,
m
o
o
d
,
a
n
d
i
n
ten
tio
n
s
.
E
m
o
tio
n
s
u
s
u
ally
d
ep
en
d
u
p
o
n
th
e
f
ac
ial
f
ea
tu
r
es
o
f
an
in
d
iv
id
u
al
alo
n
g
with
th
e
v
o
ice.
Nev
er
th
eless
,
th
er
e
ar
e
s
o
m
e
o
th
er
f
ea
tu
r
es a
s
well,
n
am
ely
p
h
y
s
io
lo
g
ical
f
ea
tu
r
es,
s
o
cial
f
ea
tu
r
es,
p
h
y
s
ical
f
ea
tu
r
es o
f
th
e
b
o
d
y
,
a
n
d
m
an
y
m
o
r
e
.
Mo
r
e
an
d
m
o
r
e
wo
r
k
h
as
b
ee
n
d
o
n
e
to
r
ec
o
g
n
ize
em
o
tio
n
s
with
m
o
r
e
ac
cu
r
ac
y
a
n
d
p
r
ec
is
io
n
.
T
h
e
tar
g
et
o
f
em
o
tio
n
r
ec
o
g
n
itio
n
ca
n
b
e
ac
h
iev
ed
b
r
o
ad
l
y
u
s
in
g
v
is
u
al
-
b
ased
tech
n
iq
u
es
o
r
s
o
u
n
d
-
b
ased
tech
n
iq
u
es.
Ar
tific
ial
in
tellig
en
ce
h
as r
ev
o
lu
tio
n
ized
th
e
f
ield
o
f
h
u
m
an
-
co
m
p
u
ter
i
n
ter
ac
tio
n
an
d
p
r
o
v
id
es m
an
y
m
ac
h
in
e
lear
n
in
g
tec
h
n
iq
u
es
t
o
r
ea
c
h
o
u
r
aim
.
T
h
er
e
ar
e
m
an
y
m
ac
h
i
n
e
lear
n
in
g
t
ec
h
n
iq
u
es
to
r
ec
o
g
n
ize
th
e
em
o
tio
n
,
b
u
t
th
is
p
ap
e
r
will
m
o
s
tly
f
o
c
u
s
o
n
v
id
e
o
-
b
ased
e
m
o
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
in
g
.
Vi
d
eo
-
b
ased
e
m
o
tio
n
r
ec
o
g
n
itio
n
is
m
u
ltid
is
cip
lin
ar
y
an
d
in
cl
u
d
es
f
ield
s
lik
e
p
s
y
c
h
o
lo
g
y
,
af
f
ec
tiv
e
c
o
m
p
u
tin
g
,
a
n
d
h
u
m
an
-
c
o
m
p
u
ter
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
1
8
,
No
.
5
,
Octo
b
e
r
2
0
2
0
:
2
4
6
3
-
2471
2464
i
n
ter
ac
tio
n
.
T
h
e
f
u
n
d
am
en
tal
p
iece
o
f
th
e
m
ess
ag
e
is
th
e
f
ac
ial
ex
p
r
ess
io
n
,
wh
ich
estab
lis
h
es
5
5
%
o
f
th
e
g
e
n
er
al
im
p
r
ess
io
n
[
1
]
.
T
o
m
ak
e
a
well
-
f
itted
m
o
d
el
f
o
r
v
id
eo
-
b
ased
em
o
tio
n
r
ec
o
g
n
i
tio
n
,
th
er
e
m
u
s
t
b
e
p
r
o
p
er
f
ea
t
u
r
e
f
r
am
es
o
f
th
e
f
ac
ial
ex
p
r
ess
io
n
wi
th
in
th
e
s
co
p
e.
I
n
s
tead
o
f
u
s
in
g
co
n
v
en
tio
n
al
tech
n
iq
u
es,
d
ee
p
lear
n
in
g
p
r
o
v
id
es
a
v
ar
iety
in
ter
m
s
o
f
ac
cu
r
ac
y
,
lear
n
in
g
r
ate,
a
n
d
p
r
ed
ictio
n
.
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
is
o
n
e
o
f
th
e
d
ee
p
lear
n
in
g
tech
n
iq
u
es
wh
ich
h
av
e
p
r
o
v
id
ed
s
u
p
p
o
r
t
an
d
p
lat
f
o
r
m
f
o
r
a
n
aly
zi
n
g
v
is
u
al
im
ag
er
y
.
C
o
n
v
o
lu
tio
n
is
th
e
f
u
n
d
am
en
tal
u
s
e
o
f
a
f
ilter
to
a
n
in
p
u
t
th
at
o
u
tco
m
e
in
an
ac
t
u
atio
n
.
R
eh
ash
ed
u
s
e
o
f
a
s
im
ilar
f
ilter
to
an
in
p
u
t
b
r
i
n
g
s
ab
o
u
t
a
m
ap
o
f
en
ac
tm
e
n
ts
ca
lled
a
f
ea
tu
r
e
m
ap
,
s
h
o
win
g
th
e
ar
ea
s
an
d
q
u
ality
o
f
a
r
ec
o
g
n
ized
ele
m
en
t
in
in
p
u
t,
f
o
r
e
x
am
p
le,
a
p
ictu
r
e.
T
h
e
d
ev
el
o
p
m
e
n
t
o
f
co
n
v
o
lu
tio
n
n
eu
r
al
s
y
s
tem
s
is
th
e
ca
p
ac
ity
to
co
n
s
eq
u
e
n
tly
g
ain
p
r
o
f
icien
c
y
with
an
en
o
r
m
o
u
s
n
u
m
b
e
r
o
f
f
ilter
s
in
eq
u
al
ex
p
licit
to
a
tr
ain
in
g
d
ataset
u
n
d
er
th
e
r
eq
u
ir
em
en
ts
o
f
a
p
a
r
ticu
lar
p
r
escien
t
d
is
p
lay
in
g
is
s
u
e,
f
o
r
ex
am
p
le,
p
ictu
r
e
ch
ar
ac
ter
izatio
n
.
T
h
e
o
u
tco
m
e
is
p
r
o
f
o
u
n
d
ly
e
x
p
licit
h
ig
h
lig
h
ts
th
at
ca
n
b
e
d
is
tin
g
u
is
h
ed
an
y
wh
e
r
e
o
n
i
n
p
u
t
p
ictu
r
es.
Dee
p
lear
n
in
g
h
as
ac
h
iev
e
d
g
r
ea
t
s
u
cc
ess
in
r
ec
o
g
n
izi
n
g
em
o
tio
n
s
,
an
d
C
NN
is
t
h
e
well
-
k
n
o
wn
d
ee
p
lear
n
in
g
m
et
h
o
d
th
at
h
as a
ch
iev
ed
r
em
ar
k
a
b
le
p
er
f
o
r
m
an
ce
in
im
a
g
e
p
r
o
ce
s
s
in
g
.
T
h
er
e
h
as
b
ee
n
a
g
r
ea
t
d
ea
l
o
f
wo
r
k
in
v
is
u
al
p
atter
n
r
ec
o
g
n
itio
n
f
o
r
f
ac
ial
em
o
tio
n
al
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
,
ju
s
t
as
in
s
ig
n
al
p
r
o
ce
s
s
in
g
f
o
r
s
o
u
n
d
-
b
ased
r
e
co
g
n
itio
n
o
f
f
ee
lin
g
s
.
Nu
m
er
o
u
s
m
u
ltimo
d
al
ap
p
r
o
ac
h
es
ar
e
jo
in
in
g
th
ese
p
r
o
m
p
ts
[
2
]
.
Ov
er
th
e
p
ast
d
ec
a
d
es,
th
er
e
h
as
b
ee
n
ex
ten
s
iv
e
r
esear
ch
in
co
m
p
u
ter
v
is
io
n
o
n
f
ac
ial
e
x
p
r
ess
io
n
a
n
aly
s
is
[
3
]
.
T
h
e
o
b
jectiv
e
o
f
th
is
p
ap
e
r
is
to
d
ev
el
o
p
v
i
d
eo
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
i
n
g
with
Go
o
g
le
c
o
llab
.
2.
VIDE
O
-
B
AS
E
D
E
M
O
T
I
O
N
RE
CO
G
NI
T
I
O
N
US
I
NG
D
E
E
P
L
E
AR
NING
A
L
G
O
RI
T
H
M
S
I
n
th
is
s
ec
tio
n
,
d
etails
o
f
th
e
g
en
e
r
al
ar
c
h
itectu
r
e
f
o
r
b
u
ild
in
g
a
v
i
d
eo
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
m
o
d
el
u
s
in
g
a
d
ee
p
l
ea
r
n
i
n
g
a
lg
o
r
ith
m
is
d
escr
ib
e
d
.
Mo
r
eo
v
er
,
th
e
ar
c
h
itectu
r
al
d
ia
g
r
am
,
alo
n
g
with
v
ar
io
u
s
p
r
e
-
a
n
d
p
o
s
t
-
p
r
o
ce
s
s
in
g
p
r
o
c
ess
es
,
ar
e
b
r
ief
ly
d
escr
ib
ed
.
T
h
e
o
v
er
v
iew
o
f
t
h
e
s
y
s
tem
u
s
in
g
C
NN
is
s
h
o
wn
in
Fig
u
r
e
1
.
B
ef
o
r
e
C
NN
co
m
es in
to
ac
tio
n
,
t
h
e
in
p
u
t v
id
eo
h
as to
g
o
t
h
r
o
u
g
h
s
ev
er
al
p
r
o
ce
s
s
es.
Fig
u
r
e
1
.
Vid
e
o
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
2
.
1
.
P
re
-
pro
ce
s
s
ing
T
h
is
is
th
e
f
ir
s
t p
r
o
ce
s
s
th
at
is
ap
p
lied
to
th
e
in
p
u
t
v
id
eo
s
am
p
le.
E
m
o
tio
n
s
ar
e
u
s
u
ally
ca
te
g
o
r
ized
as
h
ap
p
y
,
s
ad
,
a
n
g
er
,
p
r
i
d
e,
f
ea
r
,
s
u
r
p
r
is
e,
etc.
Hen
ce
,
f
r
am
e
s
ar
e
to
b
e
e
x
tr
ac
te
d
f
r
o
m
t
h
e
in
p
u
t
v
id
eo
[
4
]
.
T
h
e
n
u
m
b
er
o
f
f
r
am
es
v
a
r
ies
f
o
r
d
if
f
e
r
en
t
r
esear
ch
e
r
s
b
ased
o
n
co
m
p
lex
ity
an
d
co
m
p
u
tatio
n
al
tim
e.
T
h
e
f
r
a
m
es
ar
e
th
en
co
n
v
er
ted
to
th
e
g
r
a
y
s
ca
le.
T
h
e
f
r
am
e
o
b
tain
ed
a
f
ter
g
r
a
y
s
ca
lin
g
is
s
o
m
ewh
at
b
lac
k
an
d
wh
ite
o
r
g
r
a
y
m
o
n
o
c
h
r
o
m
e.
T
h
e
c
o
n
tr
ast
with
lo
w
-
in
ten
s
ity
r
esu
lts
in
g
r
ey
an
d
th
at
with
s
tr
o
n
g
i
n
ten
s
ity
r
esu
lts
in
wh
ite
[
5
].
T
h
is
s
tep
is
f
o
llo
wed
b
y
th
e
h
i
s
to
g
r
am
eq
u
aliza
tio
n
o
f
th
e
f
r
a
m
es.
His
to
g
r
am
eq
u
aliza
tio
n
is
a
co
m
p
u
ter
p
ictu
r
e
h
an
d
lin
g
s
tr
ateg
y
u
s
ed
to
im
p
r
o
v
e
co
n
tr
ast
in
p
ictu
r
es.
I
t
ac
h
iev
es
th
is
b
y
v
iab
l
y
s
p
r
e
ad
in
g
o
u
t
th
e
m
o
s
t
s
u
cc
ess
iv
e
in
ten
s
ity
esteem
s
,
f
o
r
ex
am
p
le
,
lo
o
s
en
in
g
u
p
t
h
e
in
ten
s
ity
s
co
p
e
o
f
th
e
p
ictu
r
e.
A
h
is
to
g
r
am
is
a
g
r
ap
h
ical
p
o
r
tr
ay
al
o
f
th
e
in
ten
s
ity
d
is
s
em
in
atio
n
o
f
a
p
ictu
r
e.
I
n
s
tr
aig
h
tf
o
r
war
d
te
r
m
s
,
it
r
ep
r
esen
ts
th
e
n
u
m
b
er
o
f
p
ix
els f
o
r
ev
er
y
in
ten
s
ity
v
alu
e
co
n
s
id
er
ed
[
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Dev
elo
p
men
t o
f v
id
eo
-
b
a
s
ed
e
mo
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lea
r
n
in
g
w
ith
… (
Ted
d
y
S
u
r
y
a
Gu
n
a
w
a
n
)
2465
2
.
2
.
F
a
ce
det
ec
t
io
n
E
m
o
tio
n
s
ar
e
f
ea
tu
r
ed
m
ain
l
y
f
r
o
m
th
e
f
ac
e.
T
h
e
r
ef
o
r
e
,
it
is
cr
u
cial
to
d
etec
t
th
e
f
ac
e
t
o
o
b
tain
f
ac
ial
f
ea
tu
r
es
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
in
g
an
d
r
ec
o
g
n
itio
n
.
Ma
n
y
f
ac
e
d
etec
tio
n
alg
o
r
ith
m
s
a
r
e
u
s
ed
b
y
m
an
y
r
esear
ch
er
s
lik
e
Op
en
C
V,
DL
I
B
,
E
ig
en
f
ac
es,
lo
ca
l
b
in
ar
y
p
atter
n
s
h
is
to
g
r
am
s
(
L
B
PH)
,
an
d
Vio
la
-
J
o
n
es
(
VJ)
[
7
]
.
C
o
n
v
en
ti
o
n
al
alg
o
r
ith
m
s
in
clu
d
ed
f
ac
e
ac
k
n
o
wled
g
m
e
n
t
wo
r
k
b
y
d
is
tin
g
u
is
h
in
g
f
ac
ial
h
ig
h
l
ig
h
ts
b
y
ex
tr
icatin
g
h
ig
h
lig
h
ts
,
o
r
m
iles
to
n
es,
f
r
o
m
th
e
p
ictu
r
e
o
f
th
e
f
ac
e.
Fo
r
i
n
s
tan
ce
,
to
ex
tr
icate
f
ac
ial
h
ig
h
lig
h
ts
,
a
ca
lcu
latio
n
m
ay
ex
am
in
e
th
e
s
h
ap
e
an
d
s
ize
o
f
th
e
ey
es,
th
e
s
ize
o
f
t
h
e
n
o
s
e,
an
d
its
r
elativ
e
s
itu
atio
n
with
th
e
ey
es.
I
t
m
ig
h
t
lik
ewise
d
is
s
ec
t
th
e
ch
ee
k
b
o
n
es
an
d
jaw.
T
h
ese
ex
tr
a
cted
h
ig
h
lig
h
ts
wo
u
ld
th
e
n
b
e
u
tili
ze
d
f
o
r
lo
o
k
i
n
g
th
r
o
u
g
h
d
if
f
er
en
t
p
ictu
r
es
th
at
h
av
e
m
atch
in
g
f
ea
tu
r
es.
T
h
r
o
u
g
h
o
u
t
th
e
y
e
ar
s
,
th
e
i
n
d
u
s
tr
y
h
as
m
o
v
ed
to
war
d
s
d
ee
p
l
ea
r
n
in
g
.
C
NN
ha
s
b
ee
n
u
tili
ze
d
r
ec
en
tly
to
im
p
r
o
v
e
th
e
ex
ac
tn
ess
o
f
f
ac
e
ac
k
n
o
wled
g
m
en
t c
alcu
latio
n
s
.
T
h
ese
ca
lcu
latio
n
s
ac
ce
p
t
a
p
ictu
r
e
as
in
f
o
r
m
atio
n
an
d
c
o
n
ce
n
tr
ate
a
p
r
o
f
o
u
n
d
ly
in
tr
i
ca
te
ar
r
an
g
e
m
en
t
o
f
f
ea
tu
r
e
s
o
u
t
o
f
th
e
p
ict
u
r
e.
T
h
ese
in
co
r
p
o
r
ate
f
ea
tu
r
es
lik
e
t
h
e
wid
th
o
f
th
e
f
ac
e,
t
h
e
s
tatu
r
e
o
f
f
ac
e,
t
h
e
wid
th
o
f
th
e
n
o
s
e,
lip
s
,
ey
es,
p
r
o
p
o
r
ti
o
n
o
f
wid
th
s
,
s
k
in
s
h
ad
in
g
to
n
e,
an
d
s
u
r
f
ac
e.
E
s
s
en
tially
,
a
co
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
s
ep
ar
ates
an
en
o
r
m
o
u
s
n
u
m
b
er
o
f
h
ig
h
lig
h
ts
f
r
o
m
a
p
ictu
r
e.
T
h
ese
h
ig
h
lig
h
ts
ar
e
th
en
co
o
r
d
i
n
ated
with
th
e
o
n
es p
u
t a
wa
y
in
th
e
d
atab
ase.
2
.
3
.
I
ma
g
e
cr
o
pp
ing
a
nd
re
s
izing
I
n
th
is
p
h
ase,
th
e
f
ac
e
d
etec
ted
b
y
th
e
f
ac
e
d
etec
tio
n
alg
o
r
i
th
m
is
cr
o
p
p
ed
to
o
b
tain
a
b
r
o
ad
er
a
n
d
clea
r
er
lo
o
k
o
f
th
e
f
ac
ial
im
ag
e.
C
r
o
p
p
in
g
is
th
e
ex
p
u
ls
io
n
o
f
u
n
d
esira
b
le
ex
ter
n
al
r
eg
io
n
s
f
r
o
m
a
p
h
o
to
g
r
ap
h
i
c
o
r
illu
s
tr
ated
p
ictu
r
e.
T
h
e
p
r
o
c
ed
u
r
e
,
as a
r
u
le,
co
m
p
r
is
es o
f
th
e
ex
p
u
ls
io
n
o
f
a
p
o
r
tio
n
o
f
th
e
f
r
in
g
e
r
eg
i
o
n
s
o
f
a
p
ictu
r
e
to
ex
p
el
in
cid
en
tal
r
u
b
b
is
h
f
r
o
m
th
e
im
ag
e,
to
im
p
r
o
v
e
its
s
u
r
r
o
u
n
d
i
n
g
,
to
c
h
a
n
g
e
th
e
p
e
r
s
p
ec
tiv
e
p
r
o
p
o
r
tio
n
,
o
r
to
h
ig
h
lig
h
t
o
r
d
is
en
g
ag
e
th
e
to
p
ic
f
r
o
m
its
b
ac
k
g
r
o
u
n
d
.
Af
ter
p
er
f
o
r
m
i
n
g
cr
o
p
p
in
g
o
p
er
atio
n
o
n
th
e
f
r
am
es
,
th
e
s
ize
o
f
th
e
im
a
g
es
v
ar
ies.
T
h
er
ef
o
r
e
,
to
attain
u
n
if
o
r
m
ity
,
th
ese
cr
o
p
p
ed
i
m
ag
es
ar
e
s
u
b
jecte
d
to
r
esizin
g
,
s
ay
f
o
r
o
u
r
ex
am
p
le
8
0
×8
0
p
ix
els.
A
d
ig
ital
im
a
g
e
is
ju
s
t
in
f
o
r
m
atio
n
n
u
m
b
er
s
s
h
o
win
g
v
ar
ieties
o
f
r
ed
,
g
r
ee
n
,
an
d
b
lu
e
at
a
s
p
e
cif
ic
ar
ea
o
n
a
f
r
am
ewo
r
k
o
f
p
ix
els.
Mo
r
e
o
f
ten
t
h
an
n
o
t,
we
s
ee
th
ese
p
ix
els as
s
m
aller
th
an
n
o
r
m
al
s
q
u
ar
e
s
h
ap
es
s
an
d
wich
ed
to
g
et
h
er
o
n
a
PC
s
cr
ee
n
.
W
ith
a
litt
le
in
v
en
tiv
e
r
ea
s
o
n
in
g
an
d
s
o
m
e
lo
wer
-
lev
el
co
n
tr
o
l
o
f
p
ix
els
with
co
d
e,
in
an
y
ca
s
e,
w
e
ca
n
s
h
o
w
th
at
d
ata
in
a
h
o
r
d
e
o
f
way
s
.
T
h
e
s
ize
o
f
th
e
f
r
am
e
d
eter
m
in
es
its
p
r
o
ce
s
s
in
g
tim
e.
Hen
ce
,
r
esizin
g
is
v
er
y
im
p
o
r
tan
t
to
s
h
o
r
ten
th
e
p
r
o
ce
s
s
in
g
tim
e.
Mo
r
eo
v
er
,
b
etter
r
esizin
g
tech
n
iq
u
es
s
h
o
u
ld
b
e
u
s
ed
to
p
r
eser
v
e
im
ag
e
attr
ib
u
tes
af
ter
r
esizin
g
[
8
].
T
h
e
ac
cu
r
ac
y
o
f
th
e
class
if
icat
io
n
d
ep
en
d
s
o
n
wh
eth
er
th
e
f
e
atu
r
es
ar
e
well
r
ep
r
esen
tin
g
th
e
ex
p
r
ess
io
n
o
r
n
o
t
.
T
h
er
ef
o
r
e
,
th
e
o
p
tim
izatio
n
o
f
th
e
s
elec
ted
f
ea
tu
r
es will a
u
to
m
atica
lly
im
p
r
o
v
e
class
if
icati
o
n
ac
cu
r
ac
y
[
9
]
.
2
.
4
.
CNN
s
t
ruct
ure
wit
h Co
nv
Net
A
C
NN
is
a
d
ee
p
l
ea
r
n
i
n
g
al
g
o
r
ith
m
t
h
at
ca
n
tak
e
in
a
n
i
n
f
o
p
ictu
r
e,
allo
ca
te
s
ig
n
if
ican
ce
(
lear
n
ab
le
lo
ad
s
an
d
b
ias)
to
d
if
f
er
en
t
v
iewp
o
in
ts
in
th
e
p
ictu
r
e
an
d
h
av
e
th
e
o
p
tio
n
to
s
ep
ar
ate
o
n
e
f
r
o
m
th
e
o
th
er
.
T
h
e
p
r
e
-
p
r
e
p
ar
in
g
r
eq
u
i
r
ed
in
a
C
o
n
v
Net
is
a
lo
t
o
f
lo
we
r
wh
en
co
n
t
r
asted
with
o
th
e
r
al
g
o
r
ith
m
s
.
W
h
ile
in
cr
u
d
e
te
ch
n
iq
u
es
,
f
ilter
s
ar
e
h
an
d
-
d
esig
n
e
d
,
with
en
o
u
g
h
p
r
ep
ar
atio
n
,
C
o
n
v
Nets
ca
n
g
et
f
am
iliar
with
th
ese
q
u
alities
.
T
h
e
en
g
in
ee
r
in
g
o
f
a
C
o
n
v
Net
is
p
r
etty
m
u
ch
s
im
il
ar
to
th
at
o
f
n
eu
r
o
n
s
in
th
e
h
u
m
an
b
r
ain
an
d
was
en
liv
en
ed
b
y
th
e
ass
o
ciatio
n
o
f
th
e
Vis
u
al
C
o
r
tex
.
Sin
g
u
lar
n
eu
r
o
n
s
r
ea
ct
to
im
p
r
o
v
em
e
n
ts
ju
s
t
in
a
co
n
f
in
e
d
lo
ca
le
o
f
th
e
v
is
u
al
f
ield
k
n
o
wn
as
th
e
r
ec
ep
tiv
e
f
ield
.
An
ass
o
r
tm
en
t
o
f
s
u
c
h
f
ield
s
o
v
er
lap
s
t
o
co
v
er
th
e
wh
o
le
v
is
u
al
zo
n
e
[
1
0
].
C
o
n
v
Net
is
an
ar
r
an
g
em
en
t
o
f
lay
er
s
,
an
d
ea
c
h
lay
er
o
f
a
C
o
n
v
Net
ch
an
g
es o
n
e
v
o
lu
m
e
o
f
in
itiatio
n
s
to
an
o
t
h
er
t
h
r
o
u
g
h
a
d
if
f
er
en
t
iab
le
f
u
n
cti
o
n
.
T
h
er
e
ar
e
th
r
e
e
p
r
im
ar
y
k
in
d
s
o
f
la
y
er
s
to
co
n
s
tr
u
ct
C
o
n
v
Net
m
o
d
els:
co
n
v
o
lu
tio
n
al
lay
er
,
p
o
o
lin
g
lay
e
r
,
an
d
f
u
lly
-
co
n
n
ec
ted
lay
er
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
e
g
en
er
al
ar
ch
itectu
r
e
o
f
th
e
C
o
n
v
Net
c
o
n
s
is
ts
o
f
th
e
f
o
llo
win
g
[
1
1
]:
−
I
n
p
u
t
[
8
0
×
80
×
2
]
will
h
o
l
d
th
e
r
aw
p
ix
el
esti
m
atio
n
s
o
f
t
h
e
p
ictu
r
e,
r
ig
h
t
n
o
w
a
p
ictu
r
e
o
f
wid
t
h
8
0
,
h
eig
h
t 8
0
.
−
T
h
e
c
o
n
v
o
lu
tio
n
al
la
y
er
will
ev
alu
ate
th
e
y
ield
o
f
n
e
u
r
o
n
s
th
at
ar
e
ass
o
ciate
d
with
n
ea
r
b
y
lo
ca
les
in
th
e
in
f
o
,
ea
ch
p
r
o
ce
s
s
in
g
a
d
o
t
p
r
o
d
u
ct
b
etwe
en
th
eir
lo
a
d
s
an
d
a
litt
le
ar
ea
th
e
y
ar
e
ass
o
ciate
d
with
th
e
in
p
u
t
v
o
lu
m
e
.
T
h
is
m
a
y
b
r
in
g
ab
o
u
t
v
o
lu
m
e,
f
o
r
ex
am
p
le
,
[
8
0
×
80
×
1
2
]
i
f
we
ch
o
s
e
to
u
t
ilize
1
2
f
ilter
s
.
−
R
E
L
U
lay
er
will
b
e
ap
p
l
ied
f
o
r
an
elem
en
t
-
wis
e
ac
tu
atio
n
w
o
r
k
,
f
o
r
ex
a
m
p
le,
th
e
m
ax
(
0
,
x
)
th
r
esh
o
l
d
in
g
at
ze
r
o
.
T
h
is
leav
es th
e
s
ize
o
f
th
e
v
o
lu
m
e
u
n
alter
e
d
[
8
0
×
80
×
1
2
]
.
−
POOL
lay
er
will
p
lay
o
u
t
a
d
o
wn
s
am
p
lin
g
ac
tiv
ity
alo
n
g
w
ith
th
e
s
p
atial
m
ea
s
u
r
em
en
ts
,
b
r
in
g
in
g
ab
o
u
t
v
o
lu
m
e,
f
o
r
e
x
am
p
le,
[
4
0
×
40
×
1
2
]
.
−
Fu
lly
co
n
n
ec
ted
la
y
er
(
FC
)
w
ill
p
r
o
ce
s
s
th
e
class
s
co
r
es.
T
h
e
in
p
u
t
t
o
th
is
lay
er
is
all
th
e
o
u
tp
u
ts
f
r
o
m
th
e
p
r
ev
io
u
s
la
y
er
to
all
th
e
in
d
iv
id
u
al
n
e
u
r
o
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
1
8
,
No
.
5
,
Octo
b
e
r
2
0
2
0
:
2
4
6
3
-
2471
2466
Fig
u
r
e
2
.
Gen
e
r
al
Ar
ch
itectu
r
a
l stru
ctu
r
e
o
f
c
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
3.
VIDE
O
E
M
O
T
I
O
N
DA
T
A
B
ASE
S
T
h
is
is
o
n
e
o
f
th
e
cr
itical
s
tep
s
in
b
u
ild
in
g
a
m
o
d
el.
T
h
e
d
at
a
in
v
o
lv
ed
in
th
e
f
o
r
m
o
f
v
id
e
o
s
s
h
o
u
ld
b
e
ef
f
ec
tiv
e
in
ter
m
s
o
f
d
etailin
g
an
d
u
s
ab
ilit
y
.
T
h
e
r
e
a
r
e
m
an
y
d
atasets
av
ailab
le
th
at
c
an
b
e
u
s
ed
to
f
r
am
e
th
e
v
id
eo
-
b
ased
em
o
ti
o
n
r
ec
o
g
n
itio
n
m
o
d
els.
Fo
r
e
x
am
p
le,
r
ea
l
-
w
o
r
ld
af
f
e
ctiv
e
f
ac
es
d
atab
ase
(
R
AFD
B
)
[
1
2
]
,
w
h
ich
g
at
h
e
r
s
r
ea
l
-
wo
r
ld
p
ictu
r
es
f
r
o
m
a
v
ast
n
u
m
b
er
o
f
p
e
o
p
le,
h
as
b
ee
n
d
is
ch
ar
g
ed
to
en
er
g
ize
m
o
r
e
t
r
u
e
r
esear
ch
o
n
f
ac
ial
em
o
tio
n
r
ec
o
g
n
itio
n
.
R
AF
-
D
B
co
n
tain
s
ar
o
u
n
d
1
2
2
7
1
tr
ai
n
in
g
s
am
p
les
an
d
3
0
6
8
test
d
ata
d
o
wn
l
o
a
d
ed
f
r
o
m
th
e
I
n
te
r
n
et,
g
iv
i
n
g
u
n
co
n
s
tr
ain
ed
ar
ticu
latio
n
s
u
n
d
er
d
is
tin
ctiv
e
n
atu
r
al
co
n
d
itio
n
s
.
T
ab
le
1
.
Vid
eo
d
atasets
a
v
ailab
le
f
o
r
e
m
o
tio
n
r
ec
o
g
n
itio
n
D
a
t
a
b
a
s
e
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
N
u
mb
e
r
o
f
S
u
b
j
e
c
t
s
N
u
mb
e
r
o
f
i
m
a
g
e
s/
v
i
d
e
o
s
R
y
e
r
so
n
A
u
d
i
o
-
V
i
su
a
l
D
a
t
a
b
a
se
o
f
Em
o
t
i
o
n
a
l
S
p
e
e
c
h
a
n
d
S
o
n
g
(
R
A
V
D
ESS
)
C
a
l
m,
h
a
p
p
y
,
sa
d
,
a
n
g
r
y
,
f
e
a
r
f
u
l
,
su
r
p
r
i
se,
d
i
s
g
u
st
,
a
n
d
n
e
u
t
r
a
l
.
24
7
3
5
6
v
i
d
e
o
a
n
d
a
u
d
i
o
f
i
l
e
s
F
-
M
F
A
C
S
3
.
0
(
ED
U
,
P
R
O
&
X
Y
Z
v
e
r
s
i
o
n
s)
N
e
u
t
r
a
l
,
sa
d
n
e
ss,
su
r
p
r
i
s
e
,
h
a
p
p
i
n
e
ss
,
f
e
a
r
,
a
n
g
e
r
,
c
o
n
t
e
m
p
t
a
n
d
d
i
s
g
u
st
10
4
8
7
7
v
i
d
e
o
s
a
n
d
i
ma
g
e
s
seq
u
e
n
c
e
s
Jap
a
n
e
se
F
e
ma
l
e
F
a
c
i
a
l
Ex
p
r
e
ssi
o
n
s (JA
F
F
E)
n
e
u
t
r
a
l
,
sa
d
n
e
ss,
su
r
p
r
i
se
,
h
a
p
p
i
n
e
ss,
f
e
a
r
,
a
n
g
e
r
,
a
n
d
d
i
s
g
u
s
t
10
2
1
3
s
t
a
t
i
c
i
ma
g
e
s
M
M
I
D
a
t
a
b
a
se
-
43
1
2
8
0
v
i
d
e
o
s
a
n
d
o
v
e
r
2
5
0
i
m
a
g
e
s
D
I
S
F
A
-
27
4
,
8
4
5
v
i
d
e
o
f
r
a
mes
M
u
l
t
i
me
d
i
a
U
n
d
e
r
st
a
n
d
i
n
g
G
r
o
u
p
(
M
U
G
)
N
e
u
t
r
a
l
,
sa
d
n
e
ss,
su
r
p
r
i
s
e
,
h
a
p
p
i
n
e
ss
,
f
e
a
r
,
a
n
g
e
r
,
a
n
d
d
i
s
g
u
s
t
86
1
4
6
2
se
q
u
e
n
c
e
s
I
n
d
i
a
n
S
p
o
n
t
a
n
e
o
u
s
E
x
p
r
e
s
si
o
n
D
a
t
a
b
a
s
e
(
I
S
ED
)
S
a
d
n
e
ss
,
s
u
r
p
r
i
s
e
,
h
a
p
p
i
n
e
ss
,
a
n
d
d
i
s
g
u
st
50
4
2
8
v
i
d
e
o
s
R
e
a
l
-
W
o
r
l
d
A
f
f
e
c
t
i
v
e
F
a
c
e
s
D
a
t
a
b
a
s
e
(
R
A
F
D
B
)
S
u
r
p
r
i
s
e
,
f
e
a
r
f
u
l
,
d
i
sg
u
s
t
e
d
,
h
a
p
p
y
,
sa
d
,
a
n
g
r
y
,
f
e
a
r
f
u
l
l
y
su
r
p
r
i
s
e
d
,
sa
d
l
y
a
n
g
r
y
,
sad
l
y
f
e
a
r
f
u
l
,
a
n
g
r
i
l
y
d
i
s
g
u
st
e
d
,
a
n
g
r
i
l
y
su
r
p
r
i
se
d
,
s
a
d
l
y
d
i
s
g
u
s
t
e
d
,
f
e
a
r
f
u
l
l
y
d
i
s
g
u
st
e
d
,
d
i
s
g
u
s
t
e
d
l
y
s
u
r
p
r
i
se
d
,
h
a
p
p
i
l
y
su
r
p
r
i
s
e
d
,
s
a
d
l
y
s
u
r
p
r
i
s
e
d
,
f
e
a
r
f
u
l
l
y
a
n
g
r
y
,
h
a
p
p
i
l
y
d
i
s
g
u
st
e
d
.
-
2
9
6
7
2
r
e
a
l
-
w
o
r
l
d
p
i
c
t
u
r
e
s
A
c
t
e
d
F
a
c
i
a
l
E
x
p
r
e
ss
i
o
n
s
i
n
t
h
e
W
i
l
d
(
A
F
EW)
-
-
1
8
0
9
v
i
d
e
o
s
a
mp
l
e
s
A
D
F
ES
-
B
I
V
A
n
g
e
r
,
d
i
s
g
u
st
,
f
e
a
r
,
sa
d
n
e
ss,
s
u
r
p
r
i
s
e
,
h
a
p
p
i
n
e
ss,
p
r
i
d
e
,
c
o
n
t
e
m
p
t
,
e
mb
a
r
r
a
ssm
e
n
t
,
a
n
d
n
e
u
t
r
a
l
.
-
3
7
0
s
h
o
r
t
v
i
d
e
o
sam
p
l
e
s
T
h
e
o
th
e
r
d
atab
ase,
ac
ted
f
ac
i
al
ex
p
r
ess
io
n
s
in
th
e
W
ild
(
AFEW)
[
1
3
]
,
is
b
u
ilt
u
p
f
o
r
th
e
em
o
tio
n
r
ec
o
g
n
itio
n
in
th
e
wild
c
h
allen
g
e
(
E
m
o
tiW
)
.
I
t
co
m
p
r
is
es
o
f
tr
ain
in
g
s
am
p
les
(
7
7
3
)
,
v
alid
atio
n
s
am
p
le
s
(
3
8
3
)
an
d
test
(
6
5
3
)
v
id
eo
cu
ts
g
at
h
er
ed
f
r
o
m
T
V
s
h
o
ws
o
r
m
o
t
io
n
p
ictu
r
es.
ADFE
S
-
B
I
V
is
an
au
g
m
en
tatio
n
o
f
th
e
ADFE
S
d
ataset,
wh
ich
w
as
f
ir
s
t
p
r
esen
ted
b
y
Van
d
er
Sch
alk
et
al.
[
1
4
]
.
ADFE
S
is
ac
ted
b
y
1
2
No
r
th
E
u
r
o
p
ea
n
s
u
b
jects
(
f
iv
e
f
em
a
les,
s
ev
en
g
u
y
s
)
an
d
1
0
Me
d
iter
r
an
ea
n
en
ter
tain
er
s
(
f
iv
e
f
em
ales,
f
iv
e
g
u
y
s
)
co
m
m
u
n
icatin
g
th
e
s
ix
f
u
n
d
a
m
en
tal
f
ee
lin
g
s
in
ad
d
itio
n
to
th
e
th
r
ee
co
m
p
lex
f
ee
lin
g
s
o
f
co
n
tem
p
t,
p
r
id
e
,
an
d
s
h
am
e,
wh
at
i
s
m
o
r
e
to
im
p
ar
t
ial.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Dev
elo
p
men
t o
f v
id
eo
-
b
a
s
ed
e
mo
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lea
r
n
in
g
w
ith
… (
Ted
d
y
S
u
r
y
a
Gu
n
a
w
a
n
)
2467
W
in
g
en
b
ac
h
et
al.
[
1
5
]
m
ad
e
th
e
ADFE
S
-
B
I
V
d
ataset
b
y
alter
in
g
th
e
1
2
0
r
ec
o
r
d
in
g
s
p
lay
ed
b
y
th
e
1
2
No
r
th
E
u
r
o
p
ea
n
en
ter
tai
n
er
s
to
in
clu
d
e
t
h
r
ee
d
e
g
r
ee
s
o
f
f
o
r
ce
.
T
h
ey
m
ad
e
th
r
ee
n
ew
r
ec
o
r
d
in
g
s
,
s
h
o
win
g
a
s
im
ilar
f
ee
lin
g
at
th
r
ee
u
n
iq
u
e
d
eg
r
ee
s
o
f
f
o
r
ce
-
lo
w,
m
ed
i
u
m
,
an
d
h
i
g
h
-
,
f
o
r
a
s
u
m
o
f
3
6
0
r
ec
o
r
d
i
n
g
s
.
E
ac
h
tap
e
o
f
ADFE
S
-
B
I
V
b
eg
in
s
w
ith
a
n
eu
tr
al
a
r
ticu
latio
n
a
n
d
c
lo
s
u
r
es
with
th
e
m
o
s
t
elev
ate
d
ex
p
r
ess
iv
e
ca
s
in
g
.
An
o
th
er
d
ata
s
et
is
W
SEF
F
E
P
,
wh
ich
co
n
tain
s
2
1
0
h
ig
h
-
q
u
ality
p
ictu
r
es
f
r
o
m
3
0
p
e
o
p
le
[
1
6
]
.
So
m
e
o
f
th
e
well
-
k
n
o
wn
v
id
eo
d
atab
as
es
f
o
r
em
o
tio
n
r
ec
o
g
n
itio
n
a
r
e
lis
ted
in
T
ab
le
1
.
4.
RE
L
AT
E
D
R
E
SE
ARCH
W
O
RK
S
I
n
tr
o
d
u
cin
g
d
ee
p
lear
n
in
g
tech
n
iq
u
es
in
th
e
f
ield
o
f
em
o
tio
n
r
ec
o
g
n
itio
n
with
v
id
eo
s
,
im
ag
es,
v
o
ice
,
o
r
h
an
d
wr
itten
wo
r
d
s
as
in
p
u
t
h
as
ac
h
iev
ed
a
p
r
o
m
is
in
g
r
esu
l
t.
Mo
r
e
an
d
m
o
r
e
r
esear
ch
er
s
ar
e
d
ev
elo
p
in
g
th
ei
r
in
ter
ests
in
th
eir
co
n
tr
ib
u
tio
n
to
th
is
f
ield
.
Fig
u
r
e
3
s
h
o
ws
th
e
g
r
ap
h
ical
r
e
p
r
esen
tatio
n
o
f
th
e
r
is
e
in
th
e
s
ch
o
lar
ly
wo
r
k
s
in
em
o
tio
n
r
ec
o
g
n
it
io
n
u
s
in
g
d
ee
p
lear
n
in
g
[
1
7
]
.
Ma
n
y
r
esear
ch
er
s
h
av
e
im
p
lem
en
t
ed
d
if
f
er
en
t
tr
ain
i
n
g
ar
ch
itectu
r
es
o
f
C
NN
to
im
p
r
o
v
e
th
eir
r
ec
o
g
n
itio
n
ac
c
u
r
ac
i
es.
I
n
[
1
8
,
1
9
]
,
C
NN
was
tr
ain
ed
with
m
an
y
C
NN
ar
ch
itectu
r
es
o
b
tain
ed
f
r
o
m
th
e
p
r
e
tr
ain
e
d
m
o
d
el
(
I
m
ag
eNe
t)
as
s
h
o
wn
in
T
ab
le
2
.
T
h
e
r
e
s
u
lts
b
ec
o
m
e
m
o
r
e
p
r
o
m
is
in
g
with
th
e
ad
v
e
n
t
o
f
m
o
r
e
an
d
m
o
r
e
r
esear
ch
es.
T
h
e
v
id
eo
-
b
ased
e
m
o
tio
n
r
ec
o
g
n
itio
n
m
o
d
els
d
ev
elo
p
e
d
to
b
e
m
o
r
e
an
d
m
o
r
e
ac
c
u
r
ate.
T
h
e
p
e
r
ce
n
ta
g
e
o
f
ac
cu
r
ac
ies
o
f
W
in
g
en
b
ac
h
et
al
.
[
1
5
]
a
n
d
So
n
m
ez
[
1
6
]
m
o
d
el
f
o
r
d
if
f
er
e
n
t e
m
o
tio
n
s
ar
e
lis
ted
in
T
a
b
le
3.
Fig
u
r
e
3.
Yea
r
ly
s
ch
o
lar
ly
wo
r
k
s
o
n
em
o
tio
n
r
ec
o
g
n
itio
n
u
s
i
n
g
d
ee
p
lear
n
in
g
f
r
o
m
2
0
0
9
t
o
2
0
2
0
T
ab
le
2.
T
esti
n
g
ac
c
u
r
ac
ies f
o
r
C
NN
ar
ch
itectu
r
es
A
r
c
h
i
t
e
c
t
u
r
e
U
s
ed
Pre
-
Tr
a
i
n
e
d
A
c
c
u
r
a
c
y
G
o
o
g
L
e
N
e
t
W
i
t
h
I
mag
e
N
e
t
6
2
.
9
6
C
a
f
f
e
N
e
t
W
i
t
h
I
mag
e
N
e
t
6
8
.
0
5
V
G
C
1
6
W
i
t
h
I
mag
e
N
e
t
6
8
.
2
4
R
e
si
d
u
a
l
N
e
t
w
o
r
k
W
i
t
h
I
mag
e
N
e
t
6
9
.
6
5
T
ab
le
3.
C
o
m
p
ar
is
o
n
b
etwe
en
W
in
g
en
b
ac
h
et
al.
[
1
5
]
an
d
So
n
m
ez
[
1
6
]
ac
c
u
r
ac
ies
Emo
t
i
o
n
W
i
n
g
e
n
b
a
c
h
e
t
a
l
.
(
2
0
1
6
)
A
c
c
u
r
a
c
y
(
%)
S
o
n
m
e
z
(
2
0
1
8
)
A
c
c
u
r
a
c
y
(
%)
H
a
p
p
y
8
4
.
6
86
S
a
d
7
9
.
3
67
A
n
g
r
y
7
4
.
6
9
4
.
6
S
u
r
p
r
i
s
e
9
2
.
3
8
3
.
3
D
i
sg
u
st
65
9
7
.
3
F
e
a
r
6
1
.
6
61
N
e
u
t
r
a
l
89
36
P
r
i
d
e
4
2
.
3
9
1
.
6
C
o
n
t
e
m
p
t
35
50
Emb
a
r
r
a
ssm
e
n
t
6
4
.
6
89
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
1
8
,
No
.
5
,
Octo
b
e
r
2
0
2
0
:
2
4
6
3
-
2471
2468
5.
G
O
O
G
L
E
CO
L
A
B
I
M
P
L
E
M
E
N
T
A
T
I
O
N
AN
D
RE
SUL
T
S
5
.1
.
E
x
perim
ent
a
l
s
et
up
T
h
e
ex
p
er
im
e
n
t
was
ca
r
r
ied
o
n
th
e
f
er
2
0
1
3
d
ataset.
T
h
e
d
ata
c
o
n
s
is
ts
o
f
4
8
×
48
-
p
ix
el
g
r
ay
s
ca
le
im
ag
es
o
f
f
ac
es
.
T
h
is
d
ataset
was
p
r
ep
ar
ed
b
y
Pier
r
e
-
L
u
c
C
ar
r
ier
a
n
d
Aar
o
n
C
o
u
r
v
ille
as
a
p
a
r
t
o
f
th
e
Kag
g
le
C
h
allen
g
e
n
am
ed
as
C
h
allen
g
es
in
R
ep
r
esen
tatio
n
L
ea
r
n
in
g
:
Facial
E
x
p
r
ess
io
n
R
ec
o
g
n
itio
n
C
h
allen
g
e
in
2
0
1
3
.
I
t
co
n
s
is
ts
o
f
2
8
7
0
9
tr
ain
s
am
p
les
an
d
3
5
8
9
test
s
am
p
les.
I
t
in
clu
d
e
d
s
ev
e
n
em
o
tio
n
s
,
n
am
ely
An
g
r
y
,
Di
s
g
u
s
t,
Fear
,
Hap
p
y
,
Sad
,
Su
r
p
r
is
e,
Neu
tr
al
.
T
h
e
in
teg
r
ate
d
d
e
v
elo
p
m
e
n
t
e
n
v
ir
o
n
m
en
t
(
I
DE
)
u
s
ed
f
o
r
th
e
p
r
o
ce
s
s
was
Go
o
g
le
C
o
lab
o
r
ato
r
y
o
r
Go
o
g
le
C
o
lab
in
s
h
o
r
t.
Go
o
g
le
C
o
lab
is
a
f
r
ee
J
u
p
y
ter
n
o
te
b
o
o
k
e
n
v
ir
o
n
m
en
t
t
h
at
r
eq
u
i
r
e
s
n
o
s
etu
p
an
d
r
u
n
s
en
tire
ly
in
th
e
clo
u
d
.
W
ith
Go
o
g
le
C
o
lab
,
it
is
p
o
s
s
ib
le
to
wr
ite
an
d
e
x
ec
u
te
co
d
e,
s
av
e
an
d
s
h
ar
e
o
u
r
an
al
y
s
es,
an
d
ac
ce
s
s
p
o
wer
f
u
l
co
m
p
u
tin
g
r
eso
u
r
ce
s
,
all
f
o
r
f
r
ee
f
r
o
m
t
h
e
b
r
o
wser
.
T
h
e
f
er
2
0
1
3
d
ataset
was
m
o
u
n
ted
to
th
e
Go
o
g
le
C
o
lab
u
s
in
g
Go
o
g
le
Dr
iv
e
in
th
e
f
o
r
m
o
f
a
C
SV
f
ile.
Af
ter
L
o
ad
in
g
th
e
d
ataset,
th
e
b
atch
s
ize
wa
s
s
et
to
2
5
6
,
an
d
th
e
tr
ain
in
g
ep
o
ch
s
s
et
to
2
5
.
T
h
e
s
o
f
twar
e
u
s
ed
was
Py
th
o
n
3
with
m
ac
h
in
e
lear
n
in
g
lib
r
ar
ies,
in
clu
d
in
g
K
er
as
2
.
1
.
6
an
d
T
e
n
s
o
r
f
lo
w
1
.
7
.
0
.
Af
ter
in
itializin
g
th
e
tr
ain
in
g
a
n
d
th
e
test
in
g
in
s
tan
ce
s
,
th
e
d
ata
was
g
iv
en
to
th
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
(
C
NN)
,
wh
ich
co
n
s
is
ted
o
f
3
co
n
v
o
lu
ti
o
n
al
lay
er
s
an
d
o
n
e
f
u
lly
c
o
n
n
ec
t
ed
n
eu
r
al
n
etwo
r
k
.
T
h
e
m
o
d
el
tr
ain
ed
f
o
r
ab
o
u
t
4
h
o
u
r
s
.
5
.
2
.
E
x
pe
rim
ent
a
l
re
s
ults
T
h
is
ex
p
er
im
en
t
u
s
ed
2
5
ep
o
c
h
s
f
o
r
tr
ain
in
g
th
e
d
ata
s
am
p
le
s
.
W
ith
ea
ch
ep
o
ch
,
th
e
tr
ain
in
g
ac
cu
r
ac
y
in
cr
ea
s
ed
wh
ile
r
ed
u
ci
n
g
th
e
l
o
s
s
.
Per
f
o
r
m
an
ce
ev
al
u
atio
n
i
s
s
h
o
wn
in
T
ab
le
4
,
w
h
ile
th
e
co
n
f
u
s
io
n
m
atr
ix
is
s
h
o
wn
in
Fig
u
r
e
4
.
W
h
ile
s
o
m
e
em
o
tio
n
r
ec
o
g
n
itio
n
s
am
p
les
ar
e
s
h
o
wn
in
Fig
u
r
e
5
.
T
h
e
test
in
g
r
esu
lts
wer
e
m
ad
e
m
o
r
e
ac
cu
r
ate
b
y
in
clu
d
in
g
th
e
Haa
r
ca
s
ca
d
e
f
ac
e
d
etec
tio
n
p
r
o
ce
s
s
.
I
t
is
a
m
ac
h
in
e
lear
n
in
g
o
b
ject
d
etec
tio
n
alg
o
r
ith
m
u
s
ed
to
id
en
tify
o
b
jects
in
an
im
a
g
e
o
r
v
id
eo
.
I
t
d
ete
cted
th
e
f
ac
e
f
r
o
m
th
e
im
ag
e
to
r
ed
u
ce
th
e
ad
d
itio
n
al
n
o
is
e.
I
t w
o
r
k
ed
as illu
s
tr
ated
in
Fig
u
r
e
6
.
T
h
er
e
was
a
tr
em
en
d
o
u
s
in
c
r
ea
s
e
in
ef
f
icien
cy
a
f
ter
u
s
in
g
Haa
r
ca
s
ca
d
e
o
n
a
r
an
d
o
m
test
im
a
g
e,
wh
ich
is
r
ef
lecte
d
in
Fig
u
r
e
7
.
I
t
ca
n
b
e
f
o
u
n
d
th
at
b
ef
o
r
e
u
s
in
g
Haa
r
ca
s
ca
d
e,
as
s
h
o
wn
in
Fi
g
u
r
e
7
(
a)
,
th
e
f
ea
r
em
o
tio
n
is
m
o
r
e
d
o
m
in
an
t
t
h
an
h
ap
p
y
.
W
h
ile
af
ter
u
s
in
g
Haa
r
ca
s
ca
d
e,
as
s
h
o
wn
in
Fig
u
r
e
7
(
b
)
,
t
h
e
o
n
ly
em
o
tio
n
th
at
ca
n
b
e
r
ec
o
g
n
iz
ed
is
h
a
p
p
y
.
T
h
er
ef
o
r
e,
it
ca
n
b
e
co
n
clu
d
ed
th
at
th
e
Haa
r
ca
s
ca
d
e
im
p
r
o
v
es
em
o
tio
n
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
.
T
h
e
s
co
p
e
f
o
r
f
u
tu
r
e
im
p
r
o
v
e
m
en
ts
is
v
er
y
ap
p
ea
lin
g
in
t
h
is
f
ield
.
Dif
f
er
e
n
t
m
u
ltimo
d
el
d
e
ep
lear
n
in
g
tech
n
iq
u
es
ca
n
b
e
u
s
ed
al
o
n
g
with
d
if
f
er
en
t
a
r
ch
itectu
r
es
to
i
m
p
r
o
v
e
th
e
p
e
r
f
o
r
m
an
ce
p
ar
a
m
eter
s
[
2
0
-
2
7
]
.
A
p
ar
t
f
r
o
m
r
e
co
g
n
izin
g
th
e
em
o
tio
n
s
o
n
ly
,
th
er
e
ca
n
b
e
f
u
r
th
er
a
d
d
i
tio
n
o
f
in
ten
s
ity
s
ca
le.
T
h
is
m
i
g
h
t
h
elp
to
p
r
e
d
ict
th
e
in
ten
s
ity
o
f
th
e
r
ec
o
g
n
ize
d
em
o
tio
n
.
Als
o
,
m
u
lti
m
o
d
al
s
ca
n
b
e
u
s
ed
in
f
u
t
u
r
e
wo
r
k
s
;
f
o
r
ex
am
p
le,
v
id
e
o
an
d
s
p
ee
ch
ca
n
b
o
t
h
b
e
u
s
ed
t
o
d
esig
n
a
m
o
d
el
alo
n
g
with
t
h
e
u
s
e
o
f
m
u
lti
-
d
atasets
.
T
ab
le
4
.
Per
f
o
r
m
an
ce
ev
alu
ati
o
n
b
ased
o
n
tr
ain
i
n
g
an
d
test
in
g
ac
cu
r
ac
y
an
d
l
o
s
s
P
e
r
f
o
r
ma
n
c
e
P
a
r
a
m
e
t
e
r
s
P
e
r
f
o
r
ma
n
c
e
M
e
t
r
i
c
s (
%)
Tr
a
i
n
i
n
g
L
o
ss
0
.
0
9
4
8
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
9
7
.
0
7
Te
st
i
n
g
Lo
ss
2
.
6
5
7
Te
st
i
n
g
A
c
c
u
r
a
c
y
5
7
.
5
0
9
Fig
u
r
e
4
.
C
o
n
f
u
s
io
n
m
atr
i
x
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Dev
elo
p
men
t o
f v
id
eo
-
b
a
s
ed
e
mo
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lea
r
n
in
g
w
ith
… (
Ted
d
y
S
u
r
y
a
Gu
n
a
w
a
n
)
2469
(
a)
(
b
)
(
c)
Fig
u
r
e
5
.
Sam
p
le
o
f
r
ec
o
g
n
ize
d
em
o
tio
n
ac
cu
r
ac
y
p
er
ce
n
tag
e
g
r
ap
h
;
(
a)
s
u
r
p
r
is
e
em
o
tio
n
,
(
b
)
n
e
u
tr
al
em
o
tio
n
an
d
(
c)
s
a
d
em
o
tio
n
Fig
u
r
e
6
.
Haar
c
ascad
e
f
ac
e
d
e
tectio
n
p
r
o
ce
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
1
8
,
No
.
5
,
Octo
b
e
r
2
0
2
0
:
2
4
6
3
-
2471
2470
(
a)
(
b
)
Fig
u
r
e
7
.
Acc
u
r
ac
y
im
p
r
o
v
e
m
en
t u
s
in
g
Haa
r
c
ascad
e
;
(
a
)
b
e
f
o
r
e
Haa
r
ca
s
ca
d
e
a
n
d
(
b
)
af
ter
Haa
r
ca
s
ca
d
e
6.
CO
NCLU
SI
O
N
S
T
h
is
p
ap
er
p
r
esen
te
d
th
e
d
e
v
e
lo
p
m
en
t
o
f
v
id
eo
-
b
ased
e
m
o
ti
o
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lea
r
n
in
g
with
Go
o
g
le
C
o
lab
.
T
h
e
s
u
cc
ess
o
f
th
is
ap
p
r
o
ac
h
in
r
ec
o
g
n
izin
g
em
o
tio
n
s
h
as
b
ee
n
tr
em
e
n
d
o
u
s
ly
im
p
r
o
v
in
g
o
v
e
r
tim
e
.
I
n
tr
o
d
u
ci
n
g
d
ee
p
lear
n
i
n
g
tech
n
iq
u
es
lik
e
C
NN,
DN
N,
o
r
o
th
er
m
u
ltimo
d
a
l
m
eth
o
d
s
h
as
also
b
o
o
s
te
d
th
e
p
ac
e
o
f
r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
.
Ou
r
wo
r
k
d
em
o
n
s
tr
ated
th
e
g
en
e
r
al
ar
ch
itectu
r
al
m
o
d
el
f
o
r
b
u
ild
i
n
g
a
r
ec
o
g
n
itio
n
s
y
s
tem
u
s
in
g
d
ee
p
lear
n
in
g
(
m
o
r
e
p
r
ec
i
s
ely
C
N
N)
.
T
h
e
aim
was
to
an
aly
ze
p
r
e
an
d
p
o
s
t
p
r
o
ce
s
s
es
in
v
o
lv
ed
in
t
h
e
m
eth
o
d
o
lo
g
y
o
f
th
e
m
o
d
el.
T
h
er
e
is
ex
ten
s
iv
e
wo
r
k
d
o
n
e
o
n
im
ag
e,
s
p
ee
c
h
,
o
r
v
id
e
o
as
in
p
u
t
to
r
ec
o
g
n
ize
th
e
e
m
o
tio
n
.
T
h
is
p
ap
er
also
co
v
er
e
d
th
e
d
atasets
av
ailab
le
f
o
r
t
h
e
r
esear
ch
e
r
s
to
co
n
tr
ib
u
te
to
t
h
is
f
ield
.
Dif
f
er
e
n
t
p
e
r
f
o
r
m
an
ce
p
ar
am
eter
s
wer
e
b
en
ch
m
ar
k
ed
on
d
if
f
er
en
t
r
esear
ch
es
t
o
s
h
o
w
th
e
p
r
o
g
r
ess
in
th
is
s
p
h
er
e.
T
h
e
e
xp
e
r
im
en
tal
o
b
s
e
r
v
atio
n
was
ca
r
r
ied
o
u
t
o
n
th
e
f
er
2
0
1
3
d
ataset
in
v
o
lv
in
g
s
ev
e
n
em
o
tio
n
s
,
n
am
ely
an
g
r
y
,
d
is
g
u
s
t,
f
ea
r
,
h
ap
p
y
,
s
ad
,
s
u
r
p
r
is
e,
n
e
u
tr
al
,
wh
ic
h
y
iel
d
ed
in
th
e
b
e
9
7
%
ac
c
u
r
ac
y
o
n
th
e
tr
ai
n
in
g
s
et
an
d
5
7
.
4
%
ac
cu
r
ac
y
o
n
th
e
test
in
g
s
et
wh
en
Haa
r
ca
s
ca
d
e
tech
n
iq
u
e
is
ap
p
lied
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
wo
u
ld
lik
e
to
e
x
p
r
e
s
s
th
eir
g
r
atitu
d
e
t
o
th
e
Ma
lay
s
ian
Min
is
tr
y
o
f
E
d
u
ca
tio
n
(
M
OE
)
,
wh
ich
ha
s
p
r
o
v
i
d
ed
r
esear
c
h
f
u
n
d
in
g
th
r
o
u
g
h
th
e
Fu
n
d
am
en
tal
R
esear
ch
Gr
an
t,
FR
GS1
9
-
0
7
6
-
0
6
8
4
.
T
h
e
au
t
h
o
r
s
wo
u
ld
also
lik
e
to
th
an
k
I
n
ter
n
atio
n
al
I
s
lam
ic
Un
iv
er
s
ity
Ma
lay
s
ia
(
I
I
UM
)
,
Un
iv
er
s
iti
T
ek
n
o
lo
g
i
MA
R
A
(
UiT
M)
a
nd
Un
iv
e
r
s
itas
Po
ten
s
i U
tam
a
f
o
r
p
r
o
v
id
in
g
f
ac
ilit
i
es to
s
u
p
p
o
r
t th
e
r
esear
ch
wo
r
k
.
RE
F
E
R
E
NC
E
S
[1
]
C.
H.
Wu
,
J.
C.
L
in
a
n
d
W.
L.
W
e
i,
“
S
u
rv
e
y
o
n
a
u
d
io
v
isu
a
l
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
:
d
a
tab
a
se
s,
fe
a
tu
re
s,
a
n
d
d
a
ta f
u
sio
n
stra
teg
ies
,
”
AP
S
IPA
T
ra
n
sa
c
ti
o
n
s
o
n
S
i
g
n
a
l
a
n
d
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
,
v
o
l
.
3
,
2
0
1
4
.
[2
]
Z.
Zen
g
,
M
.
P
a
n
ti
c
,
G
.
I.
Ro
ism
a
n
,
a
n
d
T.
S
.
Hu
a
n
g
,
"
A
S
u
rv
e
y
o
f
Affe
c
t
Re
c
o
g
n
it
io
n
M
e
t
h
o
d
s:
Au
d
io
,
Vis
u
a
l,
a
n
d
S
p
o
n
tan
e
o
u
s
E
x
p
re
ss
io
n
s,"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
r
n
A
n
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
tel
li
g
e
n
c
e
,
v
o
l.
3
1
,
n
o
.
1
,
p
p
.
3
9
-
5
8
,
2
0
0
8.
[3
]
P.
C
.
Va
sa
n
th
a
n
d
K.
R.
Na
tara
j
,
“
F
a
c
ial
E
x
p
re
ss
io
n
Re
c
o
g
n
it
i
o
n
u
sin
g
S
V
M
Clas
sifier,”
I
n
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
In
f
o
r
ma
ti
c
s (IJEEI)
,
v
o
l.
3
,
n
o
.
1
,
p
p
.
1
6
-
2
0
,
2
0
1
5
.
[
4
]
W
.
H
.
A
b
d
u
l
s
a
l
a
m
,
R
.
S
.
A
l
h
a
m
d
a
n
i
,
a
n
d
M
.
N
.
A
b
d
u
l
l
a
h
,
“
F
a
c
i
a
l
E
m
o
t
i
o
n
R
e
c
o
g
n
i
t
i
o
n
f
r
o
m
V
i
d
e
o
s
U
s
i
n
g
D
e
e
p
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
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
a
n
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
9
,
n
o
.
1
,
p
p
.
14
-
1
9
,
2
0
1
9
.
[5
]
S
.
Jo
h
n
so
n
,
“
S
tep
h
e
n
Jo
h
n
so
n
o
n
d
ig
it
a
l
p
h
o
t
o
g
ra
p
h
y
,
”
O’Re
il
ly
,
2
0
0
6
.
[6
]
Y.
C.
Hu
m
,
K.
W.
Lai,
a
n
d
M
.
I
.
M
.
S
a
li
m
,
“
M
u
lt
io
b
jec
ti
v
e
s
b
ih
ist
o
g
ra
m
e
q
u
a
li
z
a
ti
o
n
fo
r
ima
g
e
c
o
n
tras
t
e
n
h
a
n
c
e
m
e
n
t,
”
Co
mp
lex
it
y
,
v
o
l.
20
,
n
o
.
2
,
p
p
.
2
2
-
3
6
,
2
0
1
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
Dev
elo
p
men
t o
f v
id
eo
-
b
a
s
ed
e
mo
tio
n
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lea
r
n
in
g
w
ith
… (
Ted
d
y
S
u
r
y
a
Gu
n
a
w
a
n
)
2471
[7
]
S
.
Al
-
S
u
m
a
id
a
e
e
,
S
.
S
.
Dla
y
,
W
.
L.
W
o
o
,
a
n
d
J.
A.
Ch
a
m
b
e
rs
,
“
F
a
c
ial
e
x
p
re
ss
io
n
re
c
o
g
n
i
ti
o
n
u
si
n
g
l
o
c
a
l
G
a
b
o
r
g
ra
d
ien
t
c
o
d
e
-
h
o
rizo
n
tal
d
ia
g
o
n
a
l
d
e
sc
rip
to
r
,
”
Pro
c
e
e
d
in
g
s
o
f
th
e
2
n
d
IET
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
In
tell
ig
e
n
t
S
ig
n
a
l
Pro
c
e
ss
in
g
2
0
1
5
(IS
P)
,
2
0
1
5
.
[8
]
D.
Zh
o
u
,
X.
S
h
e
n
,
W.
Do
n
g
,
“
Im
a
g
e
z
o
o
m
in
g
u
sin
g
d
irec
ti
o
n
a
l
c
u
b
ic
c
o
n
v
o
l
u
ti
o
n
i
n
ter
p
o
lat
io
n
,
”
IET
Ima
g
e
Pro
c
e
ss
in
g
,
v
o
l
.
6
,
n
o
.
6
,
p
p
.
6
2
7
-
6
3
4
,
2
0
1
2
.
[9
]
H
.
Bo
u
b
e
n
n
a
,
D.
Lee
,
“
Im
a
g
e
-
b
a
se
d
e
m
o
ti
o
n
re
c
o
g
n
i
ti
o
n
u
sin
g
e
v
o
lu
ti
o
n
a
ry
a
lg
o
rit
h
m
s,”
Bi
o
l
o
g
ica
ll
y
I
n
sp
ire
d
Co
g
n
it
ive
Arc
h
it
e
c
tu
re
s
,
v
o
l.
24
,
p
p
.
7
0
-
7
6
,
2
0
1
8
.
[1
0
]
I.
Wallac
h
,
M.
Dz
a
m
b
a
,
a
n
d
A
.
He
ifets
,
“
Ato
m
Ne
t:
A
D
e
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
fo
r
Bio
a
c
ti
v
it
y
P
re
d
icti
o
n
in
S
tr
u
c
tu
re
-
b
a
se
d
Dr
u
g
Disc
o
v
e
r
y
,
”
a
rXiv
,
2
0
1
5
.
[
1
1
]
C
S
2
3
1
n
,
“
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
f
o
r
V
i
s
u
a
l
R
e
c
o
g
n
i
t
i
o
n
,
”
2016.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
c
s
2
3
1
n
.
g
i
t
h
u
b
.
i
o
/
c
o
n
v
o
l
u
t
i
o
n
a
l
-
n
e
t
w
o
r
k
s
/
.
[1
2
]
S
.
Li, W
.
De
n
g
,
a
n
d
J
.
Du
,
“
Re
li
a
b
le cro
wd
so
u
rc
in
g
a
n
d
d
e
e
p
l
o
c
a
li
ty
-
p
re
se
rv
i
n
g
lea
rn
i
n
g
f
o
r
e
x
p
re
s
sio
n
re
c
o
g
n
it
i
o
n
in
t
h
e
wild
,
”
Pro
c
e
e
d
in
g
s o
f
th
e
I
EE
E
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isi
o
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
i
ti
o
n
,
p
p
.
2
584
-
2
5
9
3
,
2
0
1
7
.
[1
3
]
A
.
Dh
a
ll
,
R
.
G
o
e
c
k
e
,
S
.
L
u
c
e
y
,
T
.
G
e
d
e
o
n
,
“
Co
ll
e
c
ti
n
g
larg
e
,
r
ich
ly
a
n
n
o
tate
d
fa
c
ial
-
e
x
p
re
ss
io
n
d
a
tab
a
se
s
fro
m
m
o
v
ies
,
”
IEE
E
M
u
lt
ime
d
i
a
,
v
o
l.
3
,
p
p
.
34
-
4
1
,
2
0
1
2
.
[1
4
]
J.
Va
n
De
r
S
c
h
a
lk
,
S
.
T.
Ha
wk
,
A.H.
F
isc
h
e
r,
a
n
d
B.
Do
o
se
je
,
“
M
o
v
i
n
g
fa
c
e
s,
lo
o
k
i
n
g
p
lac
e
s:
Va
li
d
a
ti
o
n
o
f
t
h
e
Am
ste
r
d
a
m
d
y
n
a
m
ic fac
ial
e
x
p
re
ss
io
n
se
t
(AD
F
ES
)
,
”
Emo
ti
o
n
,
v
o
l
.
11
,
n
o
.
4
,
p
p
.
9
0
7
-
9
2
0
,
2
0
1
1
.
[1
5
]
T.
S
.
Wi
n
g
e
n
b
a
c
h
,
C.
As
h
win
,
a
n
d
M
.
Br
o
sn
a
n
,
“
Va
li
d
a
ti
o
n
o
f
th
e
A
m
ste
rd
a
m
d
y
n
a
m
ic
fa
c
ial
e
x
p
re
ss
io
n
se
t
–
b
a
t
h
in
ten
sit
y
v
a
riatio
n
s
(AD
F
E
S
-
BIV):
A
se
t
o
f
v
i
d
e
o
s
e
x
p
re
ss
in
g
lo
w,
in
term
e
d
iate
,
a
n
d
h
ig
h
i
n
ten
sit
y
e
m
o
ti
o
n
s,
”
Pl
o
S
On
e
,
v
o
l.
11
,
n
o
.
1
,
p
p
.
1
-
1
,
2
0
1
6
.
[1
6
]
M
.
Olsz
a
n
o
ws
k
i
,
G
.
P
o
c
h
wa
tk
o
,
K.
Ku
k
li
n
sk
i,
M
.
S
c
ib
o
r
-
Ry
ls
k
i
,
P
.
Lew
in
sk
i
,
a
n
d
R.
K.
O
h
m
e
,
“
Wars
a
w
s
e
t
o
f
e
m
o
ti
o
n
a
l
fa
c
ial
e
x
p
re
ss
io
n
p
ict
u
re
s:
a
v
a
l
id
a
ti
o
n
stu
d
y
o
f
fa
c
ial
d
is
p
lay
p
h
o
t
o
g
ra
p
h
s,
”
Fr
o
n
ti
e
rs
in
Psy
c
h
o
l
o
g
y
,
v
o
l.
5
,
2
0
1
5
.
[1
7
]
“
S
e
a
rc
h
Article
s
th
e
Len
s
-
F
re
e
&
Op
e
n
P
a
ten
t
a
n
d
S
c
h
o
larl
y
S
e
a
rc
h
,
”
Th
e
Len
s
-
F
re
e
&
Op
e
n
P
a
t
e
n
t
a
n
d
S
c
h
o
larly
S
e
a
rc
h
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s
:/
/www
.
len
s.o
r
g
[Ac
c
e
ss
e
d
:
1
5
-
F
e
b
-
2
0
2
0
]
[1
8
]
Y.
Ca
i,
W.
Z
h
e
n
g
,
T.
Z
h
a
n
g
,
Q.
Li
,
Z.
C
u
i,
a
n
d
J.
Ye
,
“
Vid
e
o
Ba
se
d
Emo
ti
o
n
Re
c
o
g
n
it
i
o
n
U
sin
g
C
NN
a
n
d
BRNN
,
”
Ch
in
e
se
Co
n
fer
e
n
c
e
o
n
Pa
tt
e
rn
R
e
c
o
g
n
it
i
o
n
,
p
p
.
6
7
9
-
6
9
1
,
2
0
1
6
.
[1
9
]
Y.
F
a
n
,
X.
Lu
,
D.
L
i,
a
n
d
Y.
Li
u
,
“
Vid
e
o
-
b
a
se
d
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
u
si
n
g
CNN
-
RNN
a
n
d
C3
D
h
y
b
rid
n
e
two
rk
s,”
Pro
c
e
e
d
in
g
s
o
f
t
h
e
1
8
t
h
ACM
In
t
e
rn
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
u
lt
im
o
d
a
l
In
ter
a
c
ti
o
n
,
p
p
.
4
4
5
-
4
5
0
,
2
0
1
6
.
[2
0
]
T.
S.
G
u
n
a
wa
n
,
M.
F.
Alg
h
ifari,
M.
A.
M
o
rsh
i
d
i,
a
n
d
M
.
Ka
rti
wi,
“
A
re
v
iew
o
n
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
a
lg
o
rit
h
m
s
u
si
n
g
sp
e
e
c
h
a
n
a
ly
sis
,”
In
d
o
n
e
sia
n
J
o
u
r
n
a
l
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
In
fo
rm
a
t
ics
(IJ
EE
I)
,
v
o
l
.
6
,
n
o
.
1
,
p
p
.
12
-
20
,
2
0
1
8
.
[2
1
]
M.F.
Al
g
h
i
fa
ri,
T.
S
.
G
u
n
a
wa
n
,
a
n
d
M
.
Ka
rti
wi,
“
S
p
e
e
c
h
e
m
o
t
io
n
re
c
o
g
n
i
ti
o
n
u
si
n
g
d
e
e
p
fe
e
d
f
o
rwa
rd
n
e
u
ra
l
n
e
tw
o
rk
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
En
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l.
10
,
n
o
.
2
,
p
p
.
5
5
4
-
5
6
1
,
2
0
1
8
.
[2
2
]
S.
A.
A.
Qa
d
ri,
T.
S.
G
u
n
a
wa
n
,
M.
F.
Alg
h
ifari,
H.
M
a
n
so
r,
M
.
Ka
rti
wi,
a
n
d
Z.
Ja
n
in
,
“
A
c
rit
ica
l
in
sig
h
t
i
n
to
m
u
lt
i
-
lan
g
u
a
g
e
s
sp
e
e
c
h
e
m
o
ti
o
n
d
a
tab
a
se
s
,”
Bu
ll
e
ti
n
o
f
El
e
c
trica
l
En
g
i
n
e
e
rin
g
a
n
d
In
f
o
rm
a
ti
c
s
,
v
o
l
.
8
,
n
o
.
4
,
p
p
.
1
3
1
2
-
1
3
2
3
,
2
0
1
9
.
[2
3
]
M.
F.
Al
g
h
ifari,
T.
S.
G
u
n
a
wa
n
,
S.
A.
A.
Qa
d
ri,
M
.
Ka
rti
wi,
a
n
d
Z.
Ja
n
in
,
“
On
th
e
u
se
o
f
v
o
ice
a
c
ti
v
it
y
d
e
tec
ti
o
n
i
n
sp
e
e
c
h
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
,”
B
u
l
letin
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
In
f
o
rm
a
ti
c
s
,
v
o
l.
8
,
n
o
.
4
,
p
p
.
1
3
2
4
-
1
3
3
2
,
2
0
1
9
.
[2
4
]
T.
S.
G
u
n
a
wa
n
,
M
.
Ka
rti
wi,
N.
A.
M
a
li
k
,
a
n
d
N.
Ism
a
il
,
“
F
o
o
d
I
n
tak
e
Ca
lo
rie
P
re
d
ictio
n
u
si
n
g
G
e
n
e
ra
li
z
e
d
Re
g
re
ss
io
n
Ne
u
ra
l
Ne
two
rk
,”
2
0
1
8
IE
EE
5
t
h
In
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
o
n
S
ma
rt I
n
stru
me
n
ta
ti
o
n
,
M
e
a
su
re
me
n
t
a
n
d
A
p
p
l
ica
ti
o
n
(ICS
IM
A)
,
p
p
.
1
-
4
,
2
0
1
8
.
[2
5
]
T.
S
.
G
u
n
a
wa
n
,
a
n
d
M
.
Ka
rti
wi,
“
On
th
e
Us
e
o
f
Ed
g
e
F
e
a
tu
re
s
a
n
d
E
x
p
o
n
e
n
ti
a
l
De
c
a
y
in
g
N
u
m
b
e
r
o
f
No
d
e
s
i
n
t
h
e
Hid
d
e
n
Lay
e
rs
fo
r
Ha
n
d
writt
e
n
S
i
g
n
a
tu
re
Re
c
o
g
n
it
io
n
,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
E
n
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l.
12
,
n
o
.
2
,
p
p
.
7
2
2
-
7
2
8
,
2
0
1
8
.
[2
6
]
E.
Ih
sa
n
to
,
K.
Ra
m
li
,
D.
S
u
d
ian
a
,
a
n
d
T
.
S
.
G
u
n
a
wa
n
,
“
An
Eff
icie
n
t
Alg
o
rit
h
m
f
o
r
Ca
rd
iac
Arr
h
y
t
h
m
i
a
Clas
sifica
ti
o
n
Us
in
g
En
se
m
b
le
o
f
De
p
t
h
wise
S
e
p
a
ra
b
le
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
s
,
”
M
DPI
Ap
p
li
e
d
S
c
ien
c
e
s
,
v
o
l.
10
,
n
o
.
2,
p
p
.
1
-
16,
2
0
2
0
.
[
2
7
]
E
.
I
h
s
a
n
t
o
,
K
.
R
a
m
l
i
,
D
.
S
u
d
i
a
n
a
,
a
n
d
T
.
S
.
G
u
n
a
w
a
n
,
“
F
a
s
t
a
n
d
A
c
c
u
r
a
t
e
A
l
g
o
r
i
t
h
m
f
o
r
E
C
G
A
u
t
h
e
n
t
i
c
a
t
i
o
n
u
s
i
n
g
R
e
s
i
d
u
a
l
D
e
p
t
h
w
i
s
e
S
e
p
a
r
a
b
l
e
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
,
”
M
D
P
I
A
p
p
l
i
e
d
S
c
i
e
n
c
e
,
v
o
l
.
10
,
n
o
.
9
,
p
p
.
1
-
1
5
,
2020.
Evaluation Warning : The document was created with Spire.PDF for Python.