I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
6
,
Decem
b
er
20
25
,
p
p
.
5
7
5
9
~
5
7
6
9
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
7
5
9
-
5
7
6
9
5759
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Fas
ter
R
-
CN
N
im
plementa
tion
f
o
r
ha
nd sig
n recog
ni
tion o
f
t
he
Indo
nesia
n
sig
n lang
ua
g
e sy
stem
(SIBI
)
P
a
ulu
s
L
e
s
t
y
o
Adhi
a
t
m
a
,
Nu
rc
a
hy
a
P
ra
da
na
T
a
ufi
k
P
ra
k
is
y
a
,
Ro
s
iha
n Ar
iy
ua
na
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
m
a
t
i
c
s
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
E
d
u
c
a
t
i
o
n
,
F
a
c
u
l
t
y
o
f
Te
a
c
h
e
r
Tr
a
i
n
i
n
g
a
n
d
E
d
u
c
a
t
i
o
n
,
S
e
b
e
l
a
s
M
a
r
e
t
U
n
i
v
e
r
si
t
y
,
S
u
r
a
k
a
r
t
a
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
2
8
,
2
0
2
4
R
ev
is
ed
Au
g
6
,
2
0
2
5
Acc
ep
ted
Sep
1
6
,
2
0
2
5
Th
e
In
d
o
n
e
sia
n
si
g
n
lan
g
u
a
g
e
sy
s
tem
(S
IBI)
is
th
e
a
u
th
o
rize
d
si
g
n
sy
ste
m
in
In
d
o
n
e
sia
th
a
t
th
e
d
e
a
f
so
c
iety
u
se
s
to
c
o
n
v
e
y
i
n
I
n
d
o
n
e
sia
n
.
Ho
we
v
e
r,
it
s
u
se
stil
l
n
e
e
d
s
t
o
b
e
e
x
p
a
n
d
e
d
a
n
d
m
o
re
wi
d
e
sp
re
a
d
i
n
t
h
e
c
o
m
m
u
n
it
y
,
c
a
u
sin
g
d
iff
icu
lt
ies
in
c
o
m
m
u
n
ica
ti
o
n
fo
r
h
a
rd
-
of
-
h
e
a
rin
g
p
e
o
p
le.
Th
e
p
ro
d
u
c
t
o
f
d
e
e
p
lea
rn
in
g
tec
h
n
o
lo
g
ies
su
c
h
a
s
fa
ste
r
re
g
io
n
-
b
a
se
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(F
a
ste
r
R
-
CNN
)
in
o
b
jec
t
re
c
o
g
n
it
i
o
n
h
a
s
th
e
p
o
ten
ti
a
l
t
o
h
e
l
p
im
p
ro
v
e
c
o
m
m
u
n
ica
ti
o
n
b
e
twe
e
n
d
e
a
f
p
e
o
p
l
e
a
n
d
t
h
e
g
e
n
e
ra
l
p
u
b
li
c
.
T
h
is
re
se
a
rc
h
will
imp
lem
e
n
t
th
e
F
a
ste
r
R
-
CNN
a
lg
o
rit
h
m
with
th
re
e
d
iffere
n
t
re
sid
u
a
l
n
e
t
wo
rk
(Re
sN
e
t)
a
rc
h
it
e
c
tu
re
s
(5
0
,
1
0
1
,
a
n
d
1
5
2
)
fo
r
S
IBI
re
c
o
g
n
it
i
o
n
.
T
h
e
c
o
m
p
a
riso
n
o
f
th
e
fa
ste
r
R
-
CNN
a
lg
o
rit
h
m
with
d
iffere
n
t
a
rc
h
it
e
c
tu
re
s
is
a
ls
o
c
o
n
d
u
c
ted
t
o
id
e
n
t
ify
th
e
b
e
st
a
rc
h
it
e
c
tu
re
fo
r
S
IBI
re
c
o
g
n
it
io
n
,
a
n
d
th
e
re
su
lt
s
a
re
e
v
a
lu
a
ted
u
sin
g
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
-
sc
o
re
m
e
tri
c
s
fr
o
m
c
o
n
fu
si
o
n
m
a
tri
x
c
a
lcu
latio
n
a
n
d
e
x
e
c
u
ti
o
n
ti
m
e
.
F
a
ste
r
R
-
CNN
m
o
d
e
l
with
Re
sN
e
t
-
5
0
a
rc
h
it
e
c
tu
re
sh
o
we
d
t
h
e
b
e
st
a
n
d
m
o
st
e
fficie
n
t
p
e
rfo
rm
a
n
c
e
with
a
c
c
u
ra
c
y
,
re
c
a
ll
,
p
re
c
isio
n
,
a
n
d
F
1
-
sc
o
re
m
e
tri
c
s
o
f
9
6
.
1
5
%
,
9
5
%
,
9
3
%
,
a
n
d
9
4
%
,
re
sp
e
c
ti
v
e
l
y
,
a
n
d
a
n
e
x
e
c
u
ti
o
n
ti
m
e
o
f
3
6
.
8
4
se
c
o
n
d
s in
t
h
e
tes
ti
n
g
p
r
o
c
e
ss
c
o
m
p
a
re
d
to
m
o
d
e
ls wi
th
R
e
sN
e
t
-
101
a
n
d
Re
sN
e
t
-
1
5
2
a
rc
h
it
e
c
tu
re
s
.
K
ey
w
o
r
d
s
:
Fas
ter
R
-
C
NN
Ob
ject
r
ec
o
g
n
itio
n
R
esNet
SIBI
Sig
n
lan
g
u
a
g
e
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Nu
r
ca
h
y
a
Pra
d
a
n
a
T
au
f
ik
Pra
k
is
y
a
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics a
n
d
C
o
m
p
u
ter
E
n
g
in
ee
r
in
g
E
d
u
c
atio
n
,
Facu
lty
o
f
T
ea
ch
er
T
r
ai
n
in
g
an
d
E
d
u
ca
tio
n
,
U
n
iv
er
s
itas
Seb
elas M
ar
et
I
r
.
Su
tam
i Str
ee
t 3
6
Ke
n
tin
g
an
,
J
eb
r
es,
Su
r
ak
ar
ta,
I
n
d
o
n
esia
E
m
ail: n
u
r
ca
h
y
a.
p
tp
@
s
taf
f
.
u
n
s
.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
Gestu
r
es
an
d
s
ig
n
s
ar
e
co
n
s
id
er
ed
th
e
co
n
s
id
er
ab
le
n
ativ
e
way
to
o
b
tain
m
ess
ag
es
b
etw
ee
n
p
eo
p
le
th
r
o
u
g
h
b
o
d
y
m
o
tio
n
s
.
Alth
o
u
g
h
g
estu
r
es
a
n
d
s
ig
n
s
ar
e
ca
teg
o
r
ize
d
as
n
o
n
-
v
er
b
al
co
n
tact,
th
ey
ca
n
ef
f
ec
tiv
ely
ex
p
r
ess
m
ess
ag
es
b
etwe
en
h
ar
d
-
of
-
h
e
ar
in
g
p
e
o
p
le.
Sig
n
lan
g
u
ag
e
is
th
e
m
o
s
t
co
m
m
o
n
ly
u
s
ed
m
eth
o
d
o
f
co
n
v
ey
in
g
wo
r
d
s
u
s
in
g
b
o
d
y
m
o
v
em
en
ts
[
1
]
.
Sig
n
lan
g
u
ag
e
is
v
ital
in
c
o
m
m
u
n
icatio
n
f
o
r
th
e
s
p
ee
ch
-
im
p
air
ed
a
n
d
d
ea
f
co
m
m
u
n
ity
[
2
]
.
I
n
d
o
n
esian
s
ig
n
lan
g
u
ag
e
s
y
s
tem
(
SIBI)
is
I
n
d
o
n
esia'
s
o
f
f
icial
s
ig
n
lan
g
u
a
g
e,
s
tan
d
a
r
d
ized
with
s
p
ec
if
ic
f
in
g
er
an
d
h
a
n
d
m
o
v
em
en
ts
th
at
f
o
llo
w
th
e
g
r
am
m
ar
an
d
s
tr
u
ctu
r
e
o
f
th
e
I
n
d
o
n
esian
lan
g
u
ag
e,
as
r
ec
o
g
n
ized
in
Min
is
ter
ial
Dec
r
ee
Nu
m
b
er
0
1
6
1
/U/1
9
9
4
[
3
]
.
Alth
o
u
g
h
SIBI
is
th
e
au
th
o
r
iz
ed
s
ig
n
lan
g
u
ag
e
in
I
n
d
o
n
esia,
its
u
s
e
i
s
s
till
lim
i
ted
an
d
h
as
y
et
to
b
e
wid
ely
s
p
r
ea
d
in
th
e
co
m
m
u
n
ity
.
No
t
all
I
n
d
o
n
esian
s
u
n
d
er
s
tan
d
th
e
s
ig
n
lan
g
u
a
g
e
u
s
ed
b
y
d
ea
f
p
eo
p
le,
s
o
it
will
ca
u
s
e
d
if
f
icu
lties
in
co
m
m
u
n
icatio
n
f
o
r
d
ea
f
p
eo
p
le
wh
e
n
in
ter
ac
tin
g
with
p
eo
p
le
wh
o
d
o
n
o
t
u
n
d
er
s
tan
d
s
ig
n
lan
g
u
ag
e
[
4
]
.
Ma
n
y
p
e
o
p
le
f
in
d
it
d
if
f
icu
lt
to
in
ter
ac
t
an
d
r
elate
to
d
ea
f
p
eo
p
le,
s
o
th
ey
d
o
n
o
t
g
et
r
eg
u
l
ar
s
o
cial
in
ter
ac
tio
n
s
b
ec
au
s
e
o
f
th
eir
lim
itatio
n
s
,
r
esu
ltin
g
in
s
o
cial
in
eq
u
ality
[
5
]
.
T
o
ad
d
r
ess
th
ese
co
m
m
u
n
icatio
n
b
a
r
r
ier
s
an
d
p
r
o
m
o
te
s
o
cial
in
clu
s
io
n
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
7
5
9
-
5
7
6
9
5760
tech
n
o
lo
g
ical
in
n
o
v
atio
n
s
in
s
ig
n
lan
g
u
a
g
e
r
ec
o
g
n
itio
n
ar
e
cr
itical.
R
ec
en
t
ad
v
a
n
ce
m
en
t
s
in
d
ee
p
lear
n
in
g
,
p
ar
ticu
lar
ly
in
c
o
m
p
u
ter
v
is
i
o
n
,
o
f
f
er
p
r
o
m
is
in
g
s
o
l
u
tio
n
s
to
au
to
m
ate
SIBI
r
ec
o
g
n
itio
n
an
d
b
r
id
g
e
th
e
co
m
m
u
n
icatio
n
g
ap
[
6
]
.
Un
lik
e
is
o
lated
g
estu
r
e
r
ec
o
g
n
itio
n
,
SIBI
r
eq
u
i
r
es
p
r
ec
is
e
d
etec
tio
n
o
f
d
y
n
am
ic
h
an
d
c
o
n
f
ig
u
r
atio
n
s
an
d
s
p
atial
r
elatio
n
s
h
ip
s
.
Fas
t
er
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Fas
ter
R
-
C
N
N)
,
with
its
ab
ilit
y
to
lo
ca
lize
an
d
class
if
y
o
b
jec
ts
in
co
m
p
lex
s
ce
n
es,
i
s
p
ar
ti
cu
lar
ly
s
u
ited
f
o
r
th
is
task
.
T
h
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
is
a
d
ee
p
lear
n
in
g
alg
o
r
ith
m
f
o
r
o
b
ject
d
etec
tio
n
alg
o
r
ith
m
th
at
co
m
b
i
n
es
a
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
to
d
etec
t
r
e
g
io
n
p
r
o
p
o
s
als
an
d
class
if
y
o
b
jects
s
im
u
ltan
eo
u
s
ly
b
y
u
t
ilizin
g
th
e
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
(
R
PN)
,
wh
ich
en
ab
les
f
aster
an
d
m
o
r
e
ac
cu
r
ate
p
r
o
ce
s
s
in
g
tim
e
[
7
]
.
Sev
er
al
p
r
ev
io
u
s
s
tu
d
ies
o
n
id
en
tific
atio
n
u
s
in
g
th
e
Fas
ter
R
-
C
NN
a
lg
o
r
ith
m
h
av
e
b
ee
n
co
n
d
u
cte
d
.
I
n
th
e
s
tu
d
y
b
y
Den
g
et
a
l.
[
8
]
,
Fas
ter
-
R
C
NN
was
ap
p
lie
d
to
d
etec
t
d
iab
etic
r
etin
al
d
is
ea
s
e
u
s
in
g
R
es
Net5
0
an
d
VGG1
6
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
.
T
h
e
r
esear
ch
d
e
m
o
n
s
tr
ated
th
at
th
e
R
esNet5
0
-
b
ased
m
o
d
el
ac
h
iev
e
d
a
h
ig
h
er
m
ea
n
a
v
er
ag
e
p
r
ec
is
io
n
(
m
AP)
o
f
9
7
.
4
2
%
a
n
d
a
p
r
ec
is
io
n
o
f
9
8
.
9
6
%.
An
o
th
er
r
esear
c
h
was
co
n
d
u
cted
b
y
Sab
ir
et
a
l.
[
9
]
u
s
in
g
a
tr
an
s
f
er
lear
n
i
n
g
s
tr
ate
g
y
with
th
e
Fas
ter
R
-
C
NN
m
o
d
el
to
d
etec
t
f
ac
es
th
at
u
s
e
m
a
s
k
s
an
d
th
o
s
e
th
at
d
o
n
o
t
u
s
e
m
ask
s
o
b
tain
e
d
th
e
m
o
s
t
elev
ated
av
e
r
ag
e
p
r
ec
i
s
io
n
(
AP)
o
f
8
1
%
an
d
th
e
m
o
s
t
elev
ated
av
er
ag
e
r
ec
all
(
AR
)
o
f
8
4
%.
R
esear
ch
b
y
C
ao
et
a
l.
[
1
0
]
p
r
esen
ts
an
im
p
r
o
v
e
d
alg
o
r
ith
m
b
ased
o
n
Fas
ter
R
-
C
N
N
f
o
r
s
m
all
o
b
ject
d
etec
tio
n
,
ad
d
r
ess
in
g
ch
allen
g
es
s
u
ch
as
co
m
p
lex
b
ac
k
g
r
o
u
n
d
s
,
o
cc
lu
s
io
n
,
a
n
d
lo
w
r
eso
lu
tio
n
,
ac
h
iev
in
g
a
r
ec
all
r
ate
o
f
9
0
%
an
d
an
ac
c
u
r
ac
y
r
ate
o
f
8
7
% f
o
r
tr
af
f
ic
s
ig
n
s
.
B
ased
o
n
t
h
ese
s
tu
d
ies,
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
h
as e
x
ce
llen
t p
o
ten
tial in
o
b
ject
r
ec
o
g
n
iti
o
n
.
Selectin
g
th
e
s
u
itab
le
alg
o
r
it
h
m
is
cr
u
cial
f
o
r
im
p
lem
e
n
tin
g
Fas
ter
R
-
C
NN
in
r
ec
o
g
n
izin
g
th
e
I
n
d
o
n
esian
lan
g
u
ag
e
s
y
s
tem
(
SIBI)
.
T
h
e
r
esid
u
al
n
etw
o
r
k
(
R
esNet)
C
NN
ar
ch
itectu
r
e,
in
t
r
o
d
u
ce
d
b
y
He
et
a
l.
[
1
1
]
,
h
as
p
r
o
v
en
ef
f
ec
tiv
e
in
o
v
er
co
m
i
n
g
th
e
p
er
f
o
r
m
an
ce
d
e
g
r
ad
atio
n
p
r
o
b
lem
in
d
ee
p
er
C
NNs.
R
esNet
ca
n
o
v
er
co
m
e
th
e
ch
al
len
g
e
o
f
lear
n
in
g
c
o
m
p
lex
a
n
d
d
ee
p
r
ep
r
esen
tatio
n
s
u
s
in
g
s
h
o
r
tcu
t c
o
n
n
ec
tio
n
s
.
Fre
q
u
en
tly
u
s
ed
v
ar
ian
ts
o
f
t
h
e
R
esNet
ar
ch
itectu
r
e
a
r
e
R
esNet
-
5
0
,
1
0
1
,
a
n
d
1
5
2
,
ea
c
h
o
f
wh
ich
h
as
an
ap
p
r
o
p
r
iate
n
u
m
b
er
o
f
lay
er
s
to
h
an
d
le
d
if
f
er
e
n
t
lev
els
o
f
c
o
m
p
lex
ity
.
T
h
e
R
esNet
ar
ch
itectu
r
e
in
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
s
er
v
es
as
th
e
b
ac
k
b
o
n
e
in
p
er
f
o
r
m
in
g
d
ig
ital
im
a
g
e
f
ea
tu
r
e
ex
tr
ac
tio
n
[
1
2
]
.
R
esNet
ar
ch
itectu
r
e
as
a
b
ac
k
b
o
n
e/f
o
u
n
d
atio
n
h
as
b
ee
n
wid
ely
u
s
e
d
in
co
m
p
le
x
task
s
,
s
u
ch
as
o
b
ject
d
etec
tio
n
an
d
in
s
tan
ce
s
eg
m
en
tatio
n
[
1
3
]
.
I
n
th
is
s
tu
d
y
,
we
em
p
lo
y
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
d
u
e
to
its
s
tr
o
n
g
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
in
o
b
ject
d
etec
tio
n
task
s
,
p
ar
ticu
lar
ly
wh
en
d
ea
lin
g
with
d
etailed
s
p
atial
f
ea
tu
r
es
s
u
ch
as
h
an
d
g
estu
r
es.
C
o
m
p
ar
e
d
to
o
t
h
er
d
etec
tio
n
m
o
d
els
lik
e
YOL
O
a
n
d
SS
D,
wh
ic
h
p
r
io
r
itize
i
n
f
e
r
en
ce
s
p
ee
d
,
Fas
ter
R
-
C
NN
i
s
m
o
r
e
s
u
itab
le
f
o
r
s
ce
n
ar
io
s
th
at
d
em
a
n
d
h
i
g
h
p
r
ec
is
io
n
.
T
h
is
m
ak
es
it
id
ea
l
f
o
r
r
ec
o
g
n
izin
g
f
in
e
-
g
r
ain
ed
h
a
n
d
s
ig
n
v
ar
iatio
n
s
in
SIBI,
wh
er
e
d
etec
tio
n
ac
cu
r
ac
y
is
p
ar
am
o
u
n
t.
Desp
ite
its
s
lo
wer
in
f
er
en
ce
s
p
ee
d
,
Fas
ter
R
-
C
NN
d
em
o
n
s
tr
ates
s
u
p
er
io
r
ac
cu
r
ac
y
,
m
ak
in
g
it
id
ea
l
f
o
r
ap
p
licatio
n
s
wh
er
e
d
etec
tio
n
p
r
ec
is
io
n
is
cr
u
cial
[
1
4
]
.
W
ith
in
th
is
f
r
am
ewo
r
k
,
we
in
t
eg
r
ate
th
r
ee
v
a
r
ian
ts
o
f
th
e
R
esNet
b
ac
k
b
o
n
e
R
esNet
-
5
0
,
R
esNet
-
1
0
1
,
an
d
R
esNet
-
1
5
2
an
d
c
o
n
d
u
ct
a
s
tr
u
ctu
r
ed
c
o
m
p
a
r
ativ
e
an
aly
s
is
to
r
ec
o
g
n
ize
m
u
lti
-
class
SIBI
h
an
d
s
ig
n
s
.
T
h
e
n
o
v
elty
o
f
th
is
r
esear
ch
is
r
ef
lect
ed
in
ev
alu
atin
g
t
h
e
tr
a
d
e
-
o
f
f
s
b
etwe
en
d
etec
tio
n
ac
cu
r
ac
y
,
in
f
er
en
ce
tim
e,
a
n
d
m
o
d
el
co
m
p
lex
ity
u
n
d
er
lim
ited
d
ata
co
n
d
itio
n
s
,
p
r
o
v
i
d
in
g
p
r
ac
tical
in
s
ig
h
ts
f
o
r
s
elec
tin
g
ef
f
icien
t
b
ac
k
b
o
n
e
ar
ch
itectu
r
es f
o
r
r
ea
l
-
wo
r
l
d
ass
is
tiv
e
tech
n
o
lo
g
y
ap
p
licatio
n
s
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
e
s
t
u
d
y
m
e
t
h
o
d
e
m
p
l
o
y
e
d
i
s
t
h
e
r
e
s
e
a
r
c
h
a
n
d
d
e
v
e
l
o
p
m
e
n
t
(
R
&
D
)
m
e
t
h
o
d
a
s
i
l
l
u
s
t
r
a
te
d
i
n
F
i
g
u
r
e
1
.
T
h
is
r
esear
ch
im
p
lem
en
ts
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
to
id
en
tify
SIBI
s
ig
n
lan
g
u
a
g
e
in
d
ig
ital
im
ag
es.
T
h
e
im
p
lem
en
tatio
n
r
esu
lts
will
th
en
b
e
co
m
p
ar
e
d
to
d
eter
m
i
n
e
th
e
b
est
ar
ch
itectu
r
e
as
a
b
ac
k
b
o
n
e
f
o
r
Fas
ter
R
-
C
NN.
T
h
e
r
esear
ch
p
r
o
ce
s
s
b
eg
in
s
with
d
ata
c
o
llectio
n
,
w
h
er
e
im
ag
es
o
f
SIBI
s
ig
n
s
ar
e
g
ath
er
ed
,
f
o
llo
wed
by
o
b
ject
a
n
n
o
tatio
n
to
lab
el
th
e
r
eg
i
o
n
s
o
f
in
ter
est
with
in
ea
ch
im
ag
e.
T
h
e
an
n
o
tated
d
ata
u
n
d
e
r
g
o
es
d
ata
p
r
ep
r
o
ce
s
s
in
g
to
s
tan
d
ar
d
ize
an
d
au
g
m
e
n
t
th
e
im
ag
es,
en
s
u
r
in
g
th
eir
s
u
itab
ilit
y
f
o
r
th
e
tr
a
in
in
g
m
o
d
el
p
h
ase,
wh
er
e
th
e
Fas
ter
R
-
C
NN
is
tr
ain
ed
to
r
ec
o
g
n
ize
SIBI
s
ig
n
s
.
Fin
ally
,
in
th
e
ev
al
u
ate
p
h
ase,
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
is
ass
es
s
ed
th
r
o
u
g
h
v
ar
i
o
u
s
m
etr
ics.
T
h
e
s
o
u
r
ce
co
d
e
f
o
r
th
is
wo
r
k
is
p
u
b
licly
av
ailab
le
at:
h
ttp
s
:
//g
ith
u
b
.
co
m/p
a
u
l
-
lest
yo
/fa
s
ter
-
r
cn
n
-
s
ib
i
-
h
a
n
d
s
ig
n
.
Fig
u
r
e
1
.
R
esear
ch
f
lo
w
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
s
ter R
-
C
N
N
imp
leme
n
ta
tio
n
fo
r
h
a
n
d
s
ig
n
r
ec
o
g
n
itio
n
o
f
th
e
…
(
P
a
u
l
u
s
Les
tyo
A
d
h
ia
tma
)
5761
2
.
1
.
Da
t
a
s
et
s
T
h
e
d
ataset
u
s
ed
in
th
is
r
esear
ch
co
n
s
is
ts
o
f
5
4
6
la
b
eled
d
i
g
ital
im
ag
es
r
ep
r
esen
tin
g
2
6
h
an
d
s
ig
n
class
es
(
le
tter
s
A
to
Z
)
f
r
o
m
th
e
I
n
d
o
n
esian
Sig
n
L
an
g
u
a
g
e
Sy
s
tem
(
SIBI)
,
with
2
1
im
ag
es
p
er
class
,
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
is
d
atas
et
was
o
b
tain
ed
f
r
o
m
a
n
o
p
en
-
s
o
u
r
ce
c
o
llectio
n
o
n
Ka
g
g
le
[
1
5
]
.
All
im
ag
es
ar
e
in
J
PG
f
o
r
m
at
with
a
u
n
if
o
r
m
s
q
u
ar
e
r
eso
lu
tio
n
o
f
2
0
0
0
×2
0
0
0
p
ix
els.
E
ac
h
im
ag
e
was
m
an
u
ally
a
n
n
o
tated
u
s
in
g
th
e
L
ab
elI
m
g
to
o
l,
p
r
o
d
u
cin
g
XM
L
f
iles
th
at
co
n
tai
n
b
o
u
n
d
in
g
b
o
x
co
o
r
d
in
ates
a
n
d
ass
o
ciate
d
class
lab
els.
T
h
e
an
n
o
tatio
n
f
iles
wer
e
th
en
co
n
v
er
ted
t
o
C
SV
f
o
r
m
at
an
d
s
u
b
s
eq
u
e
n
tly
in
to
T
FR
ec
o
r
d
f
o
r
m
at
to
p
r
ep
ar
e
th
e
d
ataset
f
o
r
T
e
n
s
o
r
Flo
w’
s
o
b
ject
d
etec
tio
n
p
ip
elin
e.
T
o
in
tr
o
d
u
ce
s
li
g
h
t
v
ar
iatio
n
,
d
ata
au
g
m
en
tatio
n
was
ap
p
lied
d
u
r
in
g
tr
ain
i
n
g
,
in
cl
u
d
in
g
h
o
r
iz
o
n
tal
f
lip
p
in
g
an
d
b
r
ig
h
tn
ess
ad
ju
s
tm
en
t.
T
h
ese
tech
n
iq
u
es
p
r
o
d
u
ce
d
v
a
r
ian
ts
im
ag
e
an
d
im
p
r
o
v
ed
m
o
d
el
r
o
b
u
s
tn
ess
b
y
s
im
u
latin
g
s
lig
h
t
en
v
i
r
o
n
m
e
n
tal
ch
an
g
es.
Ho
we
v
er
,
t
h
e
d
ataset
p
r
esen
ts
lim
itatio
n
s
in
d
iv
er
s
ity
an
d
r
ep
r
esen
tativ
en
ess
.
Mo
s
t
o
f
th
e
im
ag
es
f
ea
tu
r
e
s
im
ilar
lig
h
tin
g
co
n
d
i
tio
n
s
,
b
ac
k
g
r
o
u
n
d
s
,
an
d
h
an
d
ch
ar
ac
ter
is
tics
,
with
m
in
im
al
v
ar
iatio
n
in
s
k
in
to
n
e,
h
an
d
s
ize,
o
r
e
n
v
ir
o
n
m
en
tal
co
n
te
x
t.
T
h
ese
f
ac
to
r
s
m
ay
in
tr
o
d
u
ce
d
ataset
b
ias
a
n
d
lim
it
t
h
e
m
o
d
el’
s
g
en
er
aliza
tio
n
a
b
ilit
y
in
r
ea
l
-
wo
r
ld
s
ettin
g
s
.
A
m
o
r
e
co
m
p
r
eh
en
s
iv
e
an
d
v
ar
ied
d
ataset
wo
u
ld
b
e
n
ec
ess
ar
y
t
o
im
p
r
o
v
e
r
o
b
u
s
tn
ess
,
p
ar
ticu
lar
ly
in
p
r
ac
tical
ass
is
tiv
e
ap
p
licatio
n
s
in
ten
d
ed
f
o
r
d
iv
er
s
e
u
s
er
p
o
p
u
latio
n
s
.
Fig
u
r
e
2
.
E
x
am
p
le
o
f
th
e
SIBI
2
.
2
.
O
bje
c
t
a
nn
o
t
a
t
io
n
Ob
ject
lab
elin
g
o
r
an
n
o
tatio
n
is
u
s
ed
to
d
eter
m
in
e
in
f
o
r
m
atio
n
r
elate
d
to
th
e
class
an
d
s
p
ec
ial
f
ea
tu
r
es
th
at
r
ep
r
esen
t
o
b
jects
in
ea
c
h
im
ag
e
b
y
cr
ea
tin
g
a
b
o
u
n
d
in
g
b
o
x
.
T
h
is
aim
s
to
tr
ain
th
e
m
o
d
el
to
b
e
ab
le
to
r
ec
o
g
n
ize
o
b
jects
th
at
will
b
e
p
r
e
d
icted
.
Ob
ject
lab
elin
g
in
th
is
r
esear
ch
is
d
o
n
e
m
an
u
ally
u
s
in
g
th
e
L
ab
elim
g
to
o
l.
L
a
b
elim
g
is
r
u
n
in
th
e
An
ac
o
n
d
a
to
o
ls
en
v
ir
o
n
m
e
n
t
in
s
talled
o
n
th
e
W
in
d
o
ws
o
p
er
atin
g
s
y
s
tem
.
Ob
ject
lab
elin
g
r
esu
lts
in
a
.
x
m
l
f
ile
with
in
f
o
r
m
atio
n
r
e
g
ar
d
in
g
th
e
o
b
ject'
s
class
an
d
th
e
b
o
u
n
d
i
n
g
b
o
x
co
o
r
d
i
n
ates.
T
h
e
o
b
ject
lab
elin
g
p
r
o
ce
s
s
p
r
o
d
u
ce
s
o
n
e
o
b
ject
class
p
er
im
ag
e
with
5
4
6
o
b
jects.
Af
ter
lab
elin
g
o
b
jects
o
n
d
ig
ital
im
a
g
es
u
s
in
g
L
ab
elI
m
g
,
a
n
XM
L
f
ile
is
p
r
o
d
u
ce
d
th
at
co
n
tain
s
i
n
f
o
r
m
atio
n
r
elate
d
to
th
e
class
an
d
co
o
r
d
in
ates o
f
th
e
o
b
ject.
2
.
3
.
Da
t
a
s
et
s
prepro
ce
s
s
ing
I
n
th
e
d
ata
p
r
e
p
r
o
ce
s
s
in
g
p
h
ase,
th
e
co
llected
I
n
d
o
n
esian
Sig
n
L
an
g
u
ag
e
Sy
s
tem
(
SI
B
I
)
d
ig
ital
im
ag
es
ar
e
p
r
ep
a
r
ed
f
o
r
th
e
n
ee
d
s
o
f
m
o
d
el
tr
ain
i
n
g
u
s
in
g
T
en
s
o
r
f
lo
w.
T
h
e
i
n
itial
s
tep
in
th
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
p
r
o
ce
s
s
in
v
o
l
v
e
s
th
e
d
ata
th
at
h
as
b
ee
n
p
r
ep
a
r
ed
alo
n
g
with
th
e
co
o
r
d
in
ate
p
o
in
t
in
f
o
r
m
atio
n
o
f
o
b
jects
an
d
class
es.
B
o
th
d
ata
ar
e
co
n
v
er
ted
in
to
T
FR
ec
o
r
d
(
T
en
s
o
r
Flo
w
r
ec
o
r
d
)
f
o
r
m
at.
T
h
is
p
r
o
ce
s
s
b
eg
in
s
with
th
e
.
xml
f
ile
cr
ea
ted
f
r
o
m
th
e
o
b
ject
la
b
elin
g
p
r
o
ce
s
s
tr
an
s
f
o
r
m
ed
in
to
a
.
csv
f
ile.
Fu
r
th
er
m
o
r
e,
th
e
.
csv
f
ile
cr
ea
ted
is
u
s
ed
as
t
h
e
b
asi
s
f
o
r
t
h
e
c
o
n
v
er
s
io
n
p
r
o
ce
s
s
t
o
th
e
T
FR
ec
o
r
d
f
o
r
m
at.
I
n
th
i
s
p
r
o
ce
s
s
,
th
e
d
ig
ital
im
ag
e
in
f
o
r
m
atio
n
in
th
e
.
csv
f
ile
is
p
r
o
ce
s
s
ed
in
a
T
FR
ec
o
r
d
s
p
ec
if
icatio
n
f
o
r
m
at.
2
.
3
.
1
.
Dim
ens
io
n
re
s
izes
T
h
er
e
is
p
r
ep
r
o
ce
s
s
in
g
th
at
o
cc
u
r
s
in
th
e
d
ata
p
ip
elin
e,
wh
er
e
r
esizin
g
is
p
er
f
o
r
m
ed
o
n
b
o
th
im
ag
e
s
ize
an
d
o
b
ject
c
o
o
r
d
in
ate
p
o
in
ts
.
R
esize
o
n
im
ag
e
s
ize
aim
s
to
en
s
u
r
e
t
h
at
all
im
ag
es
h
av
e
u
n
if
o
r
m
d
im
en
s
io
n
s
,
wh
ile
r
esize
o
n
o
b
ject
co
o
r
d
i
n
ate
p
o
in
ts
aim
s
to
m
ain
tain
o
b
ject
p
r
o
p
o
r
tio
n
s
a
f
ter
im
ag
e
r
esize.
2
.
3
.
2
.
I
m
a
g
e
a
ug
m
ent
a
t
io
n
T
h
e
au
g
m
en
tatio
n
p
er
f
o
r
m
ed
d
u
r
in
g
m
o
d
el
tr
ain
in
g
co
n
s
is
ts
o
f
two
m
ain
ty
p
es
o
f
tr
an
s
f
o
r
m
atio
n
s
.
First,
h
o
r
izo
n
tal
f
lip
p
in
g
with
a
p
r
o
b
a
b
ilit
y
o
f
0
.
3
,
p
r
o
d
u
cin
g
a
h
o
r
izo
n
tally
m
ir
r
o
r
e
d
im
ag
e
to
im
p
r
o
v
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
7
5
9
-
5
7
6
9
5762
o
r
ien
tatio
n
i
n
v
ar
ian
ce
.
Seco
n
d
,
b
r
ig
h
tn
ess
ad
ju
s
tm
en
t
with
a
m
ax
im
u
m
d
elta
o
f
0
.
3
,
en
h
a
n
cin
g
r
o
b
u
s
tn
ess
to
lig
h
tin
g
v
ar
iatio
n
s
.
Fig
u
r
e
3
s
h
o
ws im
ag
e
au
g
m
en
tatio
n
wit
h
h
o
r
iz
o
n
tal
f
lip
a
n
d
b
r
ig
h
tn
es
s
ad
ju
s
tm
en
t.
Fig
u
r
e
3
.
Data
au
g
m
en
tatio
n
e
x
am
p
les h
o
r
iz
o
n
tal
f
lip
an
d
ad
ju
s
t b
r
ig
h
tn
ess
2
.
4
.
P
r
o
po
s
ed
f
a
s
t
er
R
-
CNN
wit
h
re
s
idu
a
l net
wo
rk
Fas
ter
R
-
C
NN
is
a
n
etwo
r
k
th
at
co
m
b
in
es
Fas
t
R
-
C
NN
an
d
R
PN
to
d
ec
r
ea
s
e
d
u
r
atio
n
c
o
m
p
lex
ity
an
d
g
en
e
r
ate
h
ig
h
-
q
u
ality
r
e
g
io
n
p
r
o
p
o
s
als,
b
o
u
n
d
in
g
b
o
x
es,
an
d
o
b
ject
n
ess
s
co
r
es
s
i
m
u
ltan
eo
u
s
ly
[
1
6
]
,
wh
ich
is
illu
s
tr
ated
in
Fig
u
r
e
4
.
I
n
s
tu
d
y
[
1
7
]
,
t
h
e
p
h
ases
o
f
t
h
e
Fas
ter
R
-
C
NN
ar
e
as f
o
llo
ws:
˗
T
h
e
in
p
u
t
d
ig
ital
im
ag
e
u
n
d
er
g
o
es
co
n
v
o
lu
tio
n
a
n
d
p
o
o
l
in
g
o
p
er
atio
n
s
th
r
o
u
g
h
f
ea
tu
r
e
ex
tr
ac
tio
n
to
o
b
tain
a
f
ea
tu
r
e
m
ap
.
˗
T
h
e
f
ea
tu
r
e
m
ap
is
th
en
g
iv
en
to
th
e
R
PN n
etwo
r
k
,
wh
ich
p
e
r
f
o
r
m
s
o
b
jectn
ess
p
r
ed
ictio
n
.
˗
T
h
e
R
PN
n
etwo
r
k
p
r
o
v
id
es
s
ev
er
al
p
r
o
p
o
s
ed
an
ch
o
r
s
th
r
o
u
g
h
co
o
r
d
in
ate
p
o
in
ts
an
d
o
b
ject
n
ess
p
r
ed
ictio
n
s
co
r
es.
˗
T
h
e
R
PN
o
u
tp
u
t
is
s
en
t
to
th
e
Fas
t
R
-
C
NN
n
etwo
r
k
,
wh
er
e
th
e
o
b
ject
o
u
tp
u
t
is
p
r
o
ce
s
s
ed
in
th
e
f
u
lly
co
n
n
ec
ted
la
y
er
f
o
r
class
if
icatio
n
.
Fig
u
r
e
4
.
Step
s
o
f
th
e
f
aster
R
-
C
NN
p
r
o
ce
s
s
Fas
ter
R
-
C
NN
i
s
a
d
ev
elo
p
m
en
t
o
f
th
e
p
r
e
v
io
u
s
m
eth
o
d
s
ca
lled
R
-
C
NN
an
d
Fas
t
R
-
C
N
N.
R
-
C
NN
was
f
ir
s
t
p
r
esen
ted
b
y
Gir
s
h
ick
et
a
l.
[
1
8
]
in
2
0
1
4
as
an
o
b
ject
d
etec
tio
n
m
eth
o
d
t
h
at
u
s
es
a
s
elec
tiv
e
s
ea
r
ch
alg
o
r
ith
m
to
cr
ea
te
ar
o
u
n
d
2
0
0
0
r
eg
io
n
p
r
o
p
o
s
als
p
er
im
ag
e
,
th
en
ex
tr
ac
ts
f
ea
tu
r
es
f
r
o
m
e
ac
h
r
eg
io
n
p
r
o
p
o
s
al
u
s
in
g
a
p
r
e
-
tr
ain
e
d
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
,
an
d
f
in
ally
class
if
ies th
e
r
eg
io
n
p
r
o
p
o
s
als u
tili
zin
g
a
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SV
M)
.
Af
ter
th
at,
Gir
s
h
ick
[
1
9
]
in
tr
o
d
u
ce
d
Fas
t
R
-
C
N
N.
T
h
e
Fas
t
R
-
C
N
N
alg
o
r
ith
m
was
d
ev
elo
p
e
d
to
im
p
r
o
v
e
s
o
m
e
o
f
th
e
s
h
o
r
tco
m
in
g
s
o
f
R
-
C
NN
b
y
allo
win
g
co
n
v
o
lu
tio
n
to
b
e
p
er
f
o
r
m
ed
o
n
ly
o
n
ce
o
n
ea
c
h
im
ag
e
an
d
u
s
in
g
th
e
r
esu
l
tin
g
f
ea
tu
r
e
m
ap
t
o
g
en
e
r
ate
p
r
ed
icted
r
eg
i
o
n
p
r
o
p
o
s
als,
th
u
s
im
p
r
o
v
in
g
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
a
n
d
o
b
ject
d
etec
tio
n
ac
cu
r
ac
y
[
2
0
]
.
T
h
e
alg
o
r
ith
m
was
f
u
r
th
er
im
p
r
o
v
ed
to
Fas
ter
R
-
C
NN
in
tr
o
d
u
ce
d
b
y
R
en
et
a
l.
[
2
0
]
,
u
tili
zin
g
an
ad
d
itio
n
al
C
NN
ca
lled
R
PN
to
g
en
er
ate
r
e
g
io
n
p
r
o
p
o
s
als s
tr
ai
g
h
t f
r
o
m
v
is
io
n
f
ea
tu
r
es,
th
u
s
s
to
p
p
in
g
th
e
n
ee
d
to
u
s
e
th
e
s
e
lectiv
e
s
ea
r
ch
.
2
.
4
.
1
.
Resid
ua
l
net
wo
rk
a
rc
hite
ct
ure
a
s
ba
ck
bo
ne
Fas
ter
R
-
C
NN
g
en
er
ally
u
tili
ze
s
C
N
N
ar
ch
itectu
r
es
s
u
ch
as
R
es
Net
as
a
b
ac
k
b
o
n
e
to
p
er
f
o
r
m
f
ea
tu
r
e
ex
tr
ac
tio
n
,
wh
ich
is
th
en
u
s
ed
in
th
e
R
PN
an
d
class
i
f
icatio
n
s
tag
es
[
1
2
]
.
R
esNet
-
5
0
h
as
b
ee
n
wid
ely
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
s
ter R
-
C
N
N
imp
leme
n
ta
tio
n
fo
r
h
a
n
d
s
ig
n
r
ec
o
g
n
itio
n
o
f
th
e
…
(
P
a
u
l
u
s
Les
tyo
A
d
h
ia
tma
)
5763
o
p
er
ated
as
a
b
ac
k
b
o
n
e
in
o
b
j
ec
t
d
etec
tio
n
,
s
u
ch
as
Fas
ter
R
-
C
NN
[
2
1
]
.
T
h
e
R
esNet
ar
ch
it
ec
tu
r
e
was
f
o
r
m
ed
to
o
v
e
r
co
m
e
th
e
o
b
s
tacle
s
in
d
ee
p
lear
n
i
n
g
tr
ain
in
g
b
ec
a
u
s
e,
in
g
en
er
al,
it
ta
k
es
a
lo
n
g
d
u
r
atio
n
a
n
d
is
r
estricte
d
to
a
s
p
ec
if
ic
n
u
m
b
er
o
f
la
y
er
s
[
2
2
]
.
R
esid
u
al
n
etwo
r
k
s
en
h
a
n
ce
d
ee
p
n
et
wo
r
k
tr
ain
in
g
with
s
h
o
r
tcu
ts
th
at
ea
s
e
g
r
a
d
ien
t
f
l
o
w,
r
esu
ltin
g
i
n
q
u
ick
er
tr
ai
n
in
g
an
d
in
cr
ea
s
ed
ac
cu
r
ac
y
o
f
s
ig
n
if
ican
t
d
ep
th
[
2
3
]
,
as
s
h
o
wn
in
Fig
u
r
e
5
(
a)
.
T
h
e
m
o
r
e
r
ep
r
esen
tativ
e
th
e
f
ea
tu
r
e
m
ap
ex
tr
ac
ted
b
y
th
e
b
ac
k
b
o
n
e
,
th
e
b
etter
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
etec
to
r
,
as
a
b
etter
f
ea
tu
r
e
m
a
p
will
h
elp
in
cr
ea
s
e
th
e
ac
cu
r
a
cy
lev
el
o
f
o
b
ject
d
etec
tio
n
[
2
4
]
.
T
h
er
e
ar
e
v
a
r
io
u
s
v
er
s
io
n
s
o
f
th
e
R
esid
u
al
Netwo
r
k
s
ar
ch
itectu
r
e,
in
clu
d
in
g
R
esNet
-
5
0
,
R
esNet
-
1
0
1
,
an
d
R
esNet
-
1
5
2
,
wh
ich
h
av
e
d
if
f
er
en
t
lev
els
o
f
d
ep
th
a
n
d
c
o
m
p
lex
ity
to
m
ee
t
s
p
ec
if
ic
n
e
ed
s
.
Sli
m
en
e
et
a
l.
[
2
5
]
m
en
tio
n
e
d
th
at
th
e
n
am
e
o
f
t
h
e
R
esNet
ar
ch
itectu
r
e
d
ep
en
d
s
o
n
th
e
n
u
m
b
er
o
f
lay
e
r
s
,
as
in
R
esNet5
0
,
R
esNet1
0
1
,
an
d
R
esNet1
5
2
,
wh
ich
in
d
icate
h
a
v
in
g
5
0
,
1
0
1
,
an
d
1
5
2
lay
er
s
.
T
h
ese
lay
e
r
s
ar
e
ca
lcu
lated
f
r
o
m
th
e
to
tal
n
u
m
b
er
o
f
s
k
ip
co
n
n
ec
tio
n
s
,
co
n
v
o
l
u
tio
n
al
lay
er
s
,
p
o
o
lin
g
lay
er
s
,
an
d
f
u
lly
co
n
n
ec
ted
lay
er
s
th
at
co
m
p
r
is
e
th
e
n
etwo
r
k
,
wh
ic
h
i
s
d
etailed
in
Fig
u
r
e
5
(
b
)
.
T
h
e
ap
p
licatio
n
o
f
R
esNet
ar
ch
itectu
r
e
to
d
ig
ital
im
a
g
es
h
as
b
ee
n
u
s
ed
i
n
v
ar
i
o
u
s
s
tu
d
ies.
As
in
th
e
r
esear
ch
o
f
Sar
win
d
a
et
a
l.
[
2
2
]
,
th
e
d
etec
tio
n
o
f
co
l
o
r
ec
tal
ca
n
ce
r
was
co
n
d
u
cted
u
s
in
g
R
esNet
-
5
0
,
wh
ich
d
em
o
n
s
tr
ated
a
n
ac
cu
r
ac
y
ab
o
v
e
8
0
%
an
d
a
s
en
s
itiv
ity
ab
o
v
e
8
7
%
o
n
th
e
2
0
%
an
d
2
5
%
test
s
et
s
.
An
o
th
er
s
tu
d
y
b
y
Z
h
an
g
et
a
l.
[
2
6
]
p
r
o
p
o
s
es
a
h
ea
r
tb
ea
t
class
if
icatio
n
m
eth
o
d
b
ased
o
n
h
y
b
r
id
tim
e
-
f
r
eq
u
en
c
y
an
aly
s
is
an
d
tr
an
s
f
er
lear
n
in
g
with
R
esNe
t
-
1
0
1
,
ac
h
iev
in
g
9
9
.
7
5
%
ac
cu
r
ac
y
an
d
an
F1
-
s
c
o
r
e
o
f
0
.
9
0
1
6
f
o
r
1
4
ty
p
es
o
f
h
ea
r
tb
ea
ts
f
r
o
m
th
e
MI
T
-
B
I
H
d
atab
ase.
T
h
is
s
tu
d
y
b
y
Go
h
et
a
l.
[
2
7
]
ev
al
u
ates
d
ee
p
tr
an
s
f
er
lear
n
in
g
m
o
d
els
f
o
r
d
etec
tin
g
f
o
u
r
ca
teg
o
r
ies
o
f
f
ac
e
m
ask
-
wea
r
in
g
,
with
R
esNet
-
1
5
2
ac
h
iev
in
g
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
8
6
.
6
7
% o
n
a
test
in
g
s
et
an
d
8
4
.
4
7
% o
n
s
m
ar
tp
h
o
n
e
-
ca
p
tu
r
ed
v
id
eo
s
.
(
a)
(
b
)
Fig
u
r
e
5
.
R
esid
u
al
n
etwo
r
k
ar
ch
itectu
r
e
(
a)
s
k
ip
co
n
n
ec
tio
n
s
in
R
esNet
an
d
(
b
)
n
etwo
r
k
ar
ch
itectu
r
e
o
f
R
esNet
-
5
0
,
R
esNet
-
1
0
1
,
R
esN
et
-
152
2
.
4
.
2
.
M
o
del
co
nfig
ura
t
io
n a
nd
t
ra
ini
ng
pip
el
ine
T
h
is
r
esear
ch
will
u
s
e
R
esNet
-
5
0
,
R
esNet
-
1
0
1
,
an
d
R
esNet
-
1
5
2
as
th
e
b
ac
k
b
o
n
e
o
f
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
in
SIBI
r
ec
o
g
n
itio
n
a
n
d
will
b
e
co
m
p
ar
e
d
to
g
et
th
e
b
est
ar
ch
itectu
r
e
o
f
th
e
th
r
ee
m
o
d
els.
T
h
e
m
o
d
elin
g
was
co
n
d
u
cted
u
s
in
g
T
en
s
o
r
Flo
w
v
er
s
io
n
2
.
1
0
with
b
u
ilt
-
in
GPU
s
u
p
p
o
r
t.
T
r
ain
in
g
was
p
er
f
o
r
m
ed
o
n
a
p
e
r
s
o
n
al
lap
t
o
p
eq
u
ip
p
e
d
with
a
1
2
th
Gen
I
n
tel(
R
)
C
o
r
e(
T
M)
i5
-
1
2
5
0
0
H
2
.
5
0
GHz
C
PU,
NVI
DI
A
GeFo
r
ce
R
T
X
3
0
5
0
L
ap
to
p
GPU,
an
d
1
6
GB
R
AM
.
GPU
ac
ce
ler
atio
n
was
en
ab
led
,
with
an
av
er
ag
e
GPU
m
em
o
r
y
u
s
ag
e
b
etwe
en
3
to
4
GB
.
T
o
s
tr
ea
m
lin
e
th
e
ex
p
er
im
e
n
ts
an
d
s
im
p
lify
th
e
ev
alu
atio
n
,
th
e
th
r
ee
m
o
d
els
wer
e
in
teg
r
at
ed
in
a
p
ip
elin
e
u
s
in
g
T
en
s
o
r
Flo
w
to
o
ls
.
T
h
is
p
ip
elin
e
is
d
esig
n
ed
to
au
to
m
ate
th
e
wo
r
k
f
lo
w
f
r
o
m
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
,
m
o
d
el
tr
ain
in
g
,
to
r
esu
lt
ev
al
u
atio
n
.
T
h
e
f
o
llo
win
g
is
th
e
co
n
f
ig
u
r
atio
n
o
f
th
e
3
-
m
o
d
el
t
r
ain
in
g
p
i
p
elin
e
u
s
ed
in
th
is
s
tu
d
y
,
wh
ic
h
is
d
escr
ib
ed
in
th
e
T
ab
le
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
7
5
9
-
5
7
6
9
5764
T
ab
le
1
.
C
o
n
f
i
g
u
r
atio
n
o
f
t
r
ain
in
g
p
i
p
elin
e
C
o
n
f
i
g
u
r
a
t
i
o
n
V
a
l
u
e
D
i
me
n
si
o
n
si
z
e
3
2
0
×
3
2
0
N
u
m
o
f
c
l
a
ss
26
I
n
p
u
t
l
a
b
e
l
ma
p
"
/
/
/
−
.
"
Le
a
r
n
i
n
g
r
a
t
e
C
o
s
i
n
e
D
e
c
a
y
L
e
a
r
n
i
n
g
R
a
t
e
:
b
a
s
e
=
0
.
0
4
w
a
r
mu
p
=
0
.
0
1
3
3
3
3
t
o
t
a
l
st
e
p
s
=
4
6
8
0
0
w
a
r
mu
p
_
s
t
e
p
s:
2
0
0
0
D
a
t
a
a
u
g
m
e
n
t
a
t
i
o
n
H
o
r
i
z
o
n
t
a
l
F
l
i
p
(
P
r
o
b
a
b
i
l
i
t
a
s
=
0
.
3
)
,
A
d
j
u
st
B
r
i
g
h
t
n
e
ss
(
D
e
l
t
a
ma
k
s
=
0
.
3
)
B
a
t
c
h
si
z
e
1
N
u
m
s
t
e
p
s
4
6
8
0
0
n
u
m
st
e
p
s (
1
0
0
e
p
o
c
h
)
2
.
5
.
P
er
f
o
r
m
a
nce
m
e
t
rics
T
h
i
s
s
t
u
d
y
c
o
m
p
a
r
es
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
r
e
e
d
i
f
f
e
r
e
n
t
R
es
Ne
t
a
r
c
h
it
e
c
t
u
r
es
(
R
es
N
e
t
-
5
0
,
R
e
s
N
et
-
1
0
1
,
a
n
d
R
e
s
N
e
t
-
1
5
2
)
w
h
e
n
u
s
e
d
as
t
h
e
b
a
c
k
b
o
n
e
f
o
r
t
h
e
F
a
s
te
r
R
-
C
N
N
a
l
g
o
r
i
t
h
m
.
T
h
e
g
o
al
is
to
d
etec
t
I
n
d
o
n
esian
Sig
n
L
an
g
u
a
g
e
(
SIBI)
s
ig
n
s
in
d
ig
ital im
ag
es.
T
o
ev
alu
ate
ea
ch
m
o
d
el'
s
ef
f
ec
tiv
en
ess
,
a
co
n
f
u
s
io
n
m
atr
ix
will
b
e
u
s
ed
,
wh
ich
is
a
to
o
l
th
at
co
m
p
ar
es
th
e
ac
tu
al
s
ig
n
with
th
e
m
o
d
el'
s
p
r
ed
icted
s
ig
n
[
2
8
]
.
A
co
n
f
u
s
io
n
m
atr
ix
is
a
s
q
u
ar
e
m
atr
ix
wh
er
e
th
e
r
o
ws
d
ef
i
n
e
th
e
ac
tu
a
l
class
o
f
th
e
in
s
tan
ce
,
an
d
t
h
e
co
lu
m
n
s
ar
e
th
e
p
r
ed
icted
class
[
2
9
]
.
I
d
en
tify
i
n
g
SIBI
s
ig
n
lan
g
u
ag
e
co
n
s
is
tin
g
o
f
2
6
class
es
r
eq
u
ir
es
a
m
u
lticlas
s
co
n
f
u
s
io
n
m
atr
ix
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
m
u
lticlas
s
cla
s
s
if
icatio
n
u
s
es
a
×
co
n
tin
g
e
n
cy
tab
le
wh
e
r
e
ce
lls
[
,
]
(
=
1
,
.
.
.
,
,
=
1
,
.
.
.
,
)
r
ep
r
esen
t
th
e
f
r
eq
u
en
cy
o
f
cl
ass
o
b
s
er
v
atio
n
s
with
ac
t
u
al
class
an
d
p
r
ed
icted
class
[
3
0
]
.
T
ab
le
2
illu
s
tr
ates
an
ex
am
p
le
o
f
a
c
o
n
f
u
s
io
n
m
atr
ix
f
o
r
m
u
lticlas
s
class
if
icat
io
n
,
wh
er
e
ea
c
h
ce
ll sh
o
ws th
e
r
elatio
n
s
h
ip
b
et
wee
n
th
e
ac
tu
al
an
d
p
r
e
d
icted
class
es.
I
n
th
e
c
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
m
u
lticlas
s
clas
s
if
icatio
n
,
th
er
e
a
r
e
4
s
ec
tio
n
s
th
at
s
h
o
w
th
e
r
esu
lts
o
f
th
e
test
as f
o
llo
ws.
−
T
r
u
e
Po
s
itiv
e
(
T
P)
r
ef
e
r
s
to
in
s
tan
ce
s
wh
er
e
th
e
ac
tu
al
class
is
co
r
r
ec
tly
p
r
e
d
icted
.
−
T
r
u
e
Neg
ativ
e
(
T
N)
is
th
e
ac
t
u
al
class
p
r
ed
icted
to
b
e
t
r
u
e
i
n
th
e
n
e
g
ativ
e
class
.
−
Fals
e
Po
s
i
tiv
e
(
FP
)
h
ap
p
en
s
w
h
en
a
n
e
g
ativ
e
class
is
in
co
r
r
e
ctly
p
r
ed
icted
as p
o
s
itiv
e.
−
Fals
e
Neg
ativ
e
(
FN)
is
wh
en
a
p
o
s
itiv
e
class
is
in
co
r
r
ec
tly
p
r
ed
icted
as
n
eg
ativ
e
.
T
o
m
ea
s
u
r
e
th
e
q
u
ality
o
f
a
m
u
lticlas
s
cla
s
s
if
icat
io
n
m
o
d
el,
y
o
u
ca
n
c
o
m
p
ar
e
its
ar
ch
itect
u
r
e
with
a
co
n
f
u
s
io
n
m
atr
ix
tech
n
iq
u
e.
T
h
is
p
r
o
ce
s
s
g
e
n
er
ates
ac
cu
r
a
cy
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
t
h
e
F1
-
s
co
r
e,
wh
ich
ar
e
all
u
s
ed
to
ev
alu
ate
th
e
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
.
=
+
+
+
+
=
+
=
+
1
−
=
2
∗
(
∗
)
+
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
T
ra
ini
ng
m
o
del
T
h
e
f
aster
R
-
C
NN
m
o
d
el
was
tr
ain
ed
3
tim
es
with
d
if
f
er
en
t
ar
ch
itectu
r
es.
E
ac
h
m
o
d
el
w
as
tr
ain
ed
f
o
r
1
0
0
ep
o
ch
s
o
r
4
6
.
8
0
0
s
tep
s
,
with
a
to
tal
tr
ain
in
g
tim
e
o
f
ap
p
r
o
x
im
ately
4
,
7
8
3
s
ec
o
n
d
s
(
4
7
.
8
3
s
ec
o
n
d
s
p
er
ep
o
c
h
)
f
o
r
R
esNet
-
5
0
,
7
5
3
0
s
ec
o
n
d
s
(
7
5
.
3
0
s
ec
o
n
d
s
p
er
ep
o
c
h
)
f
o
r
R
esNet
-
1
0
1
,
an
d
1
1
8
1
1
s
ec
o
n
d
s
(
1
1
8
.
1
1
s
ec
o
n
d
s
p
er
ep
o
ch
)
f
o
r
R
esNet
-
1
5
2
.
Mo
d
el
tr
ain
in
g
u
s
es
lo
s
s
as
a
m
ea
s
u
r
em
en
t
o
f
th
e
q
u
ality
an
d
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
in
m
ak
in
g
p
r
e
d
ictio
n
s
o
n
tr
ai
n
in
g
d
ata.
Fig
u
r
e
6
(
a)
to
(
c)
r
ep
r
e
s
en
ts
a
co
m
b
i
n
atio
n
o
f
lo
s
s
in
th
e
b
o
x
class
if
ier
an
d
R
PN f
o
r
ea
ch
ar
c
h
itectu
r
e.
I
n
th
e
to
tal
lo
s
s
g
r
ap
h
o
f
th
e
Fas
ter
R
-
C
NN
m
o
d
el
tr
ain
in
g
ab
o
v
e,
it
is
k
n
o
wn
th
at
th
e
th
r
ee
m
o
d
els
s
h
o
w
a
s
ig
n
if
ican
t
d
ec
r
ea
s
e
in
to
tal
lo
s
s
u
n
til
th
e
1
0
0
th
ep
o
c
h
.
T
h
e
r
esu
lts
o
f
th
e
t
o
tal
lo
s
s
o
f
m
o
d
el
tr
ain
in
g
o
b
tain
ed
a
to
tal
lo
s
s
v
alu
e
o
f
0
.
0
0
2
in
th
e
R
esNet
-
5
0
m
o
d
e
l,
0
.
0
3
7
in
th
e
R
esNet
-
1
0
1
m
o
d
el,
a
n
d
0
.
0
0
0
5
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
s
ter R
-
C
N
N
imp
leme
n
ta
tio
n
fo
r
h
a
n
d
s
ig
n
r
ec
o
g
n
itio
n
o
f
th
e
…
(
P
a
u
l
u
s
Les
tyo
A
d
h
ia
tma
)
5765
th
e
R
esNet
-
1
5
2
m
o
d
el.
T
h
is
s
h
o
ws
th
at
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
co
n
v
er
g
es
an
d
p
er
f
o
r
m
s
well
in
d
etec
tin
g
o
b
jects in
th
e
tr
ain
in
g
d
ata.
(
a)
(
b
)
(
c)
Fig
u
r
e
6
.
T
o
tal
lo
s
s
g
r
ap
h
Fas
ter
R
-
C
NN
tr
ain
in
g
(
a)
R
esNet
-
5
0
,
(
b
)
R
esNet
-
1
0
1
,
a
n
d
(
c)
R
esNet
-
152
3
.
2
.
T
esting
m
o
del
E
ac
h
m
o
d
el
with
th
r
ee
d
if
f
er
e
n
t
ar
ch
itectu
r
es
is
th
e
n
test
ed
u
s
in
g
test
d
ata.
T
h
e
test
d
ata
co
n
s
is
ts
o
f
s
ix
d
ata
in
ea
c
h
class
,
ea
ch
o
f
w
h
ich
c
o
n
tain
s
o
n
e
o
r
ig
in
al
d
ata
s
et
a
n
d
f
iv
e
d
if
f
er
en
t
d
ig
ital
im
a
g
e
au
g
m
en
tatio
n
r
esu
lts
.
T
h
er
e
f
o
r
e,
1
5
6
d
ata
will
b
e
u
s
ed
f
o
r
test
in
g
.
T
esti
n
g
is
d
o
n
e
to
s
ee
th
e
m
o
d
el's
p
er
f
o
r
m
an
ce
in
r
ec
o
g
n
izin
g
S
I
B
I
d
ig
ital
im
ag
es.
An
ex
am
p
le
o
f
m
o
d
el
p
r
e
d
ictio
n
r
esu
lt
ass
es
s
m
en
t
d
u
r
in
g
test
in
g
ca
n
b
e
s
ee
n
in
T
a
b
le
2
.
T
ab
le
2
.
Mo
d
el
p
r
e
d
ictio
n
r
es
u
lts
ass
e
s
s
m
en
t d
u
r
in
g
test
in
g
O
u
t
p
u
t
A
c
t
u
a
l
c
l
a
ss
P
r
e
d
i
c
t
i
o
n
c
l
a
ss
R
e
s
u
l
t
A
A
Tr
u
e
M
E
F
a
l
se
D
N
o
n
e
F
a
l
se
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
7
5
9
-
5
7
6
9
5766
Fig
u
r
e
7
(
a
)
to
(
c)
s
h
o
ws
co
n
f
u
s
io
n
m
atr
i
x
f
o
r
ea
c
h
m
o
d
el
u
s
ed
to
ca
lcu
late
th
e
ev
alu
at
io
n
v
alu
e
:
R
esNet
-
5
0
,
R
esNet
-
1
0
1
,
an
d
R
esNet
-
152
.
T
h
ese
r
esu
lts
ar
e
in
th
e
f
o
r
m
o
f
a
m
u
lticlas
s
co
n
f
u
s
io
n
m
at
r
ix
with
2
7
r
o
ws
an
d
2
7
co
lu
m
n
s
.
T
h
e
r
o
w
v
alu
e
in
th
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
t
h
e
ac
tu
al
class
,
wh
ile
th
e
c
o
lu
m
n
v
alu
e
s
h
o
ws
th
e
m
o
d
el'
s
p
r
ed
ictio
n
.
C
lass
es
co
n
s
is
t
o
f
th
e
letter
s
A
to
Z
an
d
p
lu
s
th
e
No
n
e
class
,
wh
ich
in
d
icate
s
th
at
th
e
m
o
d
el
d
o
es
n
o
t r
ec
o
g
n
ize
an
y
o
b
jects in
th
e
in
p
u
t im
ag
e
.
(
a)
(
b
)
(
c)
Fig
u
r
e
7
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
test
r
esu
lts
(
a)
R
es
Net
-
5
0
,
(
b
)
R
esNet
-
1
0
1
,
an
d
(
c
)
R
esNet
-
152
3
.
3
.
P
er
f
o
r
m
a
nce
C
o
n
f
u
s
io
n
m
atr
ix
f
r
o
m
th
e
t
est
r
esu
lts
i
s
ca
lcu
lated
to
o
b
tain
th
e
m
o
d
el
p
er
f
o
r
m
an
ce
ev
alu
atio
n
v
alu
e.
T
h
e
ev
alu
atio
n
m
et
r
ics
u
s
ed
ar
e
ac
c
u
r
ac
y
,
p
r
ec
is
i
o
n
,
r
ec
all,
an
d
F
1
-
s
co
r
e.
T
a
b
le
3
p
r
esen
ts
th
e
ev
alu
atio
n
m
etr
ics f
o
r
all
th
r
ee
m
o
d
els.
B
ased
o
n
T
ab
le
3
,
ar
ch
itectu
r
es
in
th
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
f
o
r
r
ec
o
g
n
izin
g
SIBI,
it
is
k
n
o
wn
th
at
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
th
r
ee
m
o
d
els
d
if
f
er
s
s
ig
n
if
ic
an
tly
.
R
esNet
-
5
0
o
u
t
p
er
f
o
r
m
ed
R
esNet
-
1
0
1
an
d
R
esNet
-
1
5
2
in
b
o
th
ac
cu
r
ac
y
an
d
ef
f
icien
c
y
,
p
r
o
ce
s
s
in
g
f
a
s
ter
an
d
d
eliv
er
in
g
s
u
p
er
io
r
r
esu
lts
.
R
e
s
Net
-
101
s
h
o
wed
th
e
lo
west p
er
f
o
r
m
an
c
e,
wh
ile
R
esNet
-
1
5
2
im
p
r
o
v
ed
u
p
o
n
it b
u
t sti
ll f
ell
s
h
o
r
t o
f
R
esNet
-
50.
T
ab
le
3
.
Per
f
o
r
m
an
ce
ev
alu
ati
o
n
o
f
th
e
m
o
d
els
M
o
d
e
l
A
r
c
h
i
t
e
c
t
u
r
e
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
S
c
o
r
e
Ti
me
(
sec
o
n
d
)
R
e
sN
e
t
-
50
9
6
%
9
5
%
9
3
%
9
4
%
3
6
.
8
4
R
e
sN
e
t
-
101
7
1
%
7
9
%
6
9
%
7
0
%
3
8
.
8
2
R
e
sN
e
t
-
152
8
1
%
8
5
%
7
8
%
7
9
%
4
0
.
2
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
F
a
s
ter R
-
C
N
N
imp
leme
n
ta
tio
n
fo
r
h
a
n
d
s
ig
n
r
ec
o
g
n
itio
n
o
f
th
e
…
(
P
a
u
l
u
s
Les
tyo
A
d
h
ia
tma
)
5767
3
.
3
.
1
.
ResNet
-
50
T
h
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
with
R
esNet
-
5
0
ar
ch
itectu
r
e
as
th
e
b
ac
k
b
o
n
e
h
as
r
eliab
le
p
e
r
f
o
r
m
a
n
ce
in
d
etec
tin
g
ea
ch
class
in
2
6
l
etter
s
o
f
th
e
I
n
d
o
n
esian
s
ig
n
lan
g
u
ag
e
s
y
s
tem
(
SIBI)
.
T
h
is
is
s
h
o
wn
th
r
o
u
g
h
th
e
ev
alu
atio
n
r
esu
lts
wh
er
e
th
e
m
o
d
el
is
ab
le
to
p
r
e
d
ict
test
d
ata
with
m
etr
ic
v
alu
es
o
f
ac
cu
r
ac
y
9
6
%,
p
r
ec
is
io
n
9
5
%,
r
ec
all
9
3
%
an
d
F
1
-
s
co
r
e
9
4
%.
I
n
ter
m
s
o
f
test
ex
e
cu
tio
n
tim
e,
it
ca
n
b
e
s
ee
n
t
h
at
th
e
m
o
d
el
with
R
esNet
-
5
0
ar
ch
itectu
r
e
is
s
u
p
er
io
r
to
th
e
m
o
d
els
with
R
es
N
et
-
1
0
1
an
d
R
esNet
-
1
5
2
ar
ch
it
ec
tu
r
es.
T
h
e
s
im
p
le
ar
ch
itectu
r
al
lay
er
s
in
R
esNet
-
5
0
p
r
o
v
id
e
t
h
e
b
en
e
f
it o
f
less
r
eso
u
r
ce
u
tili
za
tio
n
.
T
h
e
lo
wer
ar
ch
itectu
r
e
co
m
p
lex
ity
in
R
esNet
-
5
0
allo
ws
th
e
m
o
d
el
to
lear
n
ess
en
tial
f
ea
tu
r
es
with
o
u
t
o
v
er
f
itti
n
g
,
esp
ec
ially
o
n
d
ata
with
a
lim
ited
v
ar
iety
a
n
d
a
m
o
u
n
t
o
f
d
ata.
H
o
wev
er
,
test
in
g
o
n
b
lu
r
r
e
d
d
ig
ital
im
ag
e
s
ce
n
ar
io
s
s
u
c
h
as
class
es
D,
O,
S,
a
n
d
U
wer
e
id
en
t
if
ied
less
ac
cu
r
ately
an
d
wer
e
n
o
t
r
ec
o
g
n
ized
as
o
b
jects b
y
R
PN.
T
h
is
is
b
ec
au
s
e
th
e
o
b
ject
in
th
e
im
a
g
e
is
to
o
b
lu
r
r
ed
to
b
e
r
ec
o
g
n
ized
b
y
t
h
e
m
o
d
el.
3
.
3
.
2
.
ResNet
-
101
T
h
e
Fas
ter
R
-
C
NN
alg
o
r
ith
m
with
R
es
Net
-
1
0
1
ar
ch
itectu
r
e
s
h
o
ws
les
s
th
an
o
p
tim
al
p
er
f
o
r
m
a
n
ce
with
m
etr
ics
ac
cu
r
ac
y
o
f
7
1
%,
p
r
ec
is
io
n
o
f
7
9
%,
r
ec
all
o
f
6
9
%,
an
d
F
1
-
s
co
r
e
o
f
7
0
%.
Of
th
e
2
6
ex
is
tin
g
class
es,
th
is
m
o
d
el
ca
n
o
n
l
y
a
cc
u
r
ately
p
r
ed
ict
class
es
in
9
class
es,
n
am
ely
class
es
D,
F,
I
,
J
,
K,
L
,
X,
Y,
an
d
Z
.
I
n
f
ac
t,
th
er
e
is
class
d
ata
th
at
ca
n
n
o
t
b
e
p
r
e
d
icted
.
T
h
er
e
ar
e
ev
en
class
d
ata
th
at
ca
n
n
o
t
b
e
p
r
ed
icted
co
r
r
ec
tly
,
wh
ic
h
ar
e
th
e
R
an
d
U
class
d
ata.
T
h
e
ex
ec
u
tio
n
tim
e
o
f
th
e
R
esNet
-
1
0
1
ar
ch
itectu
r
e
m
o
d
el
is
3
8
.
8
2
s
ec
o
n
d
s
.
T
h
is
s
h
o
w
s
th
at
th
is
m
o
d
el
is
lo
n
g
er
th
an
th
e
R
esN
et
-
5
0
ar
ch
itectu
r
e
m
o
d
el,
with
an
in
cr
ea
s
e
in
test
ex
ec
u
tio
n
tim
e
o
f
5
.
3
7
%.
T
h
is
is
d
u
e
to
th
e
h
ig
h
er
co
m
p
lex
ity
o
f
th
e
R
esNet
-
1
0
1
ar
ch
itectu
r
e
co
m
p
ar
e
d
to
R
esNet
-
5
0
,
wh
ich
r
esu
lts
in
g
r
ea
ter
r
eso
u
r
ce
u
tili
za
tio
n
an
d
co
m
p
u
tatio
n
al
p
r
o
ce
s
s
es.
T
h
e
lack
o
f
o
p
tim
izatio
n
o
f
t
h
e
m
o
d
el
with
R
esNet
-
1
0
1
ar
ch
itectu
r
e
is
d
u
e
to
th
e
o
v
er
f
i
ttin
g
o
f
th
e
m
o
d
el
to
th
e
lim
ited
tr
ain
in
g
d
ata.
I
n
ad
d
itio
n
,
it
ap
p
ea
r
s
th
at
th
e
v
ar
iatio
n
s
in
th
e
te
s
t
d
ata
ar
e
p
o
o
r
ly
r
ec
o
g
n
ized
b
y
th
e
m
o
d
el,
in
d
icatin
g
th
at
th
e
m
o
d
el
ca
n
n
o
t
g
en
er
alize
well
to
v
ar
iatio
n
s
n
o
t
p
r
esen
t
in
th
e
tr
ain
in
g
d
ata.
A
n
o
th
er
f
ac
to
r
co
n
tr
ib
u
tin
g
to
th
is
lo
w
p
e
r
f
o
r
m
a
n
ce
is
th
e
lack
o
f
tr
a
in
in
g
d
ata
as
th
e
co
m
p
lex
ity
o
f
th
e
R
esNet
-
1
0
1
ar
ch
itectu
r
e
r
e
q
u
ir
es m
o
r
e
d
at
a
f
o
r
ef
f
ec
tiv
e
tr
ain
in
g
.
3
.
3
.
3
.
ResNet
-
152
Featu
r
e
ex
tr
ac
tio
n
o
n
R
esNet
-
1
5
2
is
th
e
m
o
s
t c
o
m
p
lex
co
m
p
ar
ed
to
R
esNet
-
5
0
an
d
R
esNet
-
1
0
1
.
T
h
e
h
ig
h
c
o
m
p
lex
ity
o
f
th
e
ar
c
h
itectu
r
e
m
ak
es
th
e
test
ex
ec
u
tio
n
tim
e
o
f
th
is
m
o
d
el
th
e
l
o
n
g
est
co
m
p
ar
ed
to
th
e
o
th
er
two
m
o
d
els.
B
ased
o
n
t
h
e
r
esear
ch
,
th
e
ex
ec
u
tio
n
tim
e
tak
es
4
0
.
2
0
s
ec
o
n
d
s
f
o
r
th
e
test
in
g
p
r
o
ce
s
s
.
T
h
e
R
esNet
-
1
5
2
ar
ch
itectu
r
e
m
o
d
el
is
g
o
o
d
in
th
e
test
in
g
p
r
o
ce
s
s
,
with
an
ac
cu
r
ac
y
m
etr
i
c
v
alu
e
o
f
8
0
.
7
5
%,
p
r
ec
is
io
n
o
f
8
5
%,
r
ec
all
o
f
7
8
%,
an
d
F
1
-
s
co
r
e
o
f
7
9
%.
T
h
e
ev
alu
atio
n
v
alu
es
s
h
o
w
th
at
th
is
m
o
d
el
is
b
etter
at
p
r
ed
ictin
g
test
d
ata
th
a
n
th
e
R
esNet
-
1
0
1
ar
ch
itectu
r
e
m
o
d
el
b
u
t lo
wer
th
a
n
th
e
R
esNet
-
50.
T
h
e
m
o
d
el
q
u
ality
with
R
es
Net
-
1
5
2
ar
ch
itectu
r
e
ten
d
s
to
ex
p
er
ien
ce
o
v
er
f
itti
n
g
,
s
im
ilar
to
th
e
R
esNet
-
1
0
1
ar
ch
itectu
r
e
m
o
d
el.
T
h
is
is
b
ec
au
s
e
th
e
s
m
all
am
o
u
n
t
o
f
d
ata
u
s
ed
in
tr
ai
n
in
g
m
a
k
es
it
less
ef
f
ec
tiv
e
f
o
r
m
o
d
els
with
h
ig
h
co
m
p
lex
ity
,
s
u
ch
as
R
esNet
-
1
5
2
,
to
g
e
n
er
alize
th
e
d
ata.
As a
r
esu
lt,
th
is
m
o
d
el
p
er
f
o
r
m
s
v
er
y
well
in
t
r
ain
in
g
b
u
t le
s
s
o
p
tim
ally
in
test
in
g
w
ith
n
ew
d
ata
v
ar
iatio
n
s
.
4.
CO
NCLU
SI
O
N
T
h
i
s
r
e
s
e
a
r
c
h
s
u
c
c
ess
f
u
l
l
y
i
m
p
l
e
m
e
n
t
s
t
h
e
R
es
N
et
-
5
0
,
R
esN
e
t
-
1
0
1
,
a
n
d
R
e
s
N
e
t
-
1
5
2
a
r
ch
i
t
e
c
t
u
r
es
i
n
t
h
e
F
as
t
e
r
R
-
C
N
N
al
g
o
r
i
t
h
m
t
o
r
e
c
o
g
n
i
z
e
t
h
e
I
n
d
o
n
e
s
i
a
n
la
n
g
u
a
g
e
s
i
g
n
s
y
s
t
e
m
(
S
I
B
I
)
.
T
h
i
s
c
a
n
b
e
p
r
o
v
e
n
t
h
r
o
u
g
h
t
h
e
s
m
o
o
t
h
t
r
a
i
n
i
n
g
p
r
o
c
e
s
s
u
n
ti
l
it
r
e
a
c
h
es
t
h
e
f
i
n
al
ep
o
c
h
a
n
d
t
h
e
e
n
t
i
r
e
m
o
d
e
l
e
x
p
er
i
e
n
c
e
s
a
c
o
n
s
is
t
e
n
t
d
e
c
r
e
a
s
e
i
n
l
o
s
s
i
n
t
h
e
t
r
ai
n
i
n
g
p
r
o
c
e
s
s
.
F
r
o
m
t
h
e
e
v
a
l
u
a
ti
o
n
r
e
s
u
l
ts
,
t
h
e
Fa
s
t
e
r
R
-
C
NN
m
o
d
e
l
w
i
t
h
R
es
N
e
t
-
50
a
r
c
h
i
t
e
c
t
u
r
e
s
h
o
w
e
d
t
h
e
b
e
s
t
a
n
d
m
o
s
t
e
f
f
i
ci
e
n
t
p
e
r
f
o
r
m
a
n
c
e
wi
t
h
a
n
a
c
c
u
r
a
c
y
v
a
l
u
e
o
f
9
6
.
1
5
%
a
n
d
a
n
e
x
e
c
u
ti
o
n
t
i
m
e
o
f
3
6
.
8
4
s
e
c
o
n
d
s
i
n
t
h
e
t
e
s
ti
n
g
p
r
o
c
e
s
s
.
T
h
e
r
e
f
o
r
e
,
R
es
N
e
t
-
5
0
w
as
c
h
o
s
e
n
as
t
h
e
b
es
t
b
a
c
k
b
o
n
e
/
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
r
c
h
i
t
e
c
t
u
r
e
i
n
t
h
e
F
a
s
t
e
r
R
-
C
N
N
al
g
o
r
i
t
h
m
f
o
r
r
e
c
o
g
n
i
z
i
n
g
t
h
e
S
I
B
I
.
T
h
i
s
s
t
u
d
y
c
o
n
t
r
i
b
u
t
e
s
t
o
t
h
e
f
i
e
l
d
b
y
e
m
p
i
r
i
c
a
l
e
v
a
l
u
a
t
i
o
n
o
f
b
a
c
k
b
o
n
e
s
e
l
e
c
t
i
o
n
b
a
s
e
d
o
n
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
,
i
n
f
e
r
e
n
c
e
t
i
m
e
,
a
n
d
m
o
d
e
l
c
o
m
p
l
e
x
i
t
y
u
n
d
e
r
l
i
m
i
te
d
d
a
t
a
c
o
n
d
i
t
i
o
n
s
,
p
r
o
v
i
d
i
n
g
p
r
a
c
t
i
c
al
i
n
s
i
g
h
ts
f
o
r
r
e
a
l
-
w
o
r
l
d
d
e
p
l
o
y
m
e
n
t
.
B
ased
o
n
th
is
r
esear
ch
,
R
e
s
Net
-
5
0
is
r
ec
o
m
m
en
d
ed
as
th
e
to
p
ch
o
ice
f
o
r
th
e
Fas
ter
R
-
C
N
N
alg
o
r
ith
m
'
s
b
ac
k
b
o
n
e
in
ap
p
licatio
n
s
d
esig
n
ed
to
r
ec
o
g
n
ize
I
n
d
o
n
esian
s
ig
n
lan
g
u
ag
e
(
SIBI)
.
T
h
ese
ap
p
licatio
n
s
co
u
ld
b
e
m
o
b
il
e
ap
p
s
o
r
co
m
m
u
n
icatio
n
d
ev
ices
th
at
h
elp
b
r
id
g
e
th
e
co
m
m
u
n
icatio
n
g
a
p
b
etwe
en
th
e
d
ea
f
c
o
m
m
u
n
ity
an
d
th
e
g
e
n
er
al
p
u
b
lic,
p
r
o
m
o
tin
g
m
o
r
e
in
clu
s
iv
e
in
ter
a
ctio
n
s
.
Fo
r
f
u
tu
r
e
r
esear
ch
,
it'
s
s
u
g
g
ested
to
u
s
e
lar
g
er
d
atasets
an
d
ex
p
er
im
e
n
t
with
d
if
f
e
r
en
t
Fas
ter
R
-
C
NN
co
n
f
ig
u
r
atio
n
s
.
T
h
e
cu
r
r
en
t
f
in
d
in
g
s
lay
a
s
tr
o
n
g
f
o
u
n
d
atio
n
f
o
r
b
u
ild
in
g
SIBI
r
ec
o
g
n
itio
n
s
y
s
tem
s
th
at
ca
n
b
e
in
teg
r
ated
in
to
ass
is
tiv
e
tech
n
o
lo
g
ies.
W
h
ile
th
is
s
tu
d
y
'
s
p
r
o
to
ty
p
e
was
t
ested
in
a
co
n
tr
o
lled
e
n
v
ir
o
n
m
en
t,
f
u
t
u
r
e
wo
r
k
s
h
o
u
ld
f
o
c
u
s
o
n
r
ea
l
-
tim
e
d
ep
lo
y
m
e
n
t
an
d
u
s
ab
ilit
y
s
tu
d
ies,
in
clu
d
in
g
i
m
p
lem
en
ta
tio
n
o
n
m
o
b
ile
o
r
em
b
ed
d
e
d
p
latf
o
r
m
s
f
o
r
p
r
ac
ti
ca
l,
r
ea
l
-
wo
r
ld
u
s
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
7
5
9
-
5
7
6
9
5768
RE
F
E
R
E
NC
E
S
[
1
]
W
.
A
l
y
,
S
.
A
l
y
,
a
n
d
S
.
A
l
mo
t
a
i
r
i
,
“
U
ser
-
i
n
d
e
p
e
n
d
e
n
t
a
mer
i
c
a
n
s
i
g
n
l
a
n
g
u
a
g
e
a
l
p
h
a
b
e
t
r
e
c
o
g
n
i
t
i
o
n
b
a
se
d
o
n
d
e
p
t
h
i
ma
g
e
a
n
d
P
C
A
N
e
t
f
e
a
t
u
r
e
s
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
7
,
p
p
.
1
2
3
1
3
8
–
1
2
3
1
5
0
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
9
.
2
9
3
8
8
2
9
.
[
2
]
S
u
h
a
r
j
i
t
o
,
N
.
T
h
i
r
a
c
i
t
t
a
,
a
n
d
H
.
G
u
n
a
w
a
n
,
“
S
I
B
I
si
g
n
l
a
n
g
u
a
g
e
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
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
c
o
mb
i
n
e
d
w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
n
d
n
o
n
-
t
r
a
i
n
a
b
l
e
p
a
r
a
met
e
r
s,”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
7
9
,
n
o
.
2
0
1
9
,
p
p
.
7
2
–
8
0
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s
.
2
0
2
0
.
1
2
.
0
1
1
.
[
3
]
E.
R
a
k
u
n
,
I
.
G
.
B
.
H
.
W
i
d
h
i
n
u
g
r
a
h
a
,
a
n
d
N
.
F
.
P
u
t
r
a
S
e
t
y
o
n
o
,
“
W
o
r
d
r
e
c
o
g
n
i
t
i
o
n
a
n
d
a
u
t
o
ma
t
e
d
e
p
e
n
t
h
e
si
s
r
e
m
o
v
a
l
f
o
r
I
n
d
o
n
e
si
a
n
si
g
n
s
y
st
e
m
se
n
t
e
n
c
e
g
e
st
u
r
e
s,
”
I
n
d
o
n
e
si
a
n
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
i
c
a
l
En
g
i
n
e
e
ri
n
g
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
6
,
no
.
3
,
pp
.
1
4
0
2
-
1
4
1
4
,
Ju
n
e
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
e
c
s.
v
2
6
.
i
3
.
p
p
1
4
0
2
-
1
4
1
4
.
[
4
]
A
.
R
.
S
y
u
l
i
st
y
o
,
D
.
S
.
H
o
r
ma
n
s
y
a
h
,
a
n
d
P
.
Y
.
S
a
p
u
t
r
a
,
“
S
I
B
I
(
S
i
st
e
m
I
sy
a
r
a
t
B
a
h
a
s
a
I
n
d
o
n
e
si
a
)
t
r
a
n
sl
a
t
i
o
n
u
si
n
g
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
(
C
N
N
)
,
”
I
O
P
C
o
n
f
e
r
e
n
c
e
S
e
ri
e
s:
M
a
t
e
ri
a
l
s
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
7
3
2
,
n
o
.
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
5
7
-
8
9
9
X
/
7
3
2
/
1
/
0
1
2
0
8
2
.
[
5
]
P
.
S
h
a
r
ma,
A
.
R
a
n
j
a
n
,
A
.
S
h
e
k
h
a
w
a
t
,
a
n
d
M
.
R
a
j
,
“
A
n
o
v
e
l
a
p
p
r
o
a
c
h
f
o
r
i
d
e
n
t
i
f
i
c
a
t
i
o
n
o
f
si
g
n
l
a
n
g
u
a
g
e
s
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
v
o
l
.
5
,
n
o
.
4
,
p
p
.
1
2
8
4
–
1
2
8
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
5
6
2
9
/
5
2
5
2
-
0
5
0
4
1
2
8
4
1
2
8
8
.
[
6
]
N
.
E.
K
h
a
l
i
f
a
,
M
.
L
o
e
y
,
a
n
d
S
.
M
i
r
j
a
l
i
l
i
,
“
A
c
o
m
p
r
e
h
e
n
si
v
e
s
u
r
v
e
y
o
f
r
e
c
e
n
t
t
r
e
n
d
s
i
n
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
d
i
g
i
t
a
l
i
m
a
g
e
s
a
u
g
me
n
t
a
t
i
o
n
,
”
A
rt
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
v
i
e
w
,
v
o
l
.
5
5
,
n
o
.
3
,
p
p
.
2
3
5
1
–
2
3
7
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
4
6
2
-
021
-
1
0
0
6
6
-
4.
[
7
]
Y
.
Li
u
,
P
.
S
u
n
,
N
.
W
e
r
g
e
l
e
s,
a
n
d
Y
.
S
h
a
n
g
,
“
A
su
r
v
e
y
a
n
d
p
e
r
f
o
r
ma
n
c
e
e
v
a
l
u
a
t
i
o
n
o
f
d
e
e
p
l
e
a
r
n
i
n
g
m
e
t
h
o
d
s
f
o
r
smal
l
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
Ex
p
e
rt
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
7
2
,
n
o
.
A
p
r
i
l
2
0
2
0
,
p
.
1
1
4
6
0
2
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
1
.
1
1
4
6
0
2
.
[
8
]
L.
D
e
n
g
,
S
.
L
i
u
,
Y
.
C
h
e
n
g
,
G
.
Z
h
a
o
,
a
n
d
J
.
X
u
,
“
A
l
g
o
r
i
t
h
m
f
o
r
d
i
a
b
e
t
i
c
r
e
t
i
n
a
l
i
m
a
g
e
a
n
a
l
y
s
i
s
b
a
s
e
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
2
,
n
o
.
3
0
,
p
p
.
4
7
5
5
9
–
4
7
5
8
4
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
0
2
3
-
1
5
5
0
3
-
w.
[
9
]
M
.
F
.
S
.
S
a
b
i
r
e
t
a
l
.
,
“
A
n
a
u
t
o
mat
e
d
r
e
a
l
-
t
i
m
e
f
a
c
e
mas
k
d
e
t
e
c
t
i
o
n
s
y
st
e
m
u
si
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
w
i
t
h
F
a
st
e
r
-
R
C
N
N
i
n
t
h
e
e
r
a
o
f
t
h
e
C
O
V
I
D
-
1
9
p
a
n
d
e
m
i
c
,
”
C
o
m
p
u
t
e
rs
,
M
a
t
e
ri
a
l
s
a
n
d
C
o
n
t
i
n
u
a
,
v
o
l
.
7
1
,
n
o
.
2
,
p
p
.
4
1
5
1
–
4
1
6
6
,
2
0
2
2
,
d
o
i
:
1
0
.
3
2
6
0
4
/
c
mc
.
2
0
2
2
.
0
1
7
8
6
5
.
[
1
0
]
C
.
C
a
o
e
t
a
l
.
,
“
A
n
i
m
p
r
o
v
e
d
F
a
st
e
r
R
-
C
N
N
f
o
r
s
mal
l
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
I
EEE
A
c
c
e
s
s
,
v
o
l
.
7
,
p
p
.
1
0
6
8
3
8
–
1
0
6
8
4
6
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
9
.
2
9
3
2
7
3
1
.
[
1
1
]
K
.
H
e
,
X
.
Z
h
a
n
g
,
S
.
R
e
n
,
a
n
d
J
.
S
u
n
,
“
D
e
e
p
r
e
s
i
d
u
a
l
l
e
a
r
n
i
n
g
f
o
r
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
p
p
.
7
7
0
–
7
7
8
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
0
.
[
1
2
]
D
.
A
v
o
l
a
e
t
a
l
.
,
“
M
S
-
f
a
s
t
e
r
R
-
C
N
N
:
M
u
l
t
i
-
st
r
e
a
m
b
a
c
k
b
o
n
e
f
o
r
i
mp
r
o
v
e
d
f
a
st
e
r
R
-
C
N
N
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
n
d
a
e
r
i
a
l
t
r
a
c
k
i
n
g
f
r
o
m
U
A
V
i
ma
g
e
s,
”
R
e
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
3
,
n
o
.
9
,
p
p
.
1
–
1
8
,
2
0
2
1
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
3
0
9
1
6
7
0
.
[
1
3
]
I
.
C
.
D
u
t
a
,
L.
Li
u
,
F
.
Zh
u
,
a
n
d
L
.
S
h
a
o
,
“
I
mp
r
o
v
e
d
r
e
si
d
u
a
l
n
e
t
w
o
r
k
s
f
o
r
i
ma
g
e
a
n
d
v
i
d
e
o
r
e
c
o
g
n
i
t
i
o
n
,
”
Pr
o
c
e
e
d
i
n
g
s
-
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
P
a
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
p
p
.
9
4
1
5
–
9
4
2
2
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
P
R
4
8
8
0
6
.
2
0
2
1
.
9
4
1
2
1
9
3
.
[
1
4
]
D
.
D
.
A
b
o
y
o
mi
a
n
d
C
.
D
a
n
i
e
l
,
“
A
c
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
mo
d
e
r
n
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
ms
:
Y
O
LO
v
s.
S
S
D
v
s.
F
a
st
e
r
R
-
C
N
N
,
”
I
T
E
J
(
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
En
g
i
n
e
e
ri
n
g
J
o
u
r
n
a
l
s)
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
9
6
–
1
0
6
,
2
0
2
3
,
d
o
i
:
1
0
.
2
4
2
3
5
/
i
t
e
j
.
v
8
i
2
.
1
2
3
.
[
1
5
]
L.
A
f
k
a
a
r
,
“
D
a
t
a
set
s
S
I
B
I
si
g
n
l
a
n
g
u
a
g
e
a
l
p
h
a
b
e
t
s,”
K
a
g
g
l
e
,
2
0
2
1
.
h
t
t
p
s:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m/
d
a
t
a
s
e
t
s
/
ml
a
n
a
n
g
a
f
k
a
a
r
/
d
a
t
a
s
e
t
s
-
l
e
m
l
i
t
b
a
n
g
-
si
b
i
-
a
l
p
h
a
b
e
t
s
(
a
c
c
e
ss
e
d
M
a
y
3
0
,
2
0
2
4
)
.
[
1
6
]
M
.
M
a
i
t
y
,
S
.
B
a
n
e
r
j
e
e
,
a
n
d
S
.
S
i
n
h
a
C
h
a
u
d
h
u
r
i
,
“
F
a
st
e
r
R
-
C
N
N
a
n
d
Y
O
LO
b
a
se
d
v
e
h
i
c
l
e
d
e
t
e
c
t
i
o
n
:
A
su
r
v
e
y
,
”
Pro
c
e
e
d
i
n
g
s
-
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
i
n
g
Me
t
h
o
d
o
l
o
g
i
e
s
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
,
I
C
C
M
C
2
0
2
1
,
n
o
.
A
p
r
i
l
,
p
p
.
1
4
4
2
–
1
4
4
7
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
M
C
5
1
0
1
9
.
2
0
2
1
.
9
4
1
8
2
7
4
.
[
1
7
]
Y
.
S
u
,
D
.
Li
,
a
n
d
X
.
C
h
e
n
,
“
L
u
n
g
n
o
d
u
l
e
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
F
a
st
e
r
R
-
C
N
N
f
r
a
mew
o
r
k
,
”
C
o
m
p
u
t
e
r
M
e
t
h
o
d
s
a
n
d
Pr
o
g
ra
m
s
i
n
Bi
o
m
e
d
i
c
i
n
e
,
v
o
l
.
2
0
0
,
p
.
1
0
5
8
6
6
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
m
p
b
.
2
0
2
0
.
1
0
5
8
6
6
.
[
1
8
]
R
.
G
i
r
s
h
i
c
k
,
J.
D
o
n
a
h
u
e
,
T.
D
a
r
r
e
l
l
,
a
n
d
J.
M
a
l
i
k
,
“
R
i
c
h
f
e
a
t
u
r
e
h
i
e
r
a
r
c
h
i
e
s
f
o
r
a
c
c
u
r
a
t
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
n
d
sem
a
n
t
i
c
seg
m
e
n
t
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
4
,
p
p
.
5
8
0
–
5
8
7
.
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
4
.
8
1
.
[
1
9
]
R
.
G
i
r
sh
i
c
k
,
“
F
a
st
R
-
C
N
N
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EE
E
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
,
D
e
c
.
2
0
1
5
,
v
o
l
.
2
0
1
5
I
n
t
e
r
,
p
p
.
1
4
4
0
–
1
4
4
8
.
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
1
5
.
1
6
9
.
[
2
0
]
M
.
M
.
M
i
j
w
i
l
,
K
.
A
g
g
a
r
w
a
l
,
R
.
D
o
s
h
i
,
K
.
K
.
H
i
r
a
n
,
a
n
d
M
.
G
ö
k
,
“
T
h
e
d
i
st
i
n
c
t
i
o
n
b
e
t
w
e
e
n
R
-
C
N
N
a
n
d
F
a
st
R
-
C
N
N
i
n
i
ma
g
e
a
n
a
l
y
si
s
:
A
p
e
r
f
o
r
ma
n
c
e
c
o
m
p
a
r
i
s
o
n
,
”
As
i
a
n
J
o
u
r
n
a
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
0
,
n
o
.
5
,
2
0
2
2
,
d
o
i
:
1
0
.
2
4
2
0
3
/
a
j
a
s.v
1
0
i
5
.
7
0
6
4
.
[
2
1
]
Z.
L
i
,
C
.
P
e
n
g
,
G
.
Y
u
,
X
.
Z
h
a
n
g
,
Y
.
D
e
n
g
,
a
n
d
J.
S
u
n
,
“
D
e
t
N
e
t
:
A
b
a
c
k
b
o
n
e
n
e
t
w
o
r
k
f
o
r
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
a
rXi
v
p
re
p
ri
n
t
a
rXi
v
:
1
8
0
4
.
0
6
2
1
5
,
p
p
.
1
–
1
7
,
2
0
1
8
.
[
2
2
]
D
.
S
a
r
w
i
n
d
a
,
R
.
H
.
P
a
r
a
d
i
s
a
,
A
.
B
u
st
a
mam
,
a
n
d
P
.
A
n
g
g
i
a
,
“
D
e
e
p
l
e
a
r
n
i
n
g
i
n
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
r
e
si
d
u
a
l
n
e
t
w
o
r
k
(
R
e
sN
e
t
)
v
a
r
i
a
n
t
s
f
o
r
d
e
t
e
c
t
i
o
n
o
f
c
o
l
o
r
e
c
t
a
l
c
a
n
c
e
r
,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
7
9
,
n
o
.
2
0
1
9
,
p
p
.
4
2
3
–
4
3
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s
.
2
0
2
1
.
0
1
.
0
2
5
.
[
2
3
]
L.
H
.
S
h
e
h
a
b
,
O
.
M
.
F
a
h
m
y
,
S
.
M
.
G
a
sser,
a
n
d
M
.
S
.
El
-
M
a
h
a
l
l
a
w
y
,
“
A
n
e
f
f
i
c
i
e
n
t
b
r
a
i
n
t
u
m
o
r
i
m
a
g
e
se
g
me
n
t
a
t
i
o
n
b
a
se
d
o
n
d
e
e
p
r
e
si
d
u
a
l
n
e
t
w
o
r
k
s
(
R
e
sN
e
t
s)
,
”
J
o
u
rn
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
r
si
t
y
-
E
n
g
i
n
e
e
ri
n
g
S
c
i
e
n
c
e
s
,
v
o
l
.
3
3
,
n
o
.
6
,
p
p
.
4
0
4
–
4
1
2
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
su
e
s.
2
0
2
0
.
0
6
.
0
0
1
.
[
2
4
]
T.
L
i
a
n
g
e
t
a
l
.
,
“
C
B
N
e
t
:
A
c
o
m
p
o
si
t
e
b
a
c
k
b
o
n
e
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
m
a
g
e
Pro
c
e
ssi
n
g
,
v
o
l
.
3
1
,
p
p
.
6
8
9
3
–
6
9
0
6
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
P
.
2
0
2
2
.
3
2
1
6
7
7
1
.
[
2
5
]
I
.
S
l
i
men
e
,
I
.
M
e
ss
a
o
u
d
i
,
A
.
E.
O
u
e
s
l
a
t
i
,
a
n
d
Z
.
La
c
h
i
r
i
,
“
C
a
n
c
e
r
d
i
s
e
a
s
e
mu
l
t
i
n
o
mi
a
l
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
a
n
d
S
V
M
o
n
t
h
e
g
e
n
e
s’
se
q
u
e
n
c
e
s,”
E
AI
En
d
o
rs
e
d
T
r
a
n
s
a
c
t
i
o
n
s
o
n
Pe
r
v
a
s
i
v
e
H
e
a
l
t
h
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
9
,
n
o
.
1
,
p
p
.
1
–
1
1
,
2
0
2
3
,
d
o
i
:
1
0
.
4
1
0
8
/
e
e
t
p
h
t
.
9
.
3
2
2
0
.
[
2
6
]
Y
.
Zh
a
n
g
,
J
.
L
i
,
S
.
W
e
i
,
F
.
Z
h
o
u
,
a
n
d
D
.
Li
,
“
H
e
a
r
t
b
e
a
t
s
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
s
i
n
g
h
y
b
r
i
d
t
i
me
-
f
r
e
q
u
e
n
c
y
a
n
a
l
y
s
i
s
a
n
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
b
a
s
e
d
o
n
R
e
sN
e
t
,
”
I
EEE
J
o
u
rn
a
l
o
f
Bi
o
m
e
d
i
c
a
l
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
5
,
n
o
.
1
1
,
p
p
.
4
1
7
5
–
4
1
8
4
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JB
H
I
.
2
0
2
1
.
3
0
8
5
3
1
8
.
[
2
7
]
P.
-
J.
G
o
h
,
M
.
-
H
.
H
o
o
,
a
n
d
K
.
-
C
.
K
h
o
r
,
“
Ev
a
l
u
a
t
i
n
g
d
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
mo
d
e
l
s
f
o
r
d
e
t
e
c
t
i
n
g
v
a
r
i
o
u
s
f
a
c
e
m
a
sk
w
e
a
r
i
n
g
s,”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
o
f
t
C
o
m
p
u
t
i
n
g
a
n
d
D
a
t
a
Mi
n
i
n
g
,
2
0
2
4
,
p
p
.
4
3
–
5
2
.
[
2
8
]
A
.
Th
a
r
w
a
t
,
“
C
l
a
ss
i
f
i
c
a
t
i
o
n
a
ssessm
e
n
t
m
e
t
h
o
d
s
,
”
A
p
p
l
i
e
d
C
o
m
p
u
t
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
7
,
n
o
.
1
,
p
p
.
1
6
8
–
1
9
2
,
Ju
l
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
c
i
.
2
0
1
8
.
0
8
.
0
0
3
.
[
2
9
]
O
.
C
a
e
l
e
n
,
“
A
B
a
y
e
s
i
a
n
i
n
t
e
r
p
r
e
t
a
t
i
o
n
o
f
t
h
e
c
o
n
f
u
s
i
o
n
m
a
t
r
i
x
,
”
A
n
n
a
l
s
o
f
M
a
t
h
e
m
a
t
i
c
s
a
n
d
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
8
1
,
n
o
.
3
–
4
,
p
p
.
4
2
9
–
4
5
0
,
2
0
1
7
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
0
4
7
2
-
017
-
9
5
6
4
-
8.
[
3
0
]
S
.
R
u
u
s
k
a
,
W
.
H
ä
mäl
ä
i
n
e
n
,
S
.
K
a
j
a
v
a
,
M
.
M
u
g
h
a
l
,
P
.
M
a
t
i
l
a
i
n
e
n
,
a
n
d
J
.
M
o
n
o
n
e
n
,
“
Ev
a
l
u
a
t
i
o
n
o
f
t
h
e
c
o
n
f
u
s
i
o
n
mat
r
i
x
m
e
t
h
o
d
i
n
t
h
e
v
a
l
i
d
a
t
i
o
n
o
f
a
n
a
u
t
o
m
a
t
e
d
s
y
s
t
e
m
f
o
r
me
a
su
r
i
n
g
f
e
e
d
i
n
g
b
e
h
a
v
i
o
u
r
o
f
c
a
t
t
l
e
,
”
Be
h
a
v
i
o
u
ra
l
Pr
o
c
e
sses
,
v
o
l
.
1
4
8
,
p
p
.
5
6
–
6
2
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
e
p
r
o
c
.
2
0
1
8
.
0
1
.
0
0
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.