I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
,
p
p
.
4
0
8
1
~
4
0
9
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
0
8
1
-
4
0
9
5
4081
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Tem
po
ra
l conv
o
lutiona
l net
wo
rk f
o
r Indo
nesia
n word by
word
sig
n lang
ua
g
e t
ra
nsla
tor o
n Media
Pipe
framewo
rk
Sa
ra
h P
urna
m
a
wa
t
i
1
,
Albert
1
,
Ang
g
i Y
o
ha
nes
P
a
rdede
1
,
I
qb
a
l F
a
k
hriza
1
,
F
a
rha
d Na
di
2
,
M
o
ha
m
m
a
d F
a
dly
Sy
a
hp
utr
a
1
,
Ro
m
i F
a
dil
la
h Ra
hm
a
t
1
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
I
n
f
o
r
mat
i
o
n
T
e
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
si
t
a
s
S
u
ma
t
e
r
a
U
t
a
r
a
,
M
e
d
a
n
,
I
n
d
o
n
e
s
i
a
2
F
a
c
u
l
t
y
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
F
r
o
n
t
i
e
r
Te
c
h
n
o
l
o
g
i
e
s
,
U
N
I
TA
R
I
n
t
e
r
n
a
t
i
o
n
a
l
U
n
i
v
e
r
si
t
y
,
P
e
t
a
l
i
n
g
Ja
y
a
,
M
a
l
a
y
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
2
,
2
0
2
5
R
ev
is
ed
Au
g
4
,
2
0
2
6
Acc
ep
ted
Au
g
1
2
,
2
0
2
6
S
ig
n
lan
g
u
a
g
e
is
a
la
n
g
u
a
g
e
c
re
a
ted
to
h
e
lp
th
e
d
e
a
f
-
m
u
te
c
o
m
m
u
n
ica
te
with
o
th
e
rs.
I
n
I
n
d
o
n
e
sia
,
t
h
e
re
a
re
two
sig
n
lan
g
u
a
g
e
s:
In
d
o
n
e
sia
n
sig
n
lan
g
u
a
g
e
sy
ste
m
o
r
siste
m
isy
a
ra
t
b
a
h
a
s
a
In
d
o
n
e
sia
(
S
IBI)
a
n
d
I
n
d
o
n
e
sia
n
sig
n
lan
g
u
a
g
e
o
r
b
a
h
a
s
a
isy
a
ra
t
In
d
o
n
e
sia
(BIS
IND
O)
,
b
u
t
th
e
d
e
a
f
c
o
m
m
u
n
it
y
p
r
e
f
e
r
s
t
o
u
s
e
B
I
S
I
ND
O
.
De
a
f
-
m
u
t
e
p
e
o
p
l
e
u
se
s
i
g
n
l
a
n
g
u
a
g
e
t
o
c
o
m
m
u
n
i
c
a
te
,
b
u
t
n
o
rm
a
l
p
e
o
p
le
u
su
a
ll
y
a
v
o
i
d
l
e
a
rn
in
g
it
b
e
c
a
u
se
th
e
y
fe
e
l
it
'
s u
n
n
e
c
e
ss
a
ry
fo
r
d
a
il
y
c
o
m
m
u
n
ica
ti
o
n
.
Th
is
wi
ll
c
e
rtain
ly
b
e
a
p
ro
b
lem
fo
r
th
e
d
e
a
f
-
m
u
te
b
e
c
a
u
se
p
e
o
p
le
wh
o
d
o
n
o
t
lea
rn
sig
n
lan
g
u
a
g
e
will
c
e
rtain
ly
h
a
v
e
d
iffi
c
u
lt
y
u
n
d
e
rsta
n
d
i
n
g
wh
a
t
t
h
e
y
a
re
try
in
g
to
c
o
n
v
e
y
.
T
h
e
re
fo
re
,
a
n
a
p
p
li
c
a
ti
o
n
t
h
a
t
c
a
n
tran
sla
te
sig
n
lan
g
u
a
g
e
is
n
e
e
d
e
d
to
h
e
lp
o
v
e
rc
o
m
e
c
o
m
m
u
n
ica
ti
o
n
p
ro
b
lem
s
b
e
twe
e
n
th
e
d
e
a
f
-
m
u
te
a
n
d
n
o
rm
a
l
p
e
o
p
le.
T
h
is
re
se
a
rc
h
fo
c
u
se
s
o
n
BI
S
IND
O
tran
sla
ti
o
n
u
sin
g
a
tem
p
o
ra
l
c
o
n
v
o
lu
ti
o
n
a
l
n
e
two
rk
(TCN)
with
M
e
d
iaPi
p
e
m
e
th
o
d
s.
Th
e
a
m
o
u
n
t
o
f
d
a
ta
u
se
d
c
o
n
sists
o
f
2
,
7
0
0
v
i
d
e
o
s
re
p
re
se
n
ti
n
g
2
7
w
o
rd
-
lev
e
l
sig
n
s
in
BI
S
IND
O;
th
is
n
u
m
b
e
r
i
n
c
re
a
se
s
to
3
2
,
4
0
0
a
fter
u
n
d
e
rg
o
in
g
t
h
e
d
a
ta
a
u
g
m
e
n
tatio
n
p
ro
c
e
ss
.
T
h
e
m
o
d
e
l
b
u
il
t
wa
s
tes
ted
in
tra
n
sla
ti
n
g
BIS
IND
O
a
t
th
e
wo
r
d
lev
e
l
in
sim
u
lati
o
n
,
a
n
d
a
th
o
r
o
u
g
h
c
o
m
p
a
riso
n
b
e
twe
e
n
TCN
a
n
d
b
i
d
irec
ti
o
n
a
l
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(
BiL
S
TM
)
;
b
o
t
h
o
f
t
h
e
m
we
re
ru
n
n
in
g
o
n
th
e
M
e
d
iaP
ip
e
p
latfo
rm
se
p
a
ra
tely
.
F
r
o
m
t
h
e
e
x
p
e
rime
n
t,
it
is
s
h
o
wn
t
h
a
t
t
h
e
p
e
rfo
rm
a
n
c
e
o
f
TCN
e
x
c
e
e
d
s BiL
S
TM
wit
h
a
n
a
c
c
u
ra
c
y
ra
te as
h
i
g
h
a
s
8
8
.
6
4
%
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Han
d
m
o
v
e
m
en
t r
ec
o
g
n
itio
n
L
an
g
u
ag
e
au
to
m
atio
n
Neu
r
al
n
etwo
r
k
Sp
ee
ch
-
im
p
air
ed
tech
n
o
l
o
g
y
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
:
Sar
ah
Pu
r
n
am
awa
ti
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
,
Facu
lty
o
f
C
o
m
p
u
te
r
Scien
ce
an
d
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
Un
iv
er
s
itas
Su
m
ater
a
Utar
a
Me
d
an
,
I
n
d
o
n
esia
E
m
ail:
s
ar
ah
_
p
u
r
n
am
awa
ti@
u
s
u
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
As
s
o
cial
b
ein
g
s
,
it
is
h
u
m
an
n
atu
r
e
to
s
o
cialize
an
d
in
ter
ac
t
with
o
th
er
h
u
m
a
n
s
th
r
o
u
g
h
co
m
m
u
n
icatio
n
to
o
b
tain
th
e
in
f
o
r
m
atio
n
n
ee
d
e
d
to
ad
a
p
t
to
th
e
s
u
r
r
o
u
n
d
in
g
en
v
ir
o
n
m
en
t.
Ho
wev
er
,
th
e
co
m
m
u
n
icatio
n
p
r
o
ce
s
s
ca
n
b
e
d
is
r
u
p
ted
d
u
e
to
f
ac
to
r
s
s
u
ch
as
h
ea
r
in
g
lo
s
s
o
r
lim
ite
d
s
p
ea
k
in
g
a
b
ilit
y
s
u
f
f
er
ed
b
y
o
n
e
o
f
th
e
p
ar
ties
in
a
co
m
m
u
n
icatio
n
.
Du
e
to
th
is
d
if
f
icu
lty
,
a
lan
g
u
ag
e
was
cr
ea
ted
to
h
elp
th
e
co
m
m
u
n
icatio
n
p
r
o
ce
s
s
,
n
am
e
ly
s
ig
n
lan
g
u
ag
e.
Sig
n
lan
g
u
a
g
e
is
a
f
o
r
m
o
f
h
u
m
a
n
n
o
n
-
v
e
r
b
al
co
m
m
u
n
icatio
n
ex
p
r
ess
ed
th
r
o
u
g
h
h
a
n
d
g
estu
r
es
to
v
is
u
ally
co
n
v
ey
i
n
f
o
r
m
atio
n
to
r
ec
ip
ien
ts
[
1
]
.
Sig
n
lan
g
u
ag
e
u
s
u
all
y
em
er
g
es
n
atu
r
ally
an
d
ad
ap
t
s
to
th
e
cu
ltu
r
e
o
f
ea
ch
r
eg
i
o
n
;
th
is
m
ak
es
ea
ch
r
eg
io
n
h
av
e
its
o
wn
s
ig
n
lan
g
u
ag
e.
Fo
r
ex
am
p
le,
Am
e
r
ica
u
s
es
Am
er
ican
s
ig
n
lan
g
u
ag
e
(
ASL)
[
2
]
,
t
h
e
Un
ited
Kin
g
d
o
m
u
s
es
B
r
itis
h
s
ig
n
lan
g
u
ag
e
(
B
SL)
[
3
]
,
an
d
I
n
d
ia
u
s
es
I
n
d
ian
s
ig
n
lan
g
u
ag
e
(
I
SL)
[
4
]
.
I
n
d
o
n
esia
its
elf
h
as
two
s
ig
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
0
8
1
-
4
0
9
5
4082
lan
g
u
ag
es
in
u
s
e,
n
am
ely
I
n
d
o
n
esian
s
ig
n
lan
g
u
ag
e
s
y
s
tem
o
r
s
is
tem
is
ya
r
a
t
b
a
h
a
s
a
I
n
d
o
n
esia
(
SIBI)
an
d
I
n
d
o
n
esian
s
ig
n
lan
g
u
ag
e
o
r
b
a
h
a
s
a
is
ya
r
a
t
I
n
d
o
n
esia
(
B
I
SIN
DO)
.
SIBI
was
en
g
in
ee
r
ed
b
y
h
ea
r
i
n
g
in
d
iv
id
u
als
as
a
s
tan
d
ar
d
ized
s
y
s
tem
[
5
]
.
C
u
r
r
en
tly
,
SIBI
is
d
esig
n
ated
b
y
th
e
g
o
v
er
n
m
en
t
as
th
e
o
f
f
icial
s
ig
n
lan
g
u
ag
e
in
I
n
d
o
n
esia
an
d
is
tau
g
h
t
in
s
p
ec
ial
s
ch
o
o
ls
th
er
e.
B
I
SIN
DO
i
s
a
s
ig
n
lan
g
u
ag
e
f
o
r
m
ed
b
y
th
e
d
ea
f
co
m
m
u
n
ity
in
I
n
d
o
n
esia
an
d
em
er
g
ed
n
atu
r
ally
b
ased
o
n
th
e
o
b
s
er
v
atio
n
s
o
f
th
e
d
ea
f
[
6
]
.
T
h
e
ter
m
"BISI
NDO"
was
o
n
ly
u
s
ed
in
2
0
1
6
in
a
r
eso
lu
tio
n
o
f
th
e
s
ev
en
th
Ger
k
atin
c
o
n
g
r
ess
h
el
d
in
Ma
k
ass
ar
[
7
]
.
Alth
o
u
g
h
SIBI
is
an
o
f
f
icial
s
ig
n
lan
g
u
ag
e
in
I
n
d
o
n
esia,
in
f
ac
t,
d
ea
f
-
m
u
te
p
eo
p
le
p
r
ef
er
to
u
s
e
B
I
SIN
DO
r
ath
er
th
a
n
SIBI.
Deaf
-
m
u
te
p
eo
p
le
f
ee
l
th
at
B
I
SIN
DO
is
ea
s
ier
to
u
s
e
f
o
r
co
m
m
u
n
icat
io
n
b
ec
a
u
s
e
it
is
a
n
atu
r
al
s
ig
n
lan
g
u
ag
e.
T
h
e
in
ter
ac
tio
n
b
etwe
en
f
ac
ial
e
m
o
tio
n
s
an
d
n
atu
r
al
m
o
tio
n
g
estu
r
es
im
p
r
o
v
es
lan
g
u
ag
e
u
n
d
e
r
s
tan
d
in
g
f
o
r
b
o
th
d
ea
f
an
d
h
ea
r
in
g
in
d
iv
id
u
al
s
[
8
]
.
Pre
v
io
u
s
r
esear
ch
o
n
s
ig
n
lan
g
u
ag
e
r
ec
o
g
n
itio
n
h
as
b
ee
n
co
n
d
u
cted
b
y
v
ar
io
u
s
r
esear
ch
er
s
[
9
]
–
[
1
3
]
.
T
h
ese
s
tu
d
ies
h
av
e
u
s
ed
v
ar
io
u
s
m
o
d
els,
s
u
ch
as
ex
tr
em
e
le
ar
n
in
g
m
ac
h
i
n
e
(
E
L
M)
[
1
4
]
,
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
a
l
n
et
wo
r
k
(
C
NN)
,
lo
n
g
s
h
o
r
t
-
t
er
m
m
em
o
r
y
(
L
STM
)
,
a
n
d
a
co
m
b
in
atio
n
o
f
th
ese,
to
a
n
aly
ze
v
ar
io
u
s
h
a
n
d
s
ig
n
al
im
ag
es.
Ko
d
an
d
a
r
am
e
t
a
l.
[
9
]
u
s
ed
th
r
ee
C
NN
m
o
d
els
o
n
th
e
ASL
d
ataset,
wh
i
ch
r
esu
lted
in
9
7
%
ac
cu
r
ac
y
f
o
r
th
e
L
eNe
t
-
5
m
o
d
el,
9
8
.
9
% a
cc
u
r
ac
y
f
o
r
th
e
M
o
b
ileNetV2
m
o
d
el,
an
d
9
8
% a
cc
u
r
ac
y
f
o
r
th
e
b
u
ilt
m
o
d
el.
Sin
g
h
[
1
0
]
u
s
ed
th
e
3
D
C
NN
m
eth
o
d
o
n
a
d
ataset
co
n
s
is
tin
g
o
f
2
0
I
SL
h
an
d
g
e
s
tu
r
es,
wh
ich
wer
e
ca
p
tu
r
ed
u
n
d
er
d
if
f
er
en
t
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
,
s
u
ch
as
lig
h
tin
g
an
d
b
ac
k
g
r
o
u
n
d
c
o
n
d
itio
n
s
.
T
h
e
v
id
eo
s
wer
e
th
en
f
r
am
ed
a
n
d
ex
tr
ac
t
ed
,
an
d
1
6
2
,
8
4
0
im
ag
es
wer
e
o
b
tain
ed
.
T
h
e
a
v
er
ag
e
ac
cu
r
a
cy
o
b
tain
ed
was
as
h
ig
h
as
8
8
.
2
4
%.
C
o
s
ter
et
a
l.
[
1
1
]
u
s
ed
Op
e
n
Po
s
e
an
d
a
t
r
an
s
f
o
r
m
er
a
r
ch
itectu
r
e
t
o
r
ec
o
g
n
ize
th
e
Flem
is
h
s
ig
n
lan
g
u
ag
e
o
r
V
la
a
mse
g
eb
a
r
en
ta
a
l
(
VGT
)
d
ataset,
wh
ic
h
co
n
s
is
ted
o
f
1
0
0
class
es
co
n
s
is
tin
g
o
f
1
0
4
s
ig
n
wo
r
d
s
.
T
h
is
r
esear
c
h
tr
ain
e
d
t
h
r
ee
t
r
an
s
f
o
r
m
er
m
o
d
els,
ea
ch
with
a
d
if
f
er
en
t
em
b
e
d
d
in
g
s
ize,
an
d
ac
h
iev
e
d
an
ac
cu
r
ac
y
v
alu
e
o
f
7
4
.
7
%.
T
u
n
g
a
et
a
l.
[
1
2
]
u
s
ed
a
g
r
a
p
h
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
(
GC
N)
an
d
b
id
ir
ec
tio
n
a
l
en
co
d
er
r
e
p
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
to
tr
an
s
late
ASL,
with
an
ac
cu
r
ac
y
v
alu
e
o
f
6
0
.
1
5
%
an
d
4
2
.
1
8
%
o
n
th
e
f
ir
s
t
d
ataset
an
d
4
2
.
1
8
%
o
n
t
h
e
s
ec
o
n
d
d
ataset.
Gh
o
s
h
[
1
5
]
u
s
ed
Me
d
iaPip
e
as
a
cr
o
s
s
-
p
latf
o
r
m
f
r
am
ewo
r
k
,
co
m
b
in
in
g
it
with
L
STM
to
tr
a
n
s
late
ASL
.
T
h
e
b
est
p
er
f
o
r
m
i
n
g
alg
o
r
ith
m
in
th
e
test
wa
s
SVM,
with
ac
cu
r
ac
y
v
alu
es
o
f
9
9
.
1
5
%,
9
9
.
2
9
%,
9
8
.
1
9
%,
9
9
.
1
8
%,
an
d
9
6
.
2
2
%
in
th
e
f
iv
e
d
atasets
,
r
esp
ec
tiv
ely
.
Aljab
ar
a
n
d
Su
h
ar
jito
[
1
6
]
u
s
ed
th
r
ee
m
o
d
els,
n
am
ely
C
NN,
L
STM
,
an
d
a
c
o
m
b
in
atio
n
o
f
C
NN
an
d
L
STM
,
o
n
th
e
B
I
SIN
DO
d
ataset.
T
h
e
av
er
ag
e
ac
cu
r
ac
y
o
b
tain
ed
b
y
th
e
C
NN
m
o
d
el
w
as 7
3
%,
th
e
L
STM
m
o
d
el
was
8
1
%,
an
d
th
e
C
NN
-
L
STM
m
o
d
el
was
9
0
%
in
test
in
g
.
Hald
er
an
d
T
a
y
ad
e
[
1
7
]
co
n
d
u
cte
d
r
esear
ch
o
n
s
ig
n
lan
g
u
ag
e
r
ec
o
g
n
itio
n
f
r
o
m
v
ar
io
u
s
co
u
n
tr
ies,
in
clu
d
in
g
Am
er
ica,
I
n
d
ia,
I
taly
,
an
d
T
u
r
k
e
y
,
u
s
in
g
a
s
u
p
p
o
r
t v
ec
to
r
m
ac
h
in
e
(
SVM)
co
m
b
in
ed
with
Me
d
iaPip
e
f
o
r
g
estu
r
e
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
s
in
g
an
o
s
et
a
l.
[
1
8
]
p
r
o
p
o
s
ed
th
e
u
s
e
o
f
tem
p
o
r
al
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
(
T
C
Ns)
to
ad
d
r
ess
th
e
p
r
o
b
lem
o
f
s
u
r
f
ac
e
elec
tr
o
m
y
o
g
r
ap
h
y
(
s
E
MG
)
-
b
ased
h
an
d
g
estu
r
e
r
ec
o
g
n
iti
o
n
.
T
h
is
s
tu
d
y
h
ig
h
lig
h
ts
t
h
e
im
p
o
r
tan
ce
o
f
d
ev
elo
p
in
g
tailo
r
ed
d
ata
f
ee
d
in
g
m
eth
o
d
s
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
T
C
N
m
o
d
els
in
m
o
r
e
d
y
n
am
ic
u
s
ag
e
s
itu
atio
n
s
.
I
s
h
aq
et
a
l.
[
1
9
]
d
ev
elo
p
e
d
a
T
C
N
-
b
ased
em
o
tio
n
r
ec
o
g
n
itio
n
s
y
s
tem
th
at
u
tili
ze
s
a
s
p
atio
-
T
C
N
to
ascer
tain
th
e
s
p
ea
k
er
'
s
em
o
tio
n
al
s
tate
f
r
o
m
s
p
ee
ch
s
ig
n
als.
T
h
e
s
y
s
tem
au
to
m
atica
lly
ex
tr
ac
ts
v
alid
s
eq
u
en
ce
cu
es f
r
o
m
s
p
ee
ch
s
ig
n
als,
o
u
tp
er
f
o
r
m
in
g
b
o
t
h
tr
ad
itio
n
al
an
d
s
tate
-
of
-
th
e
-
a
r
t
(
SOTA
)
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
.
B
ased
o
n
p
r
ev
io
u
s
r
esear
ch
,
it
ca
n
b
e
s
ee
n
th
at
th
e
m
ain
r
ea
s
o
n
s
f
o
r
u
s
in
g
Me
d
iaPip
e
[
1
5
]
,
[
1
7
]
co
m
p
ar
ed
to
o
t
h
er
f
r
a
m
ewo
r
k
s
s
u
ch
as
Op
en
Po
s
e
[
1
1
]
ar
e
r
elate
d
to
s
ev
er
al
th
i
n
g
s
.
Am
o
n
g
th
e
m
is
th
e
s
u
p
er
io
r
ea
s
e
o
f
in
teg
r
atio
n
f
o
r
Me
d
iaPip
e.
I
n
ad
d
itio
n
,
M
ed
iaPip
e
is
s
u
p
er
io
r
in
th
e
m
o
b
ile
en
v
ir
o
n
m
en
t
co
m
p
ar
ed
to
Op
e
n
Po
s
e,
an
d
Me
d
iaPip
e
also
h
as
f
ewe
r
k
ey
p
o
in
ts
co
m
p
a
r
ed
to
Op
en
Po
s
e.
W
h
ile
Me
d
iaPip
e
an
d
Op
e
n
Po
s
e
ar
e
m
o
r
e
o
n
th
e
lib
r
ar
y
s
id
e,
th
e
r
e
also
e
x
is
t
s
ig
n
lan
g
u
a
g
e
a
p
p
licatio
n
s
th
at
u
tili
ze
th
e
h
ar
d
war
e
s
ec
tio
n
,
s
u
ch
as
th
e
leap
m
o
tio
n
d
ev
ice
[
2
0
]
a
n
d
ed
g
e
d
ev
ice
[
2
1
]
.
T
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tag
e
in
C
o
s
ter
et
a
l.
[
1
1
]
d
i
d
n
o
t
in
clu
d
e
s
tep
s
s
u
ch
as
f
r
a
m
e
s
elec
tio
n
an
d
n
o
r
m
aliza
tio
n
.
Similar
ly
,
T
u
n
g
a
et
a
l.
[
1
2
]
u
s
e
d
GC
N
a
n
d
B
E
R
T
t
o
t
r
an
s
l
at
e
ASL
wi
th
o
u
t
a
n
y
d
a
ta
p
r
e
-
p
r
o
ce
s
s
i
n
g
.
T
h
e
a
u
t
h
o
r
s
[
1
5
]
,
[
1
7
]
u
s
e
d
M
e
d
i
a
Pi
p
e
a
s
a
m
o
d
e
l
li
n
g
f
r
a
m
e
w
o
r
k
.
H
o
w
e
v
e
r
,
b
o
t
h
o
f
t
h
es
e
s
t
u
d
i
es
w
e
r
e
l
i
m
i
t
e
d
t
o
m
o
d
e
lin
g
t
h
e
h
a
n
d
r
eg
io
n
.
T
h
u
s
,
t
h
e
w
o
r
k
m
o
d
els th
e
e
n
t
ir
e
b
o
d
y
,
h
an
d
s
,
a
n
d
f
a
ce
.
A
d
d
iti
o
n
all
y
,
t
h
e
ir
cl
ass
i
f
ic
ati
o
n
m
eth
o
d
s
an
d
d
atas
ets
d
i
f
f
er
,
a
n
d
t
h
e
y
d
i
d
n
o
t
p
er
f
o
r
m
a
n
y
d
ata
p
r
e
-
p
r
o
c
ess
i
n
g
.
T
h
is
s
tu
d
y
d
if
f
er
s
f
r
o
m
Al
ja
b
a
r
a
n
d
Su
h
a
r
j
it
o
[
1
6
]
i
n
its
alg
o
r
ith
m
an
d
p
r
e
-
p
r
o
ce
s
s
in
g
s
tag
es.
T
h
eir
s
tu
d
y
em
p
l
o
y
ed
th
r
e
e
m
o
d
els:
C
NN,
L
STM
,
an
d
a
h
y
b
r
id
C
NN
-
L
STM
with
p
r
e
-
p
r
o
ce
s
s
in
g
th
at
in
clu
d
e
d
b
ac
k
g
r
o
u
n
d
s
u
b
tr
ac
tio
n
,
g
r
ay
s
ca
lin
g
,
an
d
r
esizin
g
.
L
ik
ewise,
th
e
r
esear
ch
b
y
Dwijay
an
ti
et
a
l.
[
6
]
em
p
lo
y
e
d
a
d
if
f
er
en
t
C
NN
alg
o
r
ith
m
an
d
p
er
f
o
r
m
ed
o
n
ly
im
ag
e
r
esizin
g
d
u
r
in
g
p
r
e
-
p
r
o
ce
s
s
in
g
,
f
o
cu
s
in
g
o
n
ly
o
n
s
tatic
alp
h
ab
et
s
ig
n
s
.
T
h
e
m
ain
d
if
f
er
e
n
ce
s
b
et
wee
n
th
is
s
tu
d
y
an
d
s
ev
er
al
r
elate
d
s
tu
d
ies
th
at
h
av
e
b
ee
n
co
n
d
u
cte
d
ar
e
in
th
e
m
eth
o
d
,
d
ataset,
d
ata
p
r
e
-
p
r
o
c
ess
in
g
m
eth
o
d
,
an
d
th
e
ty
p
e
o
f
tr
an
s
lated
lan
g
u
a
g
e.
I
n
th
is
s
tu
d
y
,
th
e
alg
o
r
ith
m
s
u
s
ed
ar
e
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M)
[
2
2
]
,
[
2
3
]
,
T
C
N
[
1
8
]
,
a
n
d
Me
d
iaPip
e
[
2
4
]
.
Pre
v
io
u
s
s
tu
d
ies
also
in
d
icate
d
t
h
at
Me
d
iaPip
e
h
as
b
ee
n
wid
ely
u
s
ed
in
s
ig
n
lan
g
u
ag
e
d
etec
tio
n
,
p
a
r
ticu
lar
ly
in
id
en
tify
in
g
h
an
d
an
d
f
ac
ial
la
n
d
m
ar
k
s
.
H
o
wev
er
,
th
e
m
ajo
r
ity
o
f
s
tu
d
ies
s
till
co
n
f
in
e
th
em
s
elv
es
to
s
tatic
g
estu
r
e
r
ec
o
g
n
itio
n
o
r
r
el
y
o
n
im
ag
e
-
b
ased
class
if
icatio
n
m
o
d
els.
T
h
e
m
ain
d
if
f
er
en
ce
f
r
o
m
p
r
ev
io
u
s
s
tu
d
ies
is
th
e
f
o
cu
s
o
n
r
ea
l
-
tim
e
v
id
eo
-
b
ase
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Temp
o
r
a
l c
o
n
vo
lu
tio
n
a
l n
etw
o
r
k
fo
r
I
n
d
o
n
esia
n
w
o
r
d
b
y
w
o
r
d
s
ig
n
la
n
g
u
a
g
e
…
(
S
a
r
a
h
P
u
r
n
a
ma
w
a
ti
)
4083
g
estu
r
e
r
ec
o
g
n
itio
n
,
r
ath
er
th
a
n
ju
s
t
s
tatic
im
ag
es,
s
o
th
at
th
e
d
ev
elo
p
e
d
s
y
s
tem
ca
n
ca
p
tu
r
e
th
e
co
m
p
le
x
ity
o
f
h
an
d
g
estu
r
es in
r
ea
l c
o
m
m
u
n
i
ca
tio
n
.
2.
M
E
T
H
O
D
T
h
er
e
ar
e
s
ev
er
al
s
tag
es
ca
r
r
ie
d
o
u
t
in
th
is
r
esear
ch
in
o
r
d
er
to
r
ea
lize
th
e
o
b
jectiv
es
o
f
th
e
r
esear
ch
.
Fig
u
r
e
1
s
h
o
ws
th
e
s
y
s
tem
a
r
ch
itectu
r
e
o
f
th
is
r
esear
ch
.
T
h
e
s
tag
es
ca
r
r
ied
o
u
t
i
n
th
is
r
esear
ch
ar
e
v
i
d
eo
ac
q
u
is
itio
n
,
v
id
eo
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
th
e
tr
ain
i
n
g
p
r
o
c
ess
,
an
d
th
e
test
in
g
p
r
o
ce
s
s
.
At
th
e
v
id
eo
ac
q
u
is
itio
n
s
tag
e,
d
ata
in
th
e
f
o
r
m
o
f
B
I
SIN
DO
g
estu
r
e
v
id
eo
s
ar
e
co
llected
,
th
e
n
clea
n
ed
an
d
r
esized
as
n
ee
d
ed
.
Fu
r
th
er
m
o
r
e,
th
e
v
i
d
eo
g
o
es
th
r
o
u
g
h
a
p
r
ep
r
o
ce
s
s
in
g
s
tag
e
wh
ich
in
clu
d
es
th
e
p
r
o
ce
s
s
o
f
f
r
am
e
ex
tr
ac
ti
o
n
,
k
ey
f
r
am
e
s
elec
tio
n
,
an
d
r
e
d
u
cin
g
th
e
n
u
m
b
e
r
o
f
f
r
a
m
e
s
,
b
ef
o
r
e
k
ey
p
o
in
t
ex
tr
ac
tio
n
is
ca
r
r
ied
o
u
t
at
t
h
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
s
tag
e.
I
n
t
h
e
f
ea
tu
r
e
ex
tr
ac
tio
n
s
tag
e,
th
e
e
x
tr
ac
ted
k
ey
p
o
in
ts
will
g
o
th
r
o
u
g
h
au
g
m
e
n
tatio
n
an
d
n
o
r
m
aliza
tio
n
s
tag
es
to
in
cr
ea
s
e
th
e
g
en
er
aliza
tio
n
lev
el
o
f
th
e
m
o
d
el.
T
h
e
ex
tr
ac
ted
k
ey
p
o
in
ts
ar
e
th
en
p
r
o
ce
s
s
ed
ag
ain
to
p
r
o
d
u
ce
in
p
u
t
d
ata
th
at
will
b
e
u
s
ed
wh
en
tr
ain
in
g
th
e
m
o
d
el.
T
h
e
tr
ain
in
g
s
tag
e
will
p
r
o
d
u
ce
a
tr
ain
e
d
m
o
d
el
th
at
will
b
e
u
s
ed
t
o
tr
a
n
s
late
B
I
SIN
DO
in
th
e
test
in
g
s
tag
e.
T
h
e
test
in
g
p
r
o
ce
s
s
is
ca
r
r
ied
o
u
t
in
r
ea
l
tim
e
to
test
th
e
g
e
n
er
aliza
tio
n
o
f
th
e
m
o
d
el
in
t
r
a
n
s
latin
g
B
I
SIN
DO
in
s
ev
er
al
d
if
f
er
e
n
t sit
u
atio
n
s
.
Fig
u
r
e
1
.
Sy
s
tem
a
r
c
h
itectu
r
e
2
.
1
.
Da
t
a
s
et
T
h
e
d
ata
u
s
ed
ar
e
wo
r
d
-
lev
el
B
I
SIN
DO
s
ig
n
s
tak
en
u
s
in
g
a
s
m
ar
tp
h
o
n
e
ca
m
er
a
with
a
r
e
s
o
lu
tio
n
o
f
1
4
MP
an
d
a
f
r
am
e
r
ate
o
f
3
0
f
r
am
es
p
er
s
ec
o
n
d
,
m
o
u
n
ted
o
n
a
tr
ip
o
d
.
T
h
e
tr
ip
o
d
was
p
la
ce
d
at
a
d
is
tan
ce
o
f
ap
p
r
o
x
im
ately
7
5
cm
f
r
o
m
th
e
s
ig
n
er
,
with
a
h
eig
h
t
o
f
a
p
p
r
o
x
im
ately
1
.
4
m
eter
s
.
T
h
e
v
id
e
o
was
tak
en
in
m
id
s
h
o
t,
f
r
o
m
th
e
to
p
o
f
th
e
h
ea
d
to
th
e
waist
(
s
ee
Fig
u
r
e
2
)
.
E
n
s
u
r
e
th
e
en
tire
b
o
d
y
(
f
r
o
m
w
aist
to
to
p
o
f
h
ea
d
)
ap
p
ea
r
s
in
ea
c
h
v
id
e
o
f
r
am
e
as
s
h
o
wn
in
Fig
u
r
e
2
(
a)
.
T
h
is
f
ig
u
r
e
d
em
o
n
s
tr
ates
th
e
r
eq
u
ir
em
en
t
to
k
ee
p
t
h
e
en
tire
b
o
d
y
(
f
r
o
m
waist
to
to
p
o
f
h
ea
d
)
with
in
th
e
f
r
am
e
.
As
s
h
o
wn
in
Fig
u
r
e
2
(
b
)
,
th
e
g
estu
r
e
m
o
v
em
e
n
ts
m
u
s
t
b
e
clea
r
an
d
in
f
r
am
e.
T
h
is
f
ig
u
r
e
h
ig
h
lig
h
ts
th
e
i
n
co
r
r
ec
t
cu
t
-
o
f
f
o
f
t
h
e
h
a
n
d
d
u
r
i
n
g
a
s
ig
n
.
R
ec
o
r
d
i
n
g
m
u
s
t
b
e
d
o
n
e
i
n
well
-
lit
co
n
d
i
tio
n
s
,
as
s
h
o
wn
in
Fig
u
r
e
2
(
c)
,
to
en
s
u
r
e
clea
r
ca
p
tu
r
e
o
f
k
ey
p
o
in
ts
.
T
h
is
f
ig
u
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
0
8
1
-
4
0
9
5
4084
co
n
tr
asts
p
r
o
p
e
r
illu
m
in
atio
n
with
an
u
n
d
er
ex
p
o
s
ed
,
d
ar
k
e
n
v
ir
o
n
m
en
t.
Vid
e
o
r
ec
o
r
d
in
g
s
m
u
s
t
h
av
e
a
s
tab
le
b
ac
k
g
r
o
u
n
d
an
d
n
o
t
in
clu
d
e
an
y
h
an
d
s
o
r
f
ac
es
ap
ar
t
f
r
o
m
th
e
h
an
d
s
an
d
f
ac
e
o
f
th
e
p
e
r
f
o
r
m
er
.
Ad
j
u
s
t
th
e
r
ec
o
r
d
in
g
d
ev
ice
s
o
th
at
it
is
as
s
tab
le
an
d
f
o
cu
s
ed
as
p
o
s
s
ib
le.
T
h
e
r
ec
o
r
d
ed
v
id
eo
was
in
.
m
p
4
f
o
r
m
at
with
a
f
r
am
e
s
ize
o
f
1
9
2
0
×
1
0
8
0
p
ix
els
an
d
was
a
lo
n
g
v
id
eo
c
o
n
s
is
tin
g
o
f
s
ev
er
al
s
ig
n
v
id
eo
s
.
T
h
e
r
aw
v
id
eo
i
s
th
en
clea
n
e
d
a
n
d
cu
t,
th
e
n
u
m
b
er
o
f
s
ig
n
v
id
e
o
s
o
b
tain
ed
f
r
o
m
cu
ttin
g
th
e
r
aw
v
id
eo
is
2
,
7
0
0
v
i
d
eo
s
co
n
s
is
tin
g
o
f
2
7
wo
r
d
s
,
wh
ic
h
m
ea
n
s
ea
ch
s
ig
n
h
as
a
to
t
al
o
f
1
0
0
v
id
eo
s
.
E
ac
h
v
id
e
o
is
th
en
p
u
t
in
to
a
d
ir
ec
to
r
y
with
a
n
am
e
th
at
co
r
r
esp
o
n
d
s
to
th
e
m
ea
n
in
g
o
f
i
ts
r
esp
ec
tiv
e
s
ig
n
.
W
e
g
ath
er
e
d
all
th
e
d
ata
f
r
o
m
1
0
p
e
o
p
le
—
5
m
ales
a
n
d
5
f
e
m
ales
—
in
clu
d
in
g
th
o
s
e
f
r
o
m
th
e
d
ea
f
-
m
u
te
s
ch
o
o
l.
T
h
e
p
a
r
ticip
an
ts
h
av
e
b
ee
n
cr
ea
tin
g
s
ev
er
al
v
id
eo
s
d
e
m
o
n
s
tr
atin
g
ea
ch
o
f
th
e
2
7
s
ig
n
lan
g
u
ag
e
wo
r
d
s
alo
n
g
with
th
eir
v
ar
i
o
u
s
in
ter
p
r
etatio
n
s
.
T
h
u
s
,
it
m
ea
n
s
ev
er
y
p
er
s
o
n
will
p
er
f
o
r
m
1
0
s
ig
n
lan
g
u
a
g
e
v
id
eo
s
f
o
r
ev
er
y
2
7
wo
r
d
s
.
T
h
e
d
ata
was
d
is
tr
ib
u
ted
in
to
two
g
r
o
u
p
s
:
tr
ai
n
in
g
an
d
v
alid
ati
o
n
,
w
h
ile
test
in
g
will
b
e
d
o
n
e
in
r
ea
l
tim
e.
T
h
e
tr
ain
in
g
d
ata
will
tak
e
7
0
%
of
th
e
o
v
er
all
d
ata,
a
n
d
th
e
v
alid
atio
n
will
tak
e
3
0
%
o
f
t
h
e
r
est
.
T
h
e
2
7
s
ig
n
s
u
s
ed
ca
n
b
e
g
r
o
u
p
ed
in
to
s
ix
ca
teg
o
r
ies:
i)
Qu
esti
o
n
wo
r
d
s
: ‘
wh
at’
,
‘
wh
er
e’
,
‘
h
o
w
m
u
ch
’
,
‘
wh
e
n
’
,
‘
w
h
o
’
,
‘
wh
y
’
,
an
d
‘
h
o
w’
.
ii)
E
x
p
r
ess
io
n
s
: ‘
th
an
k
y
o
u
’
an
d
‘
s
o
r
r
y
’
.
iii)
Gr
ee
tin
g
s
: ‘
g
o
o
d
m
o
r
n
in
g
’
,
‘
g
o
o
d
af
te
r
n
o
o
n
’
,
a
n
d
‘
g
o
o
d
ev
e
n
in
g
’
.
iv
)
T
im
e
wo
r
d
s
: ‘
to
d
a
y
’
,
‘
to
m
o
r
r
o
w’
,
‘
n
o
w’
,
‘
y
ester
d
a
y
’
,
a
n
d
‘
d
ay
’
.
v)
Per
s
o
n
al
p
r
o
n
o
u
n
s
:
‘
I
’
,
‘
h
e/sh
e’
,
‘
y
o
u
’
,
a
n
d
‘
we
’
.
v
i)
Ad
d
itio
n
al
wo
r
d
s
: ‘
n
am
e’
,
‘
n
u
m
b
er
’
,
‘
p
h
o
n
e’
,
‘
d
ea
f
’
,
‘
ea
t’
,
a
n
d
‘
s
ir
’
.
(
a)
(
b
)
(
c)
Fig
u
r
e
2
.
C
o
r
r
ec
t
an
d
i
n
co
r
r
ec
t im
ag
e
ac
q
u
is
itio
n
s
am
p
les
: (
a
)
b
o
d
y
f
r
a
m
in
g
,
(
b
)
g
estu
r
e
m
o
v
em
en
t,
a
n
d
(
c)
lig
h
tin
g
co
n
d
itio
n
s
2
.
2
.
Video
prepro
ce
s
s
ing
:
f
ra
m
e
ex
t
ra
ct
io
n,
k
ey
f
ra
m
e
s
elec
t
io
n,
a
nd
f
ra
m
e
re
du
ct
io
n
T
h
er
e
ar
e
th
r
ee
p
r
o
ce
s
s
es in
th
e
v
id
eo
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e,
f
r
am
e
ex
tr
ac
tio
n
,
k
e
y
f
r
am
e
s
ele
ctio
n
,
an
d
f
r
am
e
r
ed
u
ctio
n
.
T
h
e
n
u
m
b
er
o
f
f
r
am
es
e
x
tr
ac
ted
f
r
o
m
2
,
7
0
0
v
id
e
o
s
is
7
7
,
4
7
0
f
r
am
es.
E
ac
h
f
r
am
e
will
b
e
s
co
r
ed
b
y
co
n
s
id
er
i
n
g
s
ev
er
al
v
ar
iab
les,
n
am
ely
h
an
d
m
o
v
em
en
t,
h
an
d
p
o
s
itio
n
,
an
d
h
a
n
d
b
lu
r
lev
el
in
th
e
f
r
am
e.
T
h
e
co
m
b
in
atio
n
o
f
th
e
s
e
th
r
ee
v
ar
iab
les
is
b
eliev
ed
to
b
e
ab
le
to
s
o
r
t
o
u
t
u
n
im
p
o
r
t
an
t
f
r
am
es,
n
am
ely
f
r
am
es
with
h
an
d
s
th
at
ar
e
tr
an
s
itio
n
in
g
,
eith
er
tr
an
s
itio
n
in
g
to
s
tar
t
th
e
g
estu
r
e
o
r
tr
an
s
itio
n
in
g
to
en
d
th
e
g
estu
r
e.
T
h
e
h
an
d
u
s
u
ally
m
o
v
es
v
er
tically
wh
en
s
tar
tin
g
th
e
s
ig
n
as
well
as
f
in
is
h
in
g
it.
W
h
en
s
tar
tin
g
a
s
ig
n
,
th
e
h
an
d
m
o
v
es
u
p
war
d
s
to
war
d
s
th
e
s
tar
tin
g
p
o
s
itio
n
o
f
th
e
s
ig
n
,
a
n
d
wh
e
n
f
in
is
h
i
n
g
a
s
ig
n
,
th
e
h
an
d
will
b
e
lo
wer
ed
to
r
etu
r
n
to
th
e
s
tar
tin
g
p
o
s
itio
n
,
i.e
.
,
th
e
r
e
ad
y
p
o
s
itio
n
.
T
h
e
r
ef
o
r
e,
h
an
d
m
o
v
em
en
ts
ca
n
b
e
u
s
ed
in
s
ea
r
ch
in
g
f
o
r
f
r
am
es
with
tr
an
s
itio
n
in
g
h
an
d
s
.
T
o
o
b
tain
th
e
v
alu
e
o
f
th
is
v
ar
ia
b
le,
th
e
ca
lcu
latio
n
in
v
o
lv
es
co
m
p
a
r
in
g
th
e
ce
n
ter
co
o
r
d
in
ates
o
f
t
h
e
two
h
an
d
s
at
f
r
am
e
f
with
th
o
s
e
at
f
r
am
e
f
+
1
;
th
e
h
ig
h
est
r
esu
lt
f
r
o
m
th
is
co
m
p
ar
is
o
n
r
e
p
r
esen
ts
th
e
v
alu
e
o
f
th
e
f
ir
s
t
v
ar
iab
le.
T
h
e
m
id
p
o
in
t
v
alu
e
is
th
e
d
is
tan
ce
f
r
o
m
th
e
m
in
im
u
m
a
n
d
m
a
x
im
u
m
y
co
o
r
d
i
n
ates d
iv
id
ed
b
y
two
.
B
lu
r
r
in
ess
in
th
e
f
r
am
e
is
d
etec
ted
b
y
v
ar
iatio
n
o
f
th
e
L
ap
l
ac
ian
m
eth
o
d
with
th
e
L
ap
lacia
n
k
er
n
el
u
s
ed
.
T
o
g
et
m
o
r
e
ac
cu
r
ate
r
e
s
u
lts
,
th
e
v
ar
iatio
n
o
f
th
e
L
ap
lacia
n
m
eth
o
d
is
o
n
ly
ap
p
lied
to
b
o
th
h
an
d
s
an
d
n
o
t
to
t
h
e
en
tire
f
r
am
e.
T
h
is
is
d
o
n
e
s
o
th
e
m
eth
o
d
ca
n
r
ea
ll
y
f
o
c
u
s
o
n
d
eter
m
in
i
n
g
th
e
le
v
el
o
f
b
lu
r
o
n
h
an
d
s
an
d
n
o
t
o
n
th
e
e
n
tire
f
r
am
e.
Me
d
iaPip
e
ca
n
f
in
d
th
e
h
an
d
's
lo
ca
tio
n
,
an
d
th
e
ax
es'
m
in
an
d
m
ax
co
o
r
d
in
ate
s
ar
e
u
s
ed
to
cr
o
p
it.
T
h
e
m
eth
o
d
is
th
en
ap
p
lied
to
th
e
c
r
o
p
p
e
d
h
an
d
im
ag
e
to
d
etec
t its
b
lu
r
r
in
ess
lev
el.
T
h
e
k
e
y
f
r
a
m
e
s
e
le
c
t
i
o
n
m
e
c
h
a
n
i
s
m
o
n
l
y
f
o
c
u
s
es
o
n
r
e
t
r
i
ev
i
n
g
f
r
a
m
e
s
t
h
a
t
c
o
n
t
a
i
n
s
i
g
n
s
w
it
h
o
u
t
c
o
n
s
i
d
e
r
i
n
g
t
h
e
p
o
s
s
i
b
i
l
it
y
o
f
d
u
p
l
i
c
a
t
i
o
n
a
n
d
r
e
d
u
n
d
a
n
c
y
i
n
t
h
e
f
r
a
m
e
s
,
s
o
t
h
e
n
u
m
b
e
r
o
f
f
r
a
m
e
s
is
s
t
i
ll
q
u
i
t
e
l
a
r
g
e
.
T
h
e
w
a
y
f
r
a
m
e
r
e
d
u
ct
i
o
n
w
o
r
k
s
i
s
b
y
c
o
m
p
a
r
i
n
g
t
h
e
s
u
m
m
a
t
i
o
n
o
f
t
h
e
c
o
o
r
d
i
n
a
t
es
i
n
th
e
h
e
a
d
f
r
a
m
e
w
i
t
h
t
h
e
s
u
m
o
f
t
h
e
c
o
o
r
d
i
n
a
t
e
s
i
n
t
h
e
f
r
a
m
e
a
f
t
e
r
t
h
e
h
e
a
d
f
r
a
m
e.
'
He
a
d
f
r
a
m
e'
i
s
t
h
e
t
e
r
m
u
s
ed
t
o
d
e
n
o
t
e
t
h
e
l
a
s
t
f
r
a
m
e
s
e
l
e
ct
e
d
b
y
t
h
e
s
y
s
t
e
m
.
A
t
t
h
e
s
t
a
r
t
o
f
t
h
e
p
r
o
c
e
s
s
,
t
h
e
f
i
r
s
t
f
r
a
m
e
o
f
t
h
e
s
i
g
n
w
i
ll
b
e
u
s
e
d
a
s
t
h
e
h
e
a
d
f
r
a
m
e
.
T
h
e
s
u
m
m
a
t
i
o
n
o
f
t
h
e
c
o
o
r
d
i
n
a
t
e
s
o
f
t
h
e
h
e
a
d
f
r
am
e
i
s
t
h
e
n
c
o
m
p
a
r
e
d
w
it
h
t
h
e
s
u
m
m
a
ti
o
n
o
f
t
h
e
c
o
o
r
d
i
n
a
t
e
s
o
f
t
h
e
n
e
x
t
f
r
a
m
e
(
i
n
t
h
i
s
c
a
s
e
,
t
h
e
s
ec
o
n
d
f
r
a
m
e
)
,
a
n
d
i
f
t
h
e
d
i
f
f
e
r
e
n
c
e
i
s
s
m
a
l
l
e
r
t
h
a
n
t
h
e
t
h
r
e
s
h
o
l
d
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Temp
o
r
a
l c
o
n
vo
lu
tio
n
a
l n
etw
o
r
k
fo
r
I
n
d
o
n
esia
n
w
o
r
d
b
y
w
o
r
d
s
ig
n
la
n
g
u
a
g
e
…
(
S
a
r
a
h
P
u
r
n
a
ma
w
a
ti
)
4085
t
h
e
f
r
a
m
e
a
f
t
e
r
t
h
e
h
e
a
d
f
r
a
m
e
w
i
l
l
b
e
r
e
m
o
v
e
d
b
e
c
a
u
s
e
i
t
i
s
c
o
n
s
i
d
e
r
e
d
s
i
m
i
la
r
.
I
f
t
h
e
d
i
f
f
e
r
e
n
c
e
e
x
c
e
e
d
s
t
h
e
t
h
r
e
s
h
o
l
d
,
t
h
e
n
e
x
t
f
r
a
m
e
wi
l
l
b
e
t
h
e
h
e
a
d
f
r
a
m
e
i
n
t
h
e
n
e
x
t
c
o
m
p
a
r
i
s
o
n
.
T
h
e
t
h
r
e
s
h
o
l
d
v
a
l
u
e
o
f
a
g
r
o
u
p
o
f
f
r
a
m
e
s
is
o
b
t
a
i
n
e
d
b
y
m
u
l
t
i
p
l
y
in
g
a
p
o
s
i
t
i
v
e
c
o
n
s
t
a
n
t
b
y
t
h
e
av
e
r
a
g
e
c
h
a
n
g
e
i
n
t
h
e
g
r
o
u
p
'
s
k
e
y
p
o
i
n
t
d
i
s
t
a
n
ce
.
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n:
la
n
dm
a
rk
c
o
nfig
ura
t
io
n,
k
ey
po
i
nt
ex
t
ra
ct
io
n,
a
nd
da
t
a
a
ug
m
ent
a
t
io
n
T
h
is
p
r
o
ce
s
s
u
s
es
Me
d
ia
P
ip
e
h
o
lis
tic
to
p
r
ed
ict
la
n
d
m
ar
k
s
an
d
ex
tr
ac
t
f
ea
tu
r
es
o
n
t
h
e
f
r
am
e.
T
h
e
p
r
o
ce
s
s
es
p
er
f
o
r
m
ed
ar
e
lan
d
m
a
r
k
c
o
n
f
ig
u
r
atio
n
,
k
e
y
p
o
in
t
ex
tr
ac
tio
n
,
d
ata
au
g
m
e
n
tatio
n
,
k
ey
p
o
i
n
t
n
o
r
m
aliza
tio
n
,
an
d
k
ey
p
o
in
t
s
av
in
g
.
B
y
d
ef
a
u
lt,
h
o
lis
tic
d
r
aws
a
to
tal
o
f
5
4
3
lan
d
m
ar
k
s
,
c
o
n
s
is
tin
g
o
f
3
3
la
n
d
m
ar
k
s
f
r
o
m
p
o
s
e
[
25
]
,
2
1
lan
d
m
ar
k
s
(
p
er
h
a
n
d
)
f
r
o
m
h
an
d
s
[
26
]
,
an
d
4
6
8
lan
d
m
ar
k
s
f
r
o
m
f
ac
e
m
esh
[
27
]
.
Ho
wev
e
r
,
in
th
is
r
esear
ch
,
n
o
t
all
lan
d
m
a
r
k
s
ar
e
n
ee
d
ed
,
s
o
s
o
m
e
lan
d
m
ar
k
s
will
n
o
t
b
e
d
r
awn
.
T
h
e
r
e
wer
e
n
o
c
h
an
g
e
s
to
h
an
d
lan
d
m
a
r
k
s
,
wh
ile
f
a
ce
m
esh
lan
d
m
ar
k
s
wer
e
n
o
t
u
s
ed
en
tire
ly
[
28
]
.
Po
s
e
o
r
ig
in
ally
d
r
ew
3
3
lan
d
m
ar
k
s
,
af
ter
b
ein
g
co
n
f
ig
u
r
ed
p
o
s
e
o
n
ly
d
r
ew
1
7
lan
d
m
ar
k
s
.
Af
ter
b
ein
g
co
n
f
ig
u
r
ed
,
p
o
s
e
o
n
ly
d
r
aws
lan
d
m
ar
k
s
o
n
th
e
f
ac
e
ar
ea
as
well
as
lan
d
m
ar
k
s
o
f
b
o
th
s
h
o
u
ld
er
s
,
elb
o
ws,
an
d
wr
is
ts
.
T
h
e
lan
d
m
ar
k
s
o
f
p
o
s
e
th
at
ar
e
n
o
t
u
s
ed
a
r
e
th
e
lan
d
m
ar
k
s
o
f
th
e
h
an
d
,
h
i
p
,
k
n
ee
,
an
d
f
o
o
t
a
r
ea
s
.
T
h
e
to
tal
n
u
m
b
er
o
f
lan
d
m
ar
k
s
a
f
ter
co
n
f
ig
u
r
atio
n
is
5
9
,
wh
i
ch
co
n
s
is
ts
o
f
1
7
p
o
s
e
lan
d
m
ar
k
s
an
d
2
1
h
a
n
d
lan
d
m
ar
k
s
o
n
ea
ch
h
an
d
.
T
h
e
Me
d
iaPip
e
co
o
r
d
in
ate
s
y
s
tem
s
tar
ts
f
r
o
m
th
e
to
p
lef
t
c
o
r
n
er
o
f
t
h
e
f
r
am
e
a
n
d
is
th
en
n
o
r
m
alize
d
in
to
a
d
is
tan
ce
o
f
0
to
1
,
r
esp
ec
tiv
ely
b
y
th
e
wid
th
an
d
h
ei
g
h
t
o
f
th
e
im
ag
e
.
E
ac
h
lan
d
m
ar
k
co
n
s
is
ts
o
f
th
r
ee
co
o
r
d
in
ates,
,
,
an
d
,
b
u
t
in
th
is
r
esear
ch
o
n
ly
an
d
co
o
r
d
in
ates
ar
e
ex
tr
ac
ted
f
o
r
u
s
e.
Sin
ce
o
n
ly
two
co
o
r
d
in
ates
ar
e
ex
tr
ac
te
d
,
th
i
s
m
ea
n
s
th
at
in
o
n
e
f
r
am
e
t
h
er
e
ar
e
a
to
tal
o
f
1
1
8
k
e
y
p
o
in
ts
th
at
ca
n
b
e
ex
tr
ac
ted
.
T
h
e
n
u
m
b
er
1
1
8
is
d
er
iv
ed
f
r
o
m
th
e
1
7
p
o
s
e
lan
d
m
ar
k
s
an
d
th
e
2
1
h
a
n
d
lan
d
m
ar
k
s
f
o
r
ea
ch
h
a
n
d
,
m
u
ltip
lied
b
y
2
,
wh
ich
r
e
p
r
ese
n
ts
th
e
n
u
m
b
er
o
f
c
o
o
r
d
i
n
ates
ex
tr
ac
ted
.
Han
d
s
th
at
d
o
n
o
t
a
p
p
ea
r
in
th
e
f
r
am
e
will b
e
r
ep
lace
d
with
a
v
al
u
e
o
f
0
.
T
o
m
in
im
ize
th
e
p
o
s
s
ib
ilit
y
o
f
o
v
e
r
f
itti
n
g
an
d
im
p
r
o
v
e
th
e
g
en
er
aliza
tio
n
o
f
th
e
m
o
d
el,
s
p
atial
d
at
a
au
g
m
en
tatio
n
o
f
th
e
co
o
r
d
i
n
a
tes
wa
s
p
er
f
o
r
m
ed
.
T
h
e
au
g
m
en
tatio
n
p
ar
am
eter
s
wer
e
ch
o
s
en
r
an
d
o
m
ly
b
u
t
r
em
ain
ed
co
n
s
is
ten
t
f
o
r
f
r
a
m
e
s
in
th
e
s
am
e
s
ig
n
g
r
o
u
p
.
T
h
e
au
g
m
en
tatio
n
tech
n
iq
u
es
p
er
f
o
r
m
ed
o
n
th
e
d
ata
ar
e
as f
o
llo
ws
.
2
.
3
.
1
.
F
lip
T
h
e
f
lip
is
p
er
f
o
r
m
e
d
to
s
im
u
late
g
estu
r
es
m
ad
e
with
th
e
o
p
p
o
s
ite
h
an
d
f
r
o
m
th
e
ac
tiv
e
h
an
d
in
th
e
in
itial
d
ata.
T
h
e
r
ea
s
o
n
f
o
r
th
e
f
lip
is
th
at
in
s
ig
n
lan
g
u
ag
e,
g
estu
r
es
ca
n
b
e
m
a
d
e
with
eit
h
er
th
e
r
ig
h
t
o
r
lef
t
h
an
d
.
Gestu
r
es
m
ad
e
with
eit
h
er
h
an
d
h
a
v
e
th
e
s
am
e
m
ea
n
in
g
.
T
h
er
ef
o
r
e,
a
h
o
r
iz
o
n
tal
f
lip
is
p
er
f
o
r
m
ed
s
o
th
at
v
ar
iatio
n
s
in
g
estu
r
es
u
s
i
n
g
th
e
r
ig
h
t
o
r
le
f
t
h
a
n
d
ca
n
b
e
r
ec
o
g
n
ized
b
y
th
e
s
y
s
tem
.
C
o
o
r
d
in
ate
d
ata
th
at
h
as u
n
d
er
g
o
n
e
f
lip
au
g
m
en
tati
o
n
will a
ls
o
u
n
d
e
r
g
o
a
n
o
th
er
a
u
g
m
en
tatio
n
p
r
o
ce
s
s
.
2
.
3
.
2
.
Ro
t
a
t
io
n
All
co
o
r
d
in
ates
a
r
e
r
o
tated
w
ith
a
r
a
n
d
o
m
l
y
s
elec
ted
an
g
l
e
θ
b
etwe
en
6
to
1
5
d
eg
r
ee
s
,
with
th
e
ce
n
ter
o
f
r
o
tatio
n
lo
ca
ted
at
th
e
ce
n
ter
o
f
th
e
f
r
am
e
,
wh
ich
i
s
th
e
co
o
r
d
in
ates
[
0
.
5
,
0
.
5
]
.
I
n
(
1
)
is
a
ca
lcu
latio
n
to
o
b
tain
th
e
r
ad
ian
v
alu
e
f
r
o
m
a
r
an
d
o
m
ly
s
elec
ted
a
n
g
le
,
an
d
(
2
)
d
escr
ib
es
th
e
ca
lcu
l
atio
n
to
o
b
tain
th
e
co
o
r
d
in
ate
v
alu
e
af
te
r
g
o
i
n
g
th
r
o
u
g
h
th
e
r
o
tatio
n
p
r
o
ce
s
s
.
=
∙
180
(
1
)
(
,
)
=
(
−
0
.
5
)
c
os
−
(
−
0
.
5
)
s
in
+
0
.
5
,
(
−
0
.
5
)
c
os
+
(
−
0
.
5
)
s
in
+
0
.
5
(
2
)
W
h
er
e
,
ar
e
th
e
in
p
u
t
co
o
r
d
in
a
tes,
,
ar
e
th
e
co
o
r
d
i
n
ates
o
f
th
e
ce
n
ter
o
f
r
o
tatio
n
,
is
th
e
d
eg
r
ee
o
f
an
g
le,
is
th
e
an
g
le
o
f
r
ad
ian
.
2
.
3
.
3
.
Sq
ueez
e
Sq
u
ee
ze
is
p
er
f
o
r
m
ed
o
n
th
e
ax
is
with
two
r
an
d
o
m
r
atio
s
1
an
d
2
b
etwe
en
8
to
1
5
%
o
f
th
e
im
ag
e
wid
th
.
T
h
e
y
c
o
o
r
d
in
ate
v
alu
e
is
f
ix
ed
b
ec
au
s
e
s
q
u
ee
ze
is
o
n
ly
p
e
r
f
o
r
m
ed
o
n
th
e
h
o
r
izo
n
tal
ax
is
.
Sq
u
ee
ze
m
ak
es
th
e
p
r
o
p
o
r
tio
n
s
o
f
th
e
ac
to
r
'
s
b
o
d
y
s
m
aller
,
s
o
it
ca
n
s
im
u
late
v
id
eo
r
ec
o
r
d
i
n
g
w
ith
th
e
ac
to
r
f
u
r
th
er
awa
y
f
r
o
m
t
h
e
ca
m
er
a.
2
.
3
.
4
.
Str
et
ch
Similar
to
s
q
u
ee
ze
,
s
tr
etch
is
also
p
er
f
o
r
m
ed
o
n
th
e
h
o
r
izo
n
tal
ax
is
with
two
d
if
f
er
e
n
t
r
an
d
o
m
r
atio
s
b
etwe
en
8
to
1
5
%
o
f
th
e
im
a
g
e
wid
th
.
T
h
e
y
co
o
r
d
in
ate
v
alu
e
r
em
ain
s
u
n
c
h
an
g
e
d
.
Stre
tch
s
im
u
lates
v
id
eo
r
ec
o
r
d
in
g
with
th
e
d
is
tan
ce
o
f
th
e
m
o
d
el
clo
s
er
t
o
th
e
ca
m
e
r
a
b
y
en
lar
g
in
g
p
r
o
p
o
r
tio
n
o
f
th
e
m
o
d
el'
s
b
o
d
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
0
8
1
-
4
0
9
5
4086
2
.
3
.
5
.
P
er
s
pect
iv
e
t
ra
ns
f
o
r
m
T
h
e
p
o
s
e
co
o
r
d
i
n
ates
ar
e
p
r
o
j
ec
ted
in
to
a
s
p
atially
n
ew
p
la
n
e
with
a
r
an
d
o
m
ly
d
ef
i
n
ed
p
r
o
jectio
n
ce
n
ter
at
an
in
ter
v
al
b
etwe
en
0
an
d
1
.
T
h
e
n
ew
p
lan
e
is
d
r
awn
b
y
s
u
b
tr
ac
tin
g
t
h
e
wid
th
at
th
e
s
elec
ted
s
id
e
an
d
th
e
v
er
tical
ed
g
e
o
f
ea
c
h
ad
jace
n
t
co
r
n
e
r
.
T
h
is
au
g
m
e
n
tatio
n
s
im
u
lates
a
g
estu
r
e
v
id
eo
r
ec
o
r
d
in
g
with
a
s
lig
h
t
ca
m
er
a
tilt
.
T
h
e
p
o
s
e
co
o
r
d
in
ates
ar
e
p
r
o
jecte
d
o
n
t
o
a
n
ew
p
lan
e
s
p
atially
with
a
r
an
d
o
m
ly
d
eter
m
in
e
d
p
r
o
jectio
n
ce
n
ter
b
etwe
en
0
a
n
d
1
.
T
h
e
n
ew
p
lan
e
is
d
r
awn
b
y
s
u
b
tr
ac
tin
g
th
e
wid
th
o
f
t
h
e
s
elec
ted
s
id
e
an
d
th
e
v
er
tical
ed
g
e
o
f
ea
c
h
ad
jac
en
t c
o
r
n
e
r
.
2
.
3
.
6
.
Arm
j
o
int
ro
t
a
t
io
n
T
h
e
co
o
r
d
in
ates o
f
th
e
ar
m
jo
i
n
ts
,
wh
ich
in
cl
u
d
e
t
h
e
s
h
o
u
ld
e
r
,
elb
o
w,
an
d
wr
is
t,
ar
e
r
o
tated
b
ased
o
n
th
e
co
r
r
esp
o
n
d
i
n
g
j
o
in
ts
.
I
t
s
h
o
u
ld
b
e
n
o
ted
th
at
th
is
au
g
m
e
n
tatio
n
is
o
n
ly
p
er
f
o
r
m
ed
o
n
t
h
e
ar
m
co
o
r
d
in
ates
an
d
n
o
t
o
n
th
e
p
alm
c
o
o
r
d
in
ates,
wh
ich
m
ea
n
s
th
at
th
e
s
h
ap
e
an
d
p
o
s
itio
n
o
f
th
e
p
al
m
ar
e
n
o
t
ch
a
n
g
ed
.
Au
g
m
en
tatio
n
is
o
n
ly
p
er
f
o
r
m
ed
o
n
th
e
h
an
d
th
at
is
ac
tiv
e
in
th
e
s
ig
n
i
n
g
,
w
h
ich
i
s
in
d
icate
d
b
y
th
e
wr
is
t
b
ein
g
v
is
ib
le
in
th
e
f
r
a
m
e
o
r
n
o
t
in
th
e
r
ea
d
y
p
o
s
itio
n
.
T
h
e
r
o
tatio
n
an
g
le
r
an
g
es
f
r
o
m
2
to
4
d
e
g
r
ee
s
,
with
th
e
ch
an
ce
o
f
ea
ch
jo
in
t
r
o
tatin
g
b
ein
g
3
:
1
0
.
T
h
e
ch
a
n
ce
o
f
r
o
tatin
g
will
b
e
in
cr
ea
s
ed
in
th
e
n
ex
t
tu
r
n
if
n
o
r
o
tatio
n
o
cc
u
r
s
in
th
e
cu
r
r
e
n
t
tu
r
n
.
T
h
e
am
o
u
n
t
o
f
d
ata
af
ter
g
o
in
g
th
r
o
u
g
h
th
e
d
ata
au
g
m
en
tatio
n
p
r
o
ce
s
s
is
3
2
,
4
0
0
v
i
d
eo
s
,
o
r
1
,
2
0
0
p
e
r
s
ig
n
.
An
ex
am
p
le
o
f
th
e
im
p
lem
en
tatio
n
r
esu
lts
o
f
ea
ch
d
ata
au
g
m
e
n
tatio
n
p
r
o
ce
s
s
o
n
o
n
e
f
r
am
e
is
s
h
o
wn
in
Fig
u
r
e
3
.
I
n
Fig
u
r
e
3
,
th
er
e
ar
e
two
lan
d
m
ar
k
s
f
o
r
ea
s
y
o
b
s
er
v
atio
n
.
T
h
e
b
lack
-
c
o
lo
r
e
d
lan
d
m
ar
k
is
th
e
lan
d
m
ar
k
d
etec
ted
b
y
h
o
lis
tic
o
r
th
e
in
itial
lan
d
m
ar
k
b
e
f
o
r
e
g
o
in
g
th
r
o
u
g
h
th
e
au
g
m
e
n
tatio
n
p
r
o
ce
s
s
,
wh
ile
th
e
r
ed
-
co
l
o
r
ed
lan
d
m
ar
k
is
th
e
lan
d
m
ar
k
af
ter
g
o
in
g
th
r
o
u
g
h
th
e
au
g
m
en
tatio
n
p
r
o
ce
s
s
.
Fig
u
r
e
3
.
Data
au
g
m
en
tatio
n
a
p
p
lied
o
n
f
r
am
e
2
.4
.
K
ey
po
int
n
o
rma
liza
t
io
n
Key
p
o
in
t
co
o
r
d
in
ates
h
av
e
v
a
lu
es
r
elativ
e
to
th
e
f
r
am
e,
wh
ich
m
ak
es
p
o
s
itio
n
al
p
r
o
p
e
r
ties
s
u
ch
as
th
e
p
o
s
itio
n
o
f
th
e
p
e
r
f
o
r
m
er
,
d
is
tan
ce
f
r
o
m
th
e
ca
m
er
a
,
an
d
b
o
d
y
p
r
o
p
o
r
tio
n
s
o
f
th
e
s
ig
n
er
af
f
ec
t
th
ese
v
alu
es.
W
ith
o
u
t
n
o
r
m
aliza
tio
n
,
th
e
m
o
d
el
m
ay
lear
n
m
an
y
s
p
atial
f
ea
tu
r
es
th
at
ar
e
n
o
t
r
elate
d
t
o
th
e
s
ig
n
its
elf
.
Key
p
o
in
t n
o
r
m
aliza
tio
n
is
p
er
f
o
r
m
ed
b
y
r
estrictin
g
th
e
lan
d
m
ar
k
s
to
th
eir
r
esp
ec
tiv
e
b
o
u
n
d
in
g
b
o
x
es o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Temp
o
r
a
l c
o
n
vo
lu
tio
n
a
l n
etw
o
r
k
fo
r
I
n
d
o
n
esia
n
w
o
r
d
b
y
w
o
r
d
s
ig
n
la
n
g
u
a
g
e
…
(
S
a
r
a
h
P
u
r
n
a
ma
w
a
ti
)
4087
a
p
r
ed
ete
r
m
in
ed
s
ize.
T
h
e
s
ize
o
f
th
e
b
o
u
n
d
in
g
b
o
x
is
d
eter
m
in
ed
b
y
a
p
p
ly
in
g
t
h
e
"e
ig
h
t
-
h
ea
d
th
e
o
r
y
",
wh
ich
s
tates
th
at
th
e
h
eig
h
t
o
f
a
h
u
m
an
is
eig
h
t
tim
es
th
e
h
ei
g
h
t
o
f
h
is
h
ea
d
.
T
h
is
n
o
r
m
aliza
t
io
n
m
et
h
o
d
ca
n
b
e
ap
p
lied
r
e
g
ar
d
less
o
f
th
e
b
o
d
y
s
tr
u
ctu
r
e
o
f
th
e
p
er
f
o
r
m
er
a
n
d
th
e
d
is
tan
ce
o
f
th
e
p
er
f
o
r
m
e
r
f
r
o
m
th
e
ca
m
e
r
a.
T
h
e
h
ea
d
h
eig
h
t
is
d
ef
in
ed
as
th
e
h
ea
d
u
n
it
,
an
d
its
v
alu
e
i
s
o
b
tain
ed
b
y
h
alv
in
g
th
e
d
is
tan
ce
b
etwe
en
th
e
s
ig
n
er
’
s
s
h
o
u
ld
er
s
.
No
r
m
aliza
tio
n
o
f
th
e
p
o
s
e
lan
d
m
ar
k
is
d
o
n
e
b
y
p
r
o
v
id
in
g
a
b
o
u
n
d
in
g
b
o
x
with
a
s
ize
th
at
co
v
er
s
th
e
en
tire
s
ig
n
in
g
s
p
a
ce
.
Sig
n
in
g
s
p
ac
e
is
th
e
s
p
ac
e
u
s
ed
b
y
t
h
e
s
ig
n
er
d
u
r
in
g
s
ig
n
lan
g
u
ag
e.
Mo
s
t
liter
atu
r
e
[
29
]
,
[
30
]
m
en
tio
n
s
th
at
th
e
s
ig
n
in
g
s
p
ac
e
v
er
tical
ly
s
tar
ts
f
r
o
m
th
e
waist
ar
ea
to
th
e
ar
ea
s
lig
h
tly
ab
o
v
e
th
e
s
ig
n
er
’
s
h
ea
d
an
d
h
o
r
izo
n
tally
ex
te
n
d
s
f
r
o
m
el
b
o
w
to
elb
o
w
wh
en
b
o
th
ar
m
s
ar
e
b
en
t.
I
n
th
is
r
esear
ch
,
th
e
s
ig
n
in
g
s
p
ac
e
is
6
h
ea
d
u
n
its
wid
e
with
th
e
n
o
s
e
co
o
r
d
in
ate
as
its
h
o
r
izo
n
t
al
ce
n
ter
o
r
3
h
ea
d
u
n
its
to
eith
er
h
o
r
iz
o
n
tal
s
id
e
o
f
th
e
n
o
s
e.
T
h
e
v
e
r
tical
p
o
s
itio
n
o
f
th
e
s
p
ac
e
h
as
a
h
eig
h
t o
f
4
.
5
h
ea
d
u
n
its
an
d
u
s
es th
e
n
o
s
e
co
o
r
d
in
ate
as its
ce
n
ter
as we
ll,
with
th
e
to
p
ed
g
e
o
f
th
e
s
p
ac
e
b
ein
g
0
.
5
h
ea
d
u
n
its
awa
y
f
r
o
m
it
an
d
th
e
b
o
tto
m
ed
g
e
b
ei
n
g
4
h
ea
d
u
n
its
b
elo
w
it.
T
h
e
lan
d
m
ar
k
'
s
co
o
r
d
in
ates
will
b
e
b
a
s
ed
o
n
th
e
b
o
u
n
d
i
n
g
b
o
x
'
s
p
o
s
itio
n
.
T
h
e
h
an
d
lan
d
m
a
r
k
s
ar
e
n
o
r
m
alize
d
to
a
b
o
u
n
d
in
g
b
o
x
with
th
e
wid
th
an
d
h
eig
h
t
o
f
1
.
5
h
ea
d
u
n
its
,
with
th
e
m
id
d
le
p
o
s
itio
n
o
f
th
e
h
an
d
as
th
e
ce
n
te
r
.
T
h
e
co
o
r
d
in
ate
v
alu
e
o
f
th
e
h
an
d
ce
n
ter
p
o
s
itio
n
is
o
b
tain
ed
b
y
h
alv
in
g
th
e
d
is
tan
ce
b
etwe
en
t
h
e
m
in
im
u
m
an
d
m
ax
im
u
m
o
f
th
e
an
d
co
o
r
d
i
n
ates.
No
r
m
aliza
tio
n
allo
ws
th
e
m
o
d
el
to
ac
tu
ally
lear
n
th
e
s
h
ap
e
o
f
th
e
h
an
d
,
r
ath
er
th
an
ju
s
t
lear
n
in
g
th
e
s
p
atial
lo
ca
tio
n
o
f
th
e
h
an
d
with
in
t
h
e
en
tire
f
r
am
e.
Ho
wev
er
,
th
e
h
an
d
lo
ca
tio
n
in
f
o
r
m
atio
n
i
n
th
e
f
r
am
e
is
also
im
p
o
r
tan
t
,
a
n
d
th
e
m
o
d
el
ca
n
s
till
lear
n
th
e
h
an
d
lo
ca
tio
n
in
f
o
r
m
atio
n
f
r
o
m
t
h
e
Po
s
e
lan
d
m
ar
k
s
th
at
h
av
e
b
o
th
wr
is
t
lan
d
m
ar
k
s
.
T
h
e
lan
d
m
a
r
k
co
o
r
d
in
ates
th
at
h
av
e
b
ee
n
n
o
r
m
alize
d
in
to
th
eir
r
esp
ec
ti
v
e
b
o
u
n
d
in
g
b
o
x
es
ar
e
th
en
n
o
r
m
alize
d
a
g
ain
in
to
a
d
is
tan
ce
o
f
0
to
1
.
An
ex
am
p
le
o
f
ap
p
l
y
in
g
th
e
k
e
y
p
o
in
t
n
o
r
m
aliza
tio
n
b
o
u
n
d
in
g
b
o
x
t
o
o
n
e
o
f
t
h
e
f
r
a
m
es c
an
b
e
s
ee
n
in
Fig
u
r
e
4
.
Fig
u
r
e
4
.
Vis
u
alize
d
n
o
r
m
aliz
atio
n
ap
p
lied
t
o
a
f
r
a
m
e
2
.
5
.
Da
t
a
prepro
ce
s
s
ing
a
nd
ex
perim
ent
a
l set
up
T
h
e
k
e
y
p
o
i
n
t
s
t
h
at
h
a
v
e
b
e
e
n
ex
t
r
a
c
t
e
d
f
r
o
m
e
a
c
h
f
r
a
m
e
w
il
l
g
o
t
h
r
o
u
g
h
p
r
e
p
r
o
c
e
s
s
i
n
g
b
e
f
o
r
e
t
h
e
y
c
a
n
b
e
u
s
e
d
a
s
i
n
p
u
t
f
o
r
t
h
e
m
o
d
e
l
t
o
b
e
t
r
ai
n
e
d
.
Da
t
a
p
r
e
p
r
o
c
e
s
s
i
n
g
i
n
c
l
u
d
es
d
at
a
p
a
d
d
i
n
g
,
d
a
t
a
l
a
b
e
li
n
g
,
d
a
t
a
s
p
l
i
tt
i
n
g
,
a
n
d
c
r
e
a
ti
n
g
f
e
a
t
u
r
e
a
n
d
l
a
b
e
l
a
r
r
a
y
s
.
F
i
r
s
tl
y
,
d
a
t
a
p
a
d
d
i
n
g
i
s
a
m
u
s
t
,
b
e
c
a
u
s
e
t
h
e
d
a
t
a
s
e
r
i
e
s
u
s
e
d
d
o
n
o
t
h
a
v
e
t
h
e
s
a
m
e
d
a
t
a
s
i
z
e
.
T
h
i
s
i
s
d
u
e
t
o
e
a
c
h
s
i
g
n
h
a
v
i
n
g
a
d
i
f
f
e
r
e
n
t
n
u
m
b
e
r
o
f
f
r
a
m
e
s
.
W
it
h
d
a
t
a
p
a
d
d
i
n
g
,
a
l
l
d
a
t
a
wi
l
l
h
a
v
e
t
h
e
s
a
m
e
d
at
a
s
i
z
e
.
D
at
a
p
a
d
d
i
n
g
i
s
d
o
n
e
b
y
a
d
d
i
n
g
a
N
u
m
P
y
f
i
l
e
c
o
n
t
a
i
n
i
n
g
a
n
a
r
r
a
y
o
f
s
iz
e
(
1
1
8
)
w
i
t
h
a
s
p
e
c
i
a
l
s
y
m
b
o
l
0
.
1
i
n
t
o
e
a
c
h
s
i
g
n
d
i
r
e
ct
o
r
y
u
n
t
i
l
i
t
r
e
a
c
h
es
t
h
e
m
a
x
i
m
u
m
s
i
ze
,
w
h
i
c
h
is
4
4
.
T
h
e
a
r
r
a
y
s
i
z
e
o
f
1
1
8
i
s
b
e
c
a
u
s
e
i
t
i
s
t
h
e
n
u
m
b
e
r
o
f
k
e
y
p
o
i
n
t
s
e
x
t
r
a
c
t
e
d
i
n
o
n
e
f
r
a
m
e
(
i
n
o
n
e
s
e
q
u
e
n
c
e
)
.
D
a
t
a
p
a
d
d
i
n
g
w
i
l
l
b
e
d
o
n
e
p
o
s
t
-
p
a
d
d
i
n
g
,
w
h
i
c
h
is
t
h
e
a
d
d
it
i
o
n
o
f
p
a
d
d
i
n
g
a
t
t
h
e
e
n
d
o
f
t
h
e
d
a
ta
a
n
d
t
h
e
m
a
x
i
m
u
m
s
i
z
e
v
al
u
e
is
t
h
e
s
iz
e
o
f
t
h
e
s
i
g
n
d
a
t
a
t
h
a
t
h
a
s
t
h
e
l
o
n
g
e
s
t
d
a
t
a
s
e
q
u
e
n
c
e
.
S
e
c
o
n
d
s
t
e
p
is
d
a
t
a
l
a
b
el
i
n
g
;
d
a
ta
l
a
b
e
l
i
n
g
is
t
h
e
p
r
o
c
e
s
s
o
f
l
a
b
el
in
g
d
a
t
a
s
o
t
h
at
it
c
a
n
b
e
u
s
e
d
as
a
s
i
g
n
f
o
r
e
a
c
h
t
y
p
e
o
f
d
a
t
a
.
D
a
t
a
l
a
b
e
l
i
n
g
is
d
o
n
e
b
y
g
i
v
i
n
g
a
n
i
d
e
n
t
i
t
y
i
n
t
h
e
f
o
r
m
o
f
a
n
u
m
b
e
r
t
o
e
a
c
h
s
i
g
n
d
i
r
e
c
t
o
r
y
a
n
d
s
t
o
r
i
n
g
it
i
n
a
d
ic
t
i
o
n
a
r
y
.
T
h
e
l
a
b
e
l
n
u
m
b
e
r
s
t
a
r
t
s
f
r
o
m
0
a
n
d
is
a
s
s
i
g
n
e
d
t
o
e
a
c
h
d
i
r
e
c
t
o
r
y
t
h
a
t
h
as
b
e
e
n
s
o
r
t
e
d
a
l
p
h
a
b
e
ti
c
a
ll
y
f
r
o
m
s
m
a
l
l
t
o
l
a
r
g
e
.
T
h
en
we
g
o
th
r
o
u
g
h
d
ata
s
p
litt
in
g
,
d
ata
s
p
litt
in
g
was
d
o
n
e
with
a
s
p
lit
r
atio
o
f
7
0
:3
0
(
tr
ain
in
g
d
ata
:
v
alid
atio
n
d
ata
)
.
Data
s
elec
tio
n
d
u
r
in
g
s
p
lit
d
ata
is
n
o
t
co
m
p
letely
r
an
d
o
m
,
b
u
t
h
as
b
ee
n
s
et
s
o
th
at
ea
ch
p
ar
t
o
f
th
e
d
ata
h
as
th
e
s
am
e
n
u
m
b
er
o
f
h
an
d
v
ar
i
atio
n
s
.
T
h
is
is
d
o
n
e
s
o
th
at
b
o
th
h
an
d
v
ar
iatio
n
s
ca
n
b
e
r
ec
o
g
n
ized
b
y
t
h
e
s
y
s
tem
p
r
o
p
er
ly
.
T
h
e
f
r
a
m
es
th
at
h
av
e
b
ee
n
r
an
d
o
m
ly
s
elec
ted
ar
e
p
u
t
in
to
an
ar
r
a
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
0
8
1
-
4
0
9
5
4088
th
at
h
o
ld
s
ea
ch
d
ata
s
p
lit.
T
h
e
ar
r
ay
is
th
en
r
an
d
o
m
ized
in
o
r
d
er
,
b
e
f
o
r
e
b
ei
n
g
wr
itten
in
to
a
.
tx
t
f
ile.
L
astl
y
,
th
e
s
p
lit
d
ata
wi
ll
b
e
p
u
t
to
g
eth
er
in
to
two
p
air
s
o
f
ar
r
ay
s
th
at
h
o
ld
th
e
f
ea
tu
r
es
an
d
l
ab
els
o
f
ea
ch
d
ata
p
ar
titi
o
n
.
T
h
is
m
ea
n
s
th
at
ea
c
h
d
ata
p
a
r
titi
o
n
h
as
two
ar
r
ay
s
,
o
n
e
ar
r
a
y
co
n
tain
s
d
ata
f
ea
t
u
r
es
an
d
o
n
e
ar
r
ay
co
n
tain
s
d
ata
lab
els.
T
h
ese
ar
r
ay
s
will
b
e
u
s
ed
as
in
p
u
t
wh
en
tr
ain
in
g
th
e
m
o
d
el.
B
ef
o
r
e
c
o
n
tin
u
in
g
to
p
r
esen
t
th
e
m
o
d
el
tr
ain
i
n
g
an
d
test
in
g
r
esu
lts
,
th
e
s
p
ec
if
icatio
n
o
f
s
o
f
twar
e
an
d
h
a
r
d
war
e
u
s
ed
in
th
is
s
tu
d
y
ar
e
as
f
o
llo
ws
:
Py
th
o
n
3
.
9
.
1
0
,
Vis
u
a
l
Stu
d
io
C
o
d
e
v
1
.
7
2
.
0
,
J
u
p
y
te
r
No
teb
o
o
k
Vis
u
al
Stu
d
io
C
o
d
e,
Go
o
g
le
C
o
lab
,
Dr
o
id
C
am
v
6
.
5
.
2
,
M
ed
ia
P
ip
e
v
0
.
8
.
1
1
,
T
e
n
s
o
r
f
lo
w
-
GPU
v
2
.
1
0
.
0
,
O
p
en
C
V
-
p
y
th
o
n
v
4
.
5
.
3
.
5
6
,
g
T
T
S
v
2
.
2
.
4
,
Nu
m
Py
1
.
2
0
.
3
,
an
d
P
lay
s
o
u
n
d
v
1
.
2
.
2
.
W
h
ile
f
o
r
h
ar
d
wa
r
e
we
wer
e
u
s
in
g
s
ev
er
al
lap
to
p
an
d
PC
with
NVI
DI
A
GeFo
r
ce
GT
X
1
0
5
0
,
AM
D
R
ad
eo
n
Ve
g
a
8
Gr
ap
h
ics,
an
d
also
an
A
n
d
r
o
i
d
c
am
e
r
a
f
r
o
m
h
a
n
d
p
h
o
n
e
with
Dr
o
id
C
am
v
6
.
1
9
in
s
id
e.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
T
ra
ini
ng
re
s
ult
B
iLST
M
h
y
p
er
p
ar
am
eter
tu
n
in
g
is
p
er
f
o
r
m
e
d
with
th
e
ai
m
o
f
k
n
o
win
g
th
e
b
est
h
y
p
e
r
p
ar
am
eter
co
m
b
in
atio
n
s
o
th
at
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
ca
n
b
e
o
p
tim
ized
.
Hy
p
er
p
ar
am
eter
tu
n
in
g
f
o
cu
s
es
o
n
th
e
h
y
p
er
p
ar
am
eter
s
o
f
th
e
n
u
m
b
er
o
f
n
eu
r
o
n
u
n
its
o
f
th
e
B
iLST
M
lay
er
,
th
e
b
atch
s
ize
v
al
u
e,
an
d
th
e
v
alu
e
o
f
th
e
d
r
o
p
o
u
t
lay
er
.
T
h
e
d
ata
u
s
ed
as
in
p
u
t
f
o
r
th
e
tr
ai
n
in
g
p
r
o
ce
s
s
is
f
o
u
r
d
ata
ar
r
ay
s
,
two
as
tr
ain
in
g
d
ata
an
d
two
as
v
alid
atio
n
d
ata.
E
ac
h
d
ata
p
o
in
t
h
as
o
n
e
ar
r
ay
h
o
ld
i
n
g
th
e
f
ea
tu
r
es
o
f
th
e
d
ata
an
d
o
n
e
ar
r
ay
h
o
ld
in
g
th
e
lab
el
o
f
th
e
d
ata.
I
n
th
e
t
r
ain
in
g
p
r
o
ce
s
s
,
we
u
tili
ze
th
e
Go
o
g
le
C
o
lab
p
latf
o
r
m
with
a
GPU
b
ec
au
s
e
it
is
co
n
s
id
er
ed
to
h
av
e
f
aster
c
o
m
p
u
tatio
n
co
m
p
ar
ed
to
o
u
r
d
ev
ice.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
is
ca
r
r
ied
o
u
t
f
o
r
5
ep
o
c
h
s
,
an
d
th
e
ch
ec
k
p
o
in
t
with
th
e
h
ig
h
est ac
cu
r
ac
y
is
tak
en
.
T
h
er
e
ar
e
th
r
ee
c
o
m
b
in
atio
n
s
o
f
th
e
n
u
m
b
e
r
o
f
n
e
u
r
o
n
u
n
its
test
ed
,
th
o
s
e
co
m
b
in
atio
n
s
ar
e
6
4
,
1
2
8
,
6
4
;
1
2
8
,
2
5
6
,
1
2
8
;
a
n
d
2
5
6
,
5
1
2
,
2
5
6
.
T
h
e
f
ir
s
t
n
u
m
b
er
r
ep
r
esen
ts
th
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
in
th
e
f
ir
s
t
B
iLST
M
lay
er
,
th
e
s
ec
o
n
d
n
u
m
b
er
r
e
p
r
esen
ts
th
e
n
u
m
b
e
r
o
f
n
eu
r
o
n
s
in
th
e
s
ec
o
n
d
la
y
er
,
an
d
th
e
th
i
r
d
n
u
m
b
er
r
e
p
r
esen
ts
th
e
n
u
m
b
e
r
o
f
n
eu
r
o
n
s
in
th
e
last
lay
er
.
Fo
r
b
atch
s
ize,
test
ed
v
alu
es
a
r
e
3
2
,
6
4
,
an
d
1
2
8
.
Fo
r
th
e
d
r
o
p
o
u
t
v
al
u
e,
t
h
e
v
al
u
es
to
b
e
test
ed
ar
e
0
.
2
an
d
0
.
5
.
T
h
e
m
o
d
el
is
tr
ain
ed
with
all
h
y
p
e
r
p
ar
am
eter
co
m
b
in
atio
n
s
,
wh
ic
h
m
ea
n
s
ei
g
h
teen
tr
ain
in
g
tr
ials
will
b
e
co
n
d
u
cte
d
.
T
h
e
v
alid
atio
n
ac
cu
r
ac
y
v
alu
e
a
n
d
th
e
n
u
m
b
er
o
f
ep
o
ch
s
is
u
s
ed
as
a
r
ef
er
en
ce
in
d
eter
m
in
in
g
th
e
b
est
h
y
p
er
p
ar
a
m
eter
co
m
b
in
at
io
n
.
T
ab
le
1
is
th
e
r
esu
lt
o
f
h
y
p
er
p
ar
a
m
eter
tu
n
in
g
;
th
e
ep
o
ch
co
lu
m
n
d
is
p
lay
s
th
e
ep
o
ch
v
alu
e
wh
en
t
h
e
m
o
d
el
r
ea
ch
es
th
e
b
est
v
alid
atio
n
ac
cu
r
ac
y
.
T
ab
le
1
.
B
iLST
M
h
y
p
e
r
p
ar
am
eter
tu
n
in
g
r
esu
lts
N
u
mb
e
r
o
f
n
e
u
r
o
n
u
n
i
t
s
B
a
t
c
h
si
z
e
D
r
o
p
o
u
t
A
c
c
u
r
a
c
y
(
%)
Ep
o
c
h
(
6
4
,
1
2
8
,
6
4
)
32
0
.
2
9
5
.
0
7
5
(
6
4
,
1
2
8
,
6
4
)
32
0
.
5
7
7
.
0
5
5
(
6
4
,
1
2
8
,
6
4
)
64
0
.
2
9
3
.
0
3
5
(
6
4
,
1
2
8
,
6
4
)
64
0
.
5
7
3
.
0
1
2
(
6
4
,
1
2
8
,
6
4
)
1
2
8
0
.
2
9
2
.
0
9
4
(
6
4
,
1
2
8
,
6
4
)
1
2
8
0
.
5
7
7
.
4
7
5
(
1
2
8
,
2
5
6
,
1
2
8
)
32
0
.
2
9
4
.
8
4
(
1
2
8
,
2
5
6
,
1
2
8
)
32
0
.
5
9
0
.
3
1
4
(
1
2
8
,
2
5
6
,
1
2
8
)
64
0
.
2
9
2
.
6
3
4
(
1
2
8
,
2
5
6
,
1
2
8
)
64
0
.
5
8
6
.
5
2
5
(
1
2
8
,
2
5
6
,
1
2
8
)
1
2
8
0
.
2
8
5
.
7
7
5
(
1
2
8
,
2
5
6
,
1
2
8
)
1
2
8
0
.
5
8
4
.
7
7
5
(
2
5
6
,
5
1
2
,
2
5
6
)
32
0
.
2
9
7
.
4
8
5
(
2
5
6
,
5
1
2
,
2
5
6
)
32
0
.
5
8
8
.
2
5
5
(
2
5
6
,
5
1
2
,
2
5
6
)
64
0
.
2
9
3
.
2
3
5
(
2
5
6
,
5
1
2
,
2
5
6
)
64
0
.
5
9
3
.
1
2
4
(
2
5
6
,
5
1
2
,
2
5
6
)
1
2
8
0
.
2
9
5
.
9
9
5
(
2
5
6
,
5
1
2
,
2
5
6
)
1
2
8
0
.
5
9
1
.
8
9
5
B
ased
o
n
T
a
b
le
1
,
it
is
k
n
o
wn
th
at
t
h
e
m
o
d
el
with
a
c
o
m
b
in
atio
n
o
f
th
e
n
u
m
b
e
r
o
f
n
e
u
r
o
n
s
(
2
5
6
,
5
1
2
,
2
5
6
)
with
a
b
atc
h
s
ize
o
f
3
2
an
d
a
d
r
o
p
o
u
t
v
alu
e
o
f
0
.
2
h
as
th
e
h
ig
h
est
ac
cu
r
ac
y
v
alu
e
o
f
9
7
.
4
8
%
in
5
ep
o
c
h
s
.
T
h
u
s
,
th
is
h
y
p
er
p
ar
am
eter
co
n
f
ig
u
r
atio
n
will
b
e
u
s
ed
.
Fig
u
r
e
5
s
h
o
ws
th
e
ac
cu
r
ac
y
an
d
lo
s
s
g
r
ap
h
o
f
th
e
B
i
L
STM
m
o
d
el,
wh
ile
Fig
u
r
e
6
s
h
o
ws
th
e
ac
cu
r
ac
y
an
d
l
o
s
s
g
r
ap
h
f
o
r
th
e
b
e
s
t
T
C
N
m
o
d
el.
Fo
r
th
e
T
C
N
,
s
ep
ar
ate
tr
ain
in
g
was
co
n
d
u
cted
f
r
o
m
th
e
B
iLST
M
m
o
d
el.
At
th
is
s
tag
e,
T
C
N
h
y
p
er
p
ar
am
ete
r
tu
n
in
g
was
p
er
f
o
r
m
ed
u
s
in
g
Ker
as
t
u
n
e
r
to
f
in
d
th
e
b
es
t
h
y
p
e
r
p
ar
am
ete
r
s
f
o
r
b
u
ild
i
n
g
,
o
p
tim
izin
g
,
an
d
tr
ain
in
g
th
e
T
C
N
m
o
d
el.
T
h
e
p
ar
am
eter
s
test
ed
wer
e
th
e
n
u
m
b
er
o
f
la
y
er
s
,
f
ilter
u
n
its
,
ac
tiv
atio
n
,
d
ilatio
n
r
ate,
d
r
o
p
o
u
t r
ate,
d
en
s
e
u
n
it,
an
d
b
atch
s
ize.
T
esti
n
g
was c
ar
r
ied
o
u
t 3
0
tim
es,
with
ea
ch
ex
p
er
im
en
t h
a
v
in
g
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Temp
o
r
a
l c
o
n
vo
lu
tio
n
a
l n
etw
o
r
k
fo
r
I
n
d
o
n
esia
n
w
o
r
d
b
y
w
o
r
d
s
ig
n
la
n
g
u
a
g
e
…
(
S
a
r
a
h
P
u
r
n
a
ma
w
a
ti
)
4089
m
ax
im
u
m
o
f
1
0
e
p
o
ch
s
.
T
h
e
b
est
p
ar
am
eter
r
esu
lts
an
d
d
etails
o
f
ea
ch
ex
p
er
im
en
t
ca
n
b
e
s
ee
n
in
T
ab
le
2
.
Af
ter
h
y
p
er
p
a
r
am
eter
t
u
n
in
g
,
th
e
h
ig
h
est
ac
c
u
r
ac
y
was
ac
h
iev
ed
with
th
e
co
m
b
in
atio
n
o
f
n
e
u
r
o
n
s
(
6
4
,
6
4
,
1
2
8
)
,
a
d
ilatio
n
o
f
1
,
4
,
an
d
8
,
an
d
a
d
r
o
p
o
u
t r
ate
o
f
0
.
4
; th
er
ef
o
r
e,
th
is
co
m
b
in
atio
n
was im
p
lem
en
ted
in
th
e
m
o
d
el.
Fig
u
r
e
5
.
B
iLST
M
m
o
d
el’
s
ac
cu
r
ac
y
an
d
lo
s
s
Fig
u
r
e
6
.
T
C
N
m
o
d
el’
s
a
cc
u
r
a
cy
an
d
l
o
s
s
3
.
2
.
T
esting
re
s
ult
Mo
d
el
test
in
g
is
d
o
n
e
in
r
ea
l
t
im
e
an
d
ea
ch
wo
r
d
will
b
e
test
ed
3
0
tim
es
with
b
o
th
h
an
d
v
ar
iatio
n
s
,
1
5
tim
es
f
o
r
r
ig
h
t
-
h
an
d
v
ar
iati
o
n
s
an
d
1
5
tim
es
f
o
r
lef
t
-
h
an
d
v
ar
iatio
n
s
.
T
h
is
m
ea
n
s
th
e
s
y
s
tem
will
b
e
test
ed
8
1
0
tim
es.
T
h
e
in
co
m
in
g
f
r
a
m
es
will
g
o
th
r
o
u
g
h
a
s
er
ies
o
f
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
,
s
u
ch
as
f
r
am
e
r
esizin
g
,
f
r
am
e
s
elec
tio
n
,
k
ey
p
o
in
t
ex
t
r
ac
tio
n
,
an
d
k
ey
p
o
in
t
n
o
r
m
aliza
tio
n
.
T
ab
le
3
s
h
o
ws
all
ev
alu
atio
n
m
etr
ic
f
o
r
ea
ch
m
o
d
el,
wh
ile
Fig
u
r
e
7
s
h
o
ws
th
e
co
n
f
u
s
io
n
m
atr
ix
co
m
p
ar
is
o
n
b
etwe
en
B
iLST
M
an
d
T
C
N,
all
in
Me
d
iaPip
e
p
latf
o
r
m
.
W
h
er
e
Fig
u
r
e
7
(
a)
s
h
o
ws
th
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
B
iLST
M
an
d
Fig
u
r
e
7
(
b
)
s
h
o
ws
th
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
T
C
N.
B
ased
o
n
T
ab
le
3
,
it
ca
n
b
e
s
ee
n
th
at
o
u
t
o
f
th
e
8
1
0
d
ata
s
ets
u
s
ed
f
o
r
test
in
g
,
6
9
8
co
r
r
e
ct
d
ata
s
ets
wer
e
g
en
er
ated
f
o
r
th
e
B
iLST
M
m
o
d
el
an
d
7
1
8
co
r
r
ec
t
d
ata
s
ets
f
o
r
th
e
T
C
N.
I
n
o
th
er
wo
r
d
s
,
B
iLST
M
p
r
o
d
u
ce
d
an
ac
cu
r
ac
y
o
f
8
6
.
1
7
%,
wh
ile
T
C
N
o
u
tp
er
f
o
r
m
ed
B
iLST
M
with
an
ac
cu
r
ac
y
o
f
8
8
.
6
4
%.
I
t
is
also
p
o
s
s
ib
le
to
o
b
s
er
v
e
th
at
T
C
N
h
as
9
2
d
ata
p
o
in
ts
,
wh
e
r
ea
s
B
iLST
M
h
as
1
1
2
d
ata
p
o
i
n
ts
,
with
r
eg
a
r
d
to
th
e
d
ata
th
at
f
alls
in
to
t
h
e
f
alse
n
eg
ativ
e
g
r
o
u
p
.
T
h
is
s
u
g
g
ests
th
at
th
e
u
s
e
o
f
B
iLST
M
as
a
m
o
d
el
r
esu
lt
i
n
a
g
r
ea
ter
n
u
m
b
er
o
f
m
is
tak
es
wh
en
co
m
p
ar
ed
to
t
h
e
u
s
e
o
f
T
C
N.
I
n
ad
d
itio
n
,
wh
e
n
we
ex
a
m
in
e
th
e
p
r
ec
is
io
n
v
alu
es
f
o
r
ea
ch
lab
el
th
at
was
ap
p
lied
,
we
n
o
tice
t
h
at
T
C
N
is
m
o
r
e
d
o
m
in
a
n
t
th
an
B
iLST
M
in
alm
o
s
t
al
l
lab
els.
R
ec
all
an
d
F1
-
s
co
r
e
ar
e
also
m
o
r
e
d
o
m
in
an
t t
h
an
B
iLST
M.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
2
0
2
6
:
4
0
8
1
-
4
0
9
5
4090
T
ab
le
2
.
T
C
N
h
y
p
er
p
ar
a
m
eter
tu
n
in
g
r
esu
lts
La
y
e
r
c
o
u
n
t
F
i
l
t
e
r
1
A
c
t
1
D
i
l
1
F
i
l
t
e
r
2
A
c
t
2
D
i
l
2
F
i
l
t
e
r
2
A
c
t
3
D
i
l
3
D
e
n
se
D
r
o
p
o
u
t
A
c
c
u
r
a
c
y
(
%)
3
64
t
a
n
h
1
64
r
e
l
u
4
1
2
8
t
a
n
h
8
2
5
6
0
.
4
9
7
.
8
9
3
32
t
a
n
h
4
5
1
2
r
e
l
u
1
2
5
6
t
a
n
h
2
2
5
6
0
.
4
9
7
.
8
7
3
1
2
8
t
a
n
h
1
64
r
e
l
u
2
64
t
a
n
h
8
2
5
6
0
.
2
9
7
.
7
8
3
32
t
a
n
h
2
64
t
a
n
h
1
5
1
2
t
a
n
h
8
2
5
6
0
.
2
9
7
.
5
3
2
1
2
8
r
e
l
u
8
2
5
6
t
a
n
h
4
-
-
-
5
1
2
0
.
2
9
7
.
4
4
3
64
t
a
n
h
4
1
2
8
t
a
n
h
1
1
2
8
r
e
l
u
4
1
2
8
0
.
3
9
7
.
3
9
3
64
r
e
l
u
8
2
5
6
t
a
n
h
1
64
r
e
l
u
4
2
5
6
0
.
2
9
6
.
7
6
2
1
2
8
r
e
l
u
1
32
r
e
l
u
4
-
-
-
5
1
2
0
.
2
9
6
.
7
5
2
5
1
2
t
a
n
h
1
32
t
a
n
h
4
-
-
8
5
1
2
0
.
3
9
6
.
1
0
2
1
2
8
t
a
n
h
8
32
r
e
l
u
1
-
-
-
2
5
6
0
.
3
9
6
.
0
1
2
64
r
e
l
u
8
64
t
a
n
h
8
-
-
-
1
2
8
0
.
2
9
5
.
4
7
1
5
1
2
r
e
l
u
8
-
-
-
-
-
-
2
5
6
0
.
2
9
5
.
4
4
3
64
t
a
n
h
1
32
r
e
l
u
8
64
t
a
n
h
1
1
2
8
0
.
3
9
5
.
2
5
1
2
5
6
t
a
n
h
8
-
-
-
-
-
-
5
1
2
0
.
2
9
5
.
0
3
1
2
5
6
t
a
n
h
1
-
-
-
-
-
-
1
2
8
0
.
2
9
4
.
9
7
1
1
2
8
r
e
l
u
2
-
-
-
-
-
-
5
1
2
0
.
4
9
4
.
7
6
1
5
1
2
r
e
l
u
2
-
-
-
-
-
-
2
5
6
0
.
3
9
4
.
6
9
1
64
t
a
n
h
8
-
-
-
-
-
-
5
1
2
0
.
4
9
4
.
4
9
3
2
5
6
r
e
l
u
2
32
r
e
l
u
1
32
r
e
l
u
1
1
2
8
0
.
3
9
4
.
2
2
1
5
1
2
r
e
l
u
4
-
-
-
-
-
-
1
2
8
0
.
2
9
4
.
1
4
1
64
t
a
n
h
8
-
-
-
-
-
-
2
5
6
0
.
3
9
4
.
0
6
1
64
t
a
n
h
4
-
-
-
-
-
-
1
2
8
0
.
3
9
3
.
9
3
2
32
t
a
n
h
2
32
t
a
n
h
8
-
-
-
2
5
6
0
.
4
9
3
.
7
6
2
5
1
2
r
e
l
u
1
32
r
e
l
u
4
-
-
-
1
2
8
0
.
4
9
3
.
6
9
1
1
2
8
t
a
n
h
8
-
-
-
-
-
-
1
2
8
0
.
3
9
3
.
6
2
3
1
2
8
r
e
l
u
4
1
2
8
t
a
n
h
1
32
t
a
n
h
2
5
1
2
0
.
3
9
3
.
5
4
1
64
r
e
l
u
4
-
-
-
-
-
-
5
1
2
0
.
4
9
3
.
4
3
1
32
r
e
l
u
8
-
-
-
-
-
-
5
1
2
0
.
2
9
2
.
5
1
1
64
r
e
l
u
2
-
-
-
-
-
-
1
2
8
0
.
2
9
2
.
1
4
1
32
t
a
n
h
8
-
-
-
-
-
-
1
2
8
0
.
4
9
1
.
6
7
T
ab
le
3
.
B
iLST
M
an
d
T
C
N
co
m
p
ar
is
o
n
r
esu
lts
Lab
e
l
BiL
S
TM
TCN
TP
FN
FP
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
TP
FN
FP
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
W
h
a
t
26
4
11
0
.
7
0
0
.
8
7
0
.
7
8
26
4
3
0
.
9
0
0
.
8
7
0
.
8
8
H
o
w
19
11
4
0
.
8
3
0
.
6
3
0
.
7
2
27
3
4
0
.
8
7
0
.
9
0
0
.
8
8
S
i
r
29
1
9
0
.
7
6
0
.
9
7
0
.
8
5
30
0
6
0
.
8
3
1
.
0
0
0
.
9
1
H
o
w
ma
n
y
23
7
5
0
.
8
2
0
.
7
7
0
.
7
9
25
5
9
0
.
7
3
0
.
8
3
0
.
7
8
To
m
o
r
r
o
w
27
3
0
1
.
0
0
0
.
9
0
0
.
9
5
27
3
0
1
.
0
0
0
.
9
0
0
.
9
5
H
e
/
S
h
e
30
0
3
0
.
9
1
1
.
0
0
0
.
9
5
30
0
5
0
.
8
6
1
.
0
0
0
.
9
2
D
a
y
28
2
14
0
.
6
7
0
.
9
3
0
.
7
8
28
2
9
0
.
7
6
0
.
9
3
0
.
8
4
To
d
a
y
30
0
3
0
.
9
1
1
.
0
0
0
.
9
5
30
0
3
0
.
9
1
1
.
0
0
0
.
9
5
Y
o
u
25
5
15
0
.
6
2
0
.
8
3
0
.
7
1
27
3
10
0
.
7
3
0
.
9
0
0
.
8
1
W
h
e
n
23
7
2
0
.
9
2
0
.
7
7
0
.
8
4
21
9
0
1
.
0
0
0
.
7
0
0
.
8
2
W
h
e
r
e
21
9
2
0
.
9
1
0
.
7
0
0
.
7
9
22
8
1
0
.
9
6
0
.
7
3
0
.
8
3
Y
e
st
e
r
d
a
y
28
2
5
0
.
8
5
0
.
9
3
0
.
8
9
27
3
5
0
.
8
4
0
.
9
0
0
.
8
7
W
h
y
24
6
1
0
.
9
6
0
.
8
0
0
.
8
7
24
6
0
1
.
0
0
0
.
8
0
0
.
8
9
We
21
9
0
1
.
0
0
0
.
7
0
0
.
8
2
21
9
0
1
.
0
0
0
.
7
0
0
.
8
2
S
o
r
r
y
24
6
1
0
.
9
6
0
.
8
0
0
.
8
7
25
5
1
0
.
9
6
0
.
8
3
0
.
8
9
Ea
t
28
2
5
0
.
8
5
0
.
9
3
0
.
8
9
28
2
6
0
.
8
2
0
.
9
3
0
.
8
8
N
a
me
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
N
u
mb
e
r
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
I
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
30
0
0
1
.
0
0
1
.
0
0
1
.
0
0
N
o
w
27
3
0
1
.
0
0
0
.
9
0
0
.
9
5
27
3
0
1
.
0
0
0
.
9
0
0
.
9
5
G
o
o
d
n
i
g
h
t
21
9
0
1
.
0
0
0
.
7
0
0
.
8
2
21
9
2
0
.
9
1
0
.
7
0
0
.
8
0
G
o
o
d
m
o
r
n
i
n
g
27
3
19
0
.
5
9
0
.
9
0
0
.
7
1
27
3
13
0
.
6
8
0
.
9
0
0
.
7
7
G
o
o
d
a
f
t
e
r
n
o
o
n
25
5
4
0
.
8
6
0
.
8
3
0
.
8
5
25
5
4
0
.
8
6
0
.
8
3
0
.
8
5
W
h
o
22
8
0
1
.
0
0
0
.
7
3
0
.
8
5
26
4
0
1
.
0
0
0
.
8
7
0
.
9
3
P
h
o
n
e
30
0
3
0
.
9
1
1
.
0
0
0
.
9
5
30
0
3
0
.
9
1
1
.
0
0
0
.
9
5
Th
a
n
k
y
o
u
21
9
6
0
.
7
8
0
.
7
0
0
.
7
4
25
5
8
0
.
7
6
0
.
8
3
0
.
8
0
D
e
a
f
29
1
0
1
.
0
0
0
.
9
7
0
.
9
8
29
1
0
1
.
0
0
0
.
9
7
0
.
9
8
T
o
t
a
l
6
9
8
1
1
2
1
1
2
-
-
-
7
1
8
92
92
Me
d
iaPip
e
was
also
test
ed
a
t
v
ar
io
u
s
ca
m
er
a
d
is
tan
ce
s
,
with
th
e
test
ed
d
is
tan
ce
s
b
ein
g
5
0
cm
,
1
0
0
c
m
,
a
n
d
1
5
0
cm
.
Me
d
i
aPip
e's
ca
p
ab
ilit
y
f
o
r
in
itial
d
etec
tio
n
o
f
f
er
s
a
f
u
n
d
am
en
t
al
d
if
f
er
e
n
ce
[
1
7
]
,
p
ar
ticu
lar
ly
co
m
p
a
r
ed
to
f
r
am
ewo
r
k
s
th
at
p
er
f
o
r
m
in
itial
an
n
o
tatio
n
,
s
u
ch
as
y
o
u
o
n
ly
lo
o
k
o
n
ce
(
YOL
O
)
[
1
]
.
T
h
e
s
y
s
tem
'
s
te
s
t
r
esu
lts
f
o
r
th
ese
v
ar
io
u
s
d
is
tan
ce
s
ca
n
b
e
s
ee
n
in
T
ab
le
4
.
Me
d
iaPip
e
w
as
ab
le
to
d
etec
t
a
t
all
test
ed
d
is
tan
ce
s
,
b
u
t
it
s
p
er
f
o
r
m
a
n
ce
b
eg
a
n
to
d
ec
lin
e
at
1
5
0
cm
.
At
th
is
d
is
tan
ce
,
d
etec
tio
n
b
ec
o
m
es
Evaluation Warning : The document was created with Spire.PDF for Python.