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
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
831
~
840
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
8
3
1
-
8
4
0
831
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Ada
ptive
de
form
a
ble f
ea
tur
e aug
menta
tion a
nd
ref
inement
network
for scen
e
t
e
x
t
d
etec
tion a
n
d recog
nition
Ra
t
na
m
a
la
S
.
P
a
t
il,
G
ee
t
a
H
a
nji,
Ra
k
esh
H
ud
ud
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s
a
n
d
C
o
mm
u
n
i
c
a
t
i
o
n
,
P
o
o
j
y
a
D
o
d
d
a
p
p
a
A
p
p
a
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
K
a
l
a
b
u
r
a
g
i
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
7
,
2
0
2
4
R
ev
is
ed
De
c
2
6
,
2
0
2
5
Acc
ep
ted
J
an
1
0
,
2
0
2
6
S
c
e
n
e
tex
t
re
c
o
g
n
it
io
n
(
S
TR)
is
th
e
tas
k
o
f
d
e
tec
ti
n
g
a
n
d
i
d
e
n
ti
fy
in
g
te
x
t
with
in
ima
g
e
s
c
a
p
tu
re
d
fr
o
m
n
a
tu
ra
l
sc
e
n
e
s,
a
c
h
a
ll
e
n
g
in
g
p
ro
c
e
ss
d
u
e
to
v
a
riatio
n
s
i
n
tex
t
a
p
p
e
a
ra
n
c
e
,
o
ri
e
n
tatio
n
,
a
n
d
b
a
c
k
g
r
o
u
n
d
c
o
m
p
l
e
x
it
y
.
Th
e
p
ro
p
o
se
d
m
e
th
o
d
o
lo
g
y
,
a
d
a
p
ti
v
e
d
e
fo
rm
a
b
le
fe
a
tu
re
a
u
g
m
e
n
tatio
n
a
n
d
re
fin
e
m
e
n
t
n
e
two
r
k
(AD
F
ARN
),
is
d
e
sig
n
e
d
to
a
d
d
re
ss
th
e
se
c
h
a
l
len
g
e
s
b
y
c
o
m
b
in
i
n
g
d
e
fo
rm
a
b
le
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
two
rk
s
fo
r
r
o
b
u
st
e
n
h
a
n
c
e
d
fe
a
tu
re
e
x
trac
ti
o
n
with
a
n
o
v
e
l
d
e
e
p
fe
a
tu
re
re
fin
e
m
e
n
t
(
F
RE)
th
a
t
lev
e
ra
g
e
s
re
fin
e
m
e
n
t
fo
r
p
re
c
ise
tex
t
lo
c
a
li
z
a
ti
o
n
.
T
h
is
a
p
p
r
o
a
c
h
e
n
h
a
n
c
e
s
th
e
d
iffere
n
ti
a
ti
o
n
b
e
twe
e
n
te
x
t
a
n
d
b
a
c
k
g
r
o
u
n
d
,
si
g
n
if
ica
n
tl
y
i
m
p
ro
v
i
n
g
re
c
o
g
n
it
i
o
n
a
c
c
u
ra
c
y
.
Th
e
AD
F
A
RN m
e
th
o
d
o
lo
g
y
i
n
c
lu
d
e
s a
c
o
m
p
re
h
e
n
siv
e
p
ro
c
e
ss
o
f
fe
a
tu
re
e
x
trac
ti
o
n
,
d
e
e
p
fe
a
tu
re
a
u
g
m
e
n
tati
o
n
m
o
d
u
le
(DFAM
),
a
n
d
th
e
g
e
n
e
ra
ti
o
n
o
f
sc
o
re
a
n
d
t
h
re
sh
o
l
d
m
a
p
s
th
r
o
u
g
h
d
if
fe
re
n
ti
a
b
le
b
in
a
riza
ti
o
n
.
T
h
e
a
d
a
p
ti
v
e
n
a
tu
r
e
o
f
th
e
m
o
d
e
l
a
ll
o
ws
it
to
h
a
n
d
le
l
o
w
-
re
so
lu
ti
o
n
a
n
d
p
a
rti
a
ll
y
o
c
c
lu
d
e
d
tex
t
e
ffe
c
ti
v
e
ly
,
f
u
rth
e
r
in
c
r
e
a
sin
g
it
s
ro
b
u
stn
e
ss
.
Ad
d
i
ti
o
n
a
ll
y
,
th
e
p
r
o
p
o
se
d
m
e
th
o
d
a
l
i
g
n
s
v
isu
a
l
a
n
d
tex
t
u
a
l
fe
a
tu
re
s
se
a
m
les
sly
.
Ex
te
n
siv
e
p
e
rfo
rm
a
n
c
e
e
v
a
lu
a
ti
o
n
o
n
th
e
c
o
m
m
o
n
o
b
jec
ts
in
c
o
n
tex
t
(
COCO
)
-
T
e
x
t
da
tas
e
t
d
e
m
o
n
stra
tes
th
a
t
AD
F
A
RN
o
u
t
p
e
rfo
rm
s
e
x
isti
n
g
sta
te
-
of
-
th
e
-
a
rt
m
e
th
o
d
s
i
n
term
s
o
f
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
-
sc
o
re
s,
e
sta
b
li
sh
i
n
g
it
a
s
a
h
ig
h
ly
e
ffe
c
ti
v
e
s
o
lu
t
io
n
fo
r
S
T
R
in
re
a
l
-
wo
rld
a
p
p
li
c
a
ti
o
n
s.
K
ey
w
o
r
d
s
:
ADFAR
N
D
e
e
p
f
e
a
t
u
r
e
a
u
g
m
e
n
t
a
t
i
o
n
m
o
d
u
l
e
Dee
p
f
ea
tu
r
e
r
e
f
in
em
en
t
Scen
e
tex
t r
ec
o
g
n
itio
n
T
ex
t r
ec
o
g
n
itio
n
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
:
R
atn
am
ala
S
.
Patil
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
,
Po
o
jy
a
Do
d
d
ap
p
a
Ap
p
a
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
Kala
b
u
r
ag
i,
I
n
d
ia
E
m
ail:
r
atn
am
ala_
1
2
@
r
ed
i
f
f
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
ex
ts
p
lay
an
im
p
o
r
tan
t
r
o
le
in
th
e
cu
ltu
r
al
tr
an
s
m
is
s
io
n
p
r
o
ce
s
s
s
in
ce
th
ey
ar
e
a
s
to
r
eh
o
u
s
e
o
f
h
u
m
an
wis
d
o
m
.
Sp
o
k
en
lan
g
u
ag
e
h
as
b
ec
o
m
e
a
m
o
r
e
p
o
wer
f
u
l
m
ed
iu
m
f
o
r
th
e
d
ev
el
o
p
m
en
t
o
f
h
u
m
an
civ
ilizatio
n
b
ec
au
s
e
it
h
as
b
r
o
k
en
d
o
wn
b
o
u
n
d
a
r
ies
r
elate
d
to
tim
e
a
n
d
s
p
ac
e.
No
wa
d
ay
s
,
a
lar
g
e
am
o
u
n
t
o
f
tex
tu
al
d
ata
is
s
to
r
ed
d
ig
itally
as
d
o
cu
m
en
ts
,
m
o
v
ies,
o
r
p
h
o
to
s
.
As
a
r
esu
lt,
u
s
in
g
co
m
p
u
ter
tech
n
o
lo
g
y
f
o
r
s
ce
n
e
tex
t
p
ictu
r
e
d
etec
tio
n
an
d
en
d
-
to
-
en
d
id
en
tific
atio
n
is
m
o
r
e
ess
en
tial
th
an
ev
e
r
[
1
]
.
T
h
er
ef
o
r
e
,
it
is
m
o
r
e
im
p
o
r
tan
t
th
a
n
ev
er
t
o
u
s
e
co
m
p
u
ter
tech
n
o
lo
g
y
f
o
r
en
d
-
to
-
en
d
id
en
tific
atio
n
an
d
s
ce
n
e
te
x
t
p
ictu
r
e
d
etec
tio
n
[
1
]
.
Ap
p
licatio
n
s
f
o
r
tex
t
r
ec
o
g
n
itio
n
ca
n
b
e
f
o
u
n
d
i
n
m
an
y
d
if
f
er
e
n
t
ar
ea
s
,
s
u
ch
as
ass
is
t
iv
e
tech
n
o
l
o
g
y
f
o
r
th
e
b
lin
d
,
d
r
iv
in
g
ass
is
tan
ce
,
an
d
h
an
d
wr
itin
g
r
ec
o
g
n
itio
n
.
Scan
n
ed
d
o
cu
m
en
t
r
ec
o
g
n
it
io
n
an
d
s
ce
n
e
tex
t
r
ec
o
g
n
itio
n
(
STR)
ar
e
th
e
two
m
ain
s
u
b
ca
teg
o
r
ies
o
f
tex
t
r
e
co
g
n
itio
n
.
Desp
ite
th
e
im
p
r
ess
iv
e
ad
v
an
ce
m
en
ts
in
s
o
f
twar
e
-
d
ef
i
n
ed
r
ad
io
s
(
S
DR
)
,
STR
is
s
t
ill
a
d
if
f
icu
lt
t
ask
.
Nu
m
er
o
u
s
f
ac
to
r
s
,
s
u
c
h
as
s
lan
ted
letter
in
g
,
lin
g
u
is
tic
v
ar
ian
ce
s
,
p
o
o
r
im
a
g
e
q
u
ality
,
v
ar
ied
ty
p
e
f
ac
es,
an
d
u
n
iq
u
e
tex
t
f
o
r
m
s
,
co
n
tr
i
b
u
te
to
th
is
is
s
u
e.
STR,
a
b
r
an
ch
o
f
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
,
s
ee
k
s
to
p
r
ec
is
ely
f
in
d
an
d
id
en
tify
ch
a
r
ac
ter
s
in
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
831
-
8
4
0
832
im
ag
es
o
f
s
itu
atio
n
s
,
i
n
clu
d
in
g
tex
t
th
at
ap
p
ea
r
s
o
n
b
illb
o
ar
d
s
.
T
ex
t
t
h
at
is
twis
ted
o
r
d
ef
o
r
m
e
d
,
c
o
m
p
le
x
b
ac
k
g
r
o
u
n
d
s
,
an
d
v
ar
io
u
s
f
o
n
t
s
ty
les ar
e
am
o
n
g
th
e
f
ac
to
r
s
t
h
at
lead
to
co
m
p
u
ter
v
is
io
n
is
s
u
es.
T
ex
t
r
ep
r
esen
tatio
n
tec
h
n
iq
u
e
s
ar
e
co
n
tin
u
ally
ev
o
lv
in
g
to
k
ee
p
u
p
with
ad
v
an
c
es
in
in
f
o
r
m
atio
n
tech
n
o
lo
g
y
.
Mo
r
e
s
p
ec
if
ically
,
ef
f
ec
tiv
e
co
m
m
u
n
icatio
n
an
d
in
f
o
r
m
atio
n
ac
ce
s
s
d
ep
en
d
m
o
r
e
an
d
m
o
r
e
o
n
th
e
ab
ilit
y
to
r
ec
o
g
n
ize
tex
t
f
r
o
m
s
tar
t
to
f
in
is
h
an
d
to
id
en
t
if
y
it
in
co
n
tex
t.
As
a
r
esu
lt,
m
u
ch
m
o
r
e
r
esear
ch
in
th
is
p
ar
ticu
lar
f
ield
is
r
eq
u
ir
ed
.
Dee
p
lear
n
in
g
tech
n
iq
u
es
ar
e
b
ein
g
u
s
ed
b
ec
au
s
e
STR
r
esear
ch
er
s
ar
e
b
ein
g
u
s
ed
f
r
eq
u
en
tly
[
2
]
.
T
e
x
t
in
n
atu
r
al
s
ce
n
e
p
h
o
to
s
ca
n
b
e
ch
allen
g
in
g
to
id
en
tify
d
u
e
to
its
in
co
n
s
is
ten
t
n
atu
r
e,
ex
tr
em
e
b
lu
r
r
in
g
,
p
e
r
s
p
ec
tiv
e
d
is
to
r
tio
n
s
,
an
d
d
i
v
er
s
e
ch
ar
ac
ter
.
Fo
llo
win
g
th
e
r
e
s
u
r
g
en
ce
o
f
n
eu
r
al
n
etwo
r
k
s
an
d
im
p
r
o
v
em
en
ts
in
p
u
b
licly
ac
ce
s
s
ib
le
v
is
io
n
d
atasets
,
th
e
co
m
p
u
ter
v
is
io
n
c
o
m
m
u
n
ity
h
as
d
em
o
n
s
tr
ated
a
s
tr
o
n
g
in
ter
est in
th
e
r
esear
ch
to
p
ics o
f
tex
t r
ec
o
g
n
itio
n
f
r
o
m
lo
w
-
r
eso
lu
tio
n
im
ag
es a
n
d
s
ce
n
e
tex
t
id
en
tific
atio
n
o
f
ir
r
eg
u
lar
tex
t
f
r
o
m
n
at
u
r
alis
tic
p
h
o
to
s
.
T
h
e
f
ir
s
t
p
h
ases
o
f
s
o
p
h
is
ticated
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
h
av
e
b
ee
n
d
em
o
n
s
tr
ated
b
y
th
e
m
o
s
t
r
ec
en
t
I
n
te
r
n
atio
n
al
C
o
n
f
er
e
n
ce
o
n
Do
c
u
m
en
t
An
aly
s
is
an
d
R
ec
o
g
n
itio
n
R
o
b
u
s
t
R
ea
d
in
g
(
I
C
DAR
)
ch
allen
g
in
g
r
ea
d
in
g
ch
allen
g
es.
T
h
ese
d
ay
s
,
th
e
m
o
s
t
wid
ely
u
s
ed
d
ee
p
lear
n
in
g
r
ec
o
g
n
itio
n
tec
h
n
iq
u
es
a
r
e
p
h
o
to
r
ec
tific
ati
o
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
a
n
d
s
eq
u
en
ce
p
r
e
d
ictio
n
.
T
h
e
ac
cu
r
ac
y
o
f
tex
t
r
ec
o
g
n
i
tio
n
in
r
ea
l
-
wo
r
ld
s
itu
atio
n
s
h
as
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
with
th
e
u
s
e
o
f
d
ee
p
lear
n
in
g
in
STR
[
3
]
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
u
s
e
lo
ca
l
s
p
atial
in
f
o
r
m
atio
n
in
th
e
in
p
u
t
t
o
ef
f
icien
tly
u
n
co
v
e
r
h
id
d
e
n
p
atter
n
s
.
Ho
w
ev
er
,
in
th
e
f
ield
o
f
STR,
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
s
(
R
NNs)
ar
e
th
o
u
g
h
t
to
b
e
th
e
b
est
m
eth
o
d
f
o
r
ca
p
tu
r
in
g
co
n
tex
t
an
d
d
e
p
en
d
e
n
cy
in
s
eq
u
en
tial
d
ata
[
4
]
.
T
o
m
ak
e
p
r
ed
ictio
n
s
,
R
NNs,
m
ay
r
etain
an
d
u
tili
ze
p
ast
d
at
a
f
r
o
m
ea
r
lier
tim
e
s
tep
s
,
as
a
r
esu
lt,
th
ey
wo
r
k
well
with
s
eq
u
en
tial
in
p
u
t,
s
u
ch
as
tex
t
d
ata.
R
NN
s
ef
f
icien
tly
ca
p
tu
r
e
th
e
co
n
tex
tu
al
r
elatio
n
s
h
ip
s
b
etwe
en
elem
en
ts
in
S
T
R
task
s
,
al
lo
win
g
p
r
ec
is
e
tex
t
id
e
n
tific
atio
n
a
n
d
co
m
p
r
eh
en
s
io
n
.
T
e
x
t
is
f
r
e
q
u
en
tly
d
is
p
lay
ed
i
n
STR
as
a
p
atch
o
r
s
tr
in
g
o
f
ch
ar
ac
ter
s
.
C
o
n
v
er
s
ely
,
C
NN
s
s
h
o
w
co
m
p
eten
ce
in
id
en
tify
in
g
im
p
o
r
tan
t
v
is
u
al
ch
ar
ac
ter
is
tics
in
in
p
u
t
im
ag
es.
C
NNs
ar
e
ca
p
ab
le
o
f
h
ier
ar
c
h
ically
d
ev
elo
p
i
n
g
c
o
m
p
licated
r
ep
r
esen
tatio
n
s
a
n
d
ca
p
tu
r
in
g
lo
ca
l
s
p
atial
p
atter
n
s
.
C
o
n
v
o
lu
tio
n
al
lay
er
s
an
d
p
o
o
lin
g
tech
n
i
q
u
es
ar
e
em
p
lo
y
ed
f
o
r
th
is
[
5
]
.
T
h
e
f
o
llo
win
g
m
eth
o
d
s
ca
n
b
e
u
s
ed
t
o
id
en
ti
f
y
f
ea
tu
r
es
f
r
o
m
te
x
tu
al
im
ag
es
an
d
to
class
if
y
o
r
id
en
tify
o
b
jects
in
th
e
s
h
o
r
t
tan
d
em
r
ep
ea
ts
f
ield
.
Alth
o
u
g
h
d
ee
p
lear
n
in
g
wo
r
k
s
in
cr
ed
i
b
ly
well,
it
s
u
f
f
er
s
g
r
ea
tly
f
r
o
m
p
ar
tially
o
b
s
cu
r
e
d
o
r
p
o
o
r
-
q
u
ality
im
ag
es.
T
h
e
p
u
b
lic
d
atab
ases
co
n
tain
a
r
a
n
g
e
o
f
im
ag
e
ty
p
es,
s
u
ch
as
r
eg
u
lar
,
lo
w
-
r
eso
lu
tio
n
,
an
d
p
ar
tially
o
cc
lu
d
e
d
p
h
o
to
s
.
T
h
er
e
ar
e
s
ev
er
al
r
ea
s
o
n
s
wh
y
tex
t
g
r
ap
h
ics
with
lo
w
r
es
o
lu
tio
n
m
ig
h
t
ex
is
t.
On
e
ca
u
s
e
m
ig
h
t b
e
th
at
th
e
i
m
ag
e
was c
o
m
p
r
ess
ed
to
r
ed
u
ce
s
to
r
ag
e
s
p
ac
e
[
6
]
.
An
o
th
er
p
o
s
s
ib
ilit
y
is
th
at
th
e
p
ictu
r
e
was
tak
e
n
u
s
in
g
a
c
am
er
a
th
at
h
as
a
lim
ited
a
m
o
u
n
t
o
f
f
o
c
u
s
p
o
i
n
ts
.
I
n
r
ec
o
g
n
itio
n
s
y
s
tem
s
,
lo
w
-
r
eso
lu
tio
n
p
ictu
r
es
ar
e
o
f
ten
h
an
d
led
with
b
icu
b
ic
an
d
b
ilin
ea
r
in
ter
p
o
latio
n
m
eth
o
d
s
.
T
h
e
u
p
-
s
am
p
led
p
ictu
r
es
ar
e
s
till
o
u
t
o
f
f
o
c
u
s
.
Fu
r
th
er
m
o
r
e
,
alth
o
u
g
h
th
ese
t
ec
h
n
iq
u
es g
r
ea
tly
e
n
h
an
ce
p
er
f
o
r
m
an
ce
o
n
ty
p
ical
s
ce
n
e
tex
t,
th
ey
ar
e
u
n
ab
le
to
y
ield
s
atis
f
ac
to
r
y
o
u
tco
m
es
o
n
d
if
f
icu
lt
ir
r
e
g
u
lar
te
x
t,
wh
ich
h
as
lo
n
g
b
ee
n
a
p
r
o
b
lem
f
o
r
STR.
T
h
e
in
c
o
r
p
o
r
atio
n
o
f
d
ee
p
lea
r
n
in
g
m
eth
o
d
o
lo
g
ies
to
im
p
r
o
v
e
tex
t
id
en
tific
atio
n
an
d
r
ec
o
g
n
itio
n
i
n
n
atu
r
al
p
h
o
t
o
s
is
h
ig
h
lig
h
ted
in
th
is
p
ap
er
'
s
th
o
r
o
u
g
h
an
al
y
s
is
o
f
ad
v
an
ce
d
tech
n
i
q
u
es
in
STR.
T
h
e
s
tu
d
y
in
v
esti
g
ates
th
e
ef
f
icac
y
o
f
C
NNs
an
d
R
NNs
in
en
h
an
cin
g
tex
t
id
en
tific
atio
n
ac
c
u
r
ac
y
in
o
r
d
er
to
ad
d
r
ess
th
e
d
if
f
icu
lties
p
r
esen
ted
b
y
ir
r
eg
u
lar
tex
t
s
h
ap
es,
lo
w
im
ag
e
q
u
ality
,
an
d
co
m
p
licated
b
ac
k
d
r
o
p
s
.
A
n
ew
m
eth
o
d
is
p
r
esen
ted
th
at
in
co
r
p
o
r
ates
a
d
ee
p
f
ea
tu
r
e
a
u
g
m
e
n
tatio
n
m
o
d
u
le
(
DFAM)
an
d
d
ee
p
f
ea
tu
r
e
r
ef
in
e
m
en
t
m
o
d
u
le
(
DFR
M)
f
o
r
ac
cu
r
ate
tex
t
lo
ca
lizatio
n
,
alo
n
g
with
a
d
ef
o
r
m
a
b
le
co
n
v
o
lu
tio
n
al
n
et
wo
r
k
f
o
r
im
p
r
o
v
ed
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
m
eth
o
d
o
lo
g
y
in
clu
d
es
a
co
m
p
lex
f
e
atu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
,
th
e
DFAM,
an
d
th
e
u
s
e
o
f
d
if
f
e
r
en
tiab
le
b
in
a
r
izatio
n
to
cr
ea
te
s
co
r
e
an
d
th
r
esh
o
ld
m
ap
s
.
T
h
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
s
u
g
g
ested
STR
tech
n
iq
u
es
in
p
r
ac
tical
ap
p
lic
atio
n
s
is
d
em
o
n
s
tr
ated
b
y
ex
t
en
s
iv
e
ex
p
er
im
e
n
ts
ca
r
r
ied
o
u
t
o
n
th
e
co
m
m
o
n
o
b
jects
in
co
n
te
x
t
(
C
OC
O
)
-
T
ex
t
d
ataset,
wh
ich
s
h
o
w
n
o
ta
b
le
g
ain
s
in
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
es
wh
e
n
co
m
p
ar
ed
to
cu
r
r
en
t state
-
of
-
t
h
e
-
ar
t m
eth
o
d
s
.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
p
ap
er
ca
n
b
e
s
u
m
m
ar
ized
as f
o
llo
ws:
i)
E
n
h
an
ce
d
tex
t
lo
ca
lizatio
n
:
t
h
e
p
r
o
p
o
s
ed
ad
a
p
tiv
e
d
ef
o
r
m
ab
le
f
ea
tu
r
e
au
g
m
en
tatio
n
a
n
d
r
ef
i
n
em
en
t
n
etwo
r
k
(
ADF
A
R
N)
m
eth
o
d
o
lo
g
y
in
tr
o
d
u
ce
s
a
n
o
v
el
d
ee
p
f
ea
tu
r
e
r
e
f
in
em
en
t
(
FR
E
)
th
at
s
ig
n
if
ican
tly
im
p
r
o
v
es tex
t lo
ca
lizatio
n
b
y
l
ev
er
ag
in
g
r
ef
in
em
e
n
t
.
ii)
R
o
b
u
s
t
en
h
an
ce
d
f
ea
tu
r
e
ex
t
r
ac
tio
n
:
ADF
A
R
N
u
tili
ze
s
a
d
ef
o
r
m
ab
le
co
n
v
o
lu
tio
n
al
n
etw
o
r
k
to
p
er
f
o
r
m
en
h
an
ce
d
f
ea
tu
r
e
e
x
tr
ac
tio
n
,
c
ap
tu
r
in
g
i
n
tr
icate
tex
t p
atter
n
s
ac
r
o
s
s
v
ar
io
u
s
s
ca
les an
d
r
eso
l
u
tio
n
s
.
iii)
State
-
of
-
th
e
-
ar
t
p
er
f
o
r
m
an
ce
:
ADF
A
R
N
o
u
tp
er
f
o
r
m
s
c
u
r
r
en
t
s
tate
-
of
-
th
e
-
a
r
t
tech
n
i
q
u
e
s
in
ter
m
s
o
f
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
s
af
ter
th
o
r
o
u
g
h
test
in
g
o
n
t
h
e
C
OC
O
-
T
ex
t
d
ataset.
A
s
tr
o
n
g
an
d
ef
f
ec
tiv
e
tex
t
r
ec
o
g
n
itio
n
s
y
s
tem
is
p
r
o
d
u
ce
d
b
y
co
m
b
in
in
g
im
p
r
o
v
e
d
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
b
o
u
n
d
ar
y
au
g
m
en
tatio
n
ap
p
r
o
ac
h
es,
estab
lis
h
in
g
a
n
ew
s
tan
d
ar
d
in
th
e
f
ield
o
f
STR
.
T
h
is
p
ap
er
'
s
r
esear
ch
is
d
iv
i
d
e
d
in
to
f
o
u
r
s
ec
tio
n
s
:
a
q
u
ick
s
u
m
m
ar
y
is
co
v
er
e
d
in
th
e
s
ec
t
io
n
1
,
an
d
r
elate
d
wo
r
k
is
co
v
er
e
d
in
th
e
s
ec
tio
n
2
.
C
r
ea
tin
g
a
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
is
th
e
f
o
cu
s
o
f
th
e
s
ec
tio
n
3
.
T
h
e
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
is
co
v
er
ed
in
th
e
s
ec
tio
n
4
,
wh
e
r
e
th
e
f
in
d
in
g
s
ar
e
d
is
p
lay
ed
as tab
les an
d
g
r
ap
h
s
.
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
A
d
a
p
tive
d
efo
r
ma
b
le
fea
tu
r
e
a
u
g
men
ta
tio
n
a
n
d
r
efin
eme
n
t n
etw
o
r
k
fo
r
s
ce
n
e
text
… (
R
a
tn
a
ma
la
S
.
P
a
til
)
833
2.
RE
L
AT
E
D
WO
RK
T
h
e
id
e
n
tific
atio
n
o
f
te
x
t
in
a
n
y
f
o
r
m
at
ca
n
b
e
ac
h
iev
ed
b
y
u
tili
zin
g
a
n
e
n
co
d
e
r
th
at
lev
e
r
ag
es
lo
ca
l
d
ep
en
d
e
n
cy
m
o
d
elin
g
,
as
p
r
o
p
o
s
ed
b
y
L
ee
et
a
l
.
[
6
]
.
T
h
e
en
c
o
d
er
was
in
teg
r
ate
d
w
ith
an
ad
ap
tiv
e
2
D
s
elf
-
atten
tio
n
m
ec
h
an
is
m
to
ef
f
icien
tly
ca
p
tu
r
e
s
p
atial
in
ter
ac
tio
n
s
.
T
h
e
lim
itatio
n
o
f
tr
ain
in
g
s
p
atial
tr
an
s
f
o
r
m
er
n
etwo
r
k
(
STN
)
-
b
ased
ir
r
eg
u
lar
te
x
t
r
ec
o
g
n
itio
n
s
y
s
tem
s
is
d
is
cu
s
s
ed
b
y
C
h
en
g
et
a
l
.
[
7
]
.
T
h
e
m
eth
o
d
u
tili
ze
s
weig
h
t
co
m
b
i
n
atio
n
s
to
co
n
s
tr
u
ct
s
eq
u
en
ce
s
an
d
in
co
r
p
o
r
ates
f
ea
tu
r
e
ex
t
r
ac
tio
n
in
f
o
u
r
te
x
t
d
ir
ec
tio
n
s
.
T
h
e
ap
p
r
o
ac
h
u
s
ed
b
y
th
e
r
o
b
u
s
t
s
ca
n
n
er
[
8
]
to
r
e
d
u
ce
er
r
o
n
e
o
u
s
r
ec
o
g
n
itio
n
o
f
s
em
an
tic
-
f
r
ee
d
ata
in
v
o
lv
es
th
e
u
tili
za
tio
n
o
f
p
o
s
itio
n
-
en
h
an
ce
d
an
d
h
y
b
r
i
d
b
r
an
ch
es
in
th
e
d
ec
o
d
e
r
.
T
h
e
m
er
g
in
g
o
f
th
ese
b
r
an
ch
es
to
p
r
o
d
u
ce
p
r
e
d
ictio
n
r
esu
lts
is
ac
co
m
p
lis
h
ed
u
s
in
g
a
d
y
n
a
m
ic
f
u
s
io
n
m
o
d
u
le.
T
h
e
u
tili
za
tio
n
o
f
m
er
g
in
g
m
o
d
u
les
an
d
m
ix
i
n
g
b
lo
ck
s
was
im
p
lem
en
ted
b
y
Du
et
a
l
.
[
9
]
in
th
eir
s
tu
d
y
to
en
h
an
ce
th
e
p
r
o
ce
s
s
o
f
m
u
lti
-
g
r
an
u
lar
ity
f
ea
tu
r
e
e
x
tr
ac
tio
n
in
p
u
r
e
v
ir
t
u
al
m
ac
h
i
n
e
ar
ch
itectu
r
es.
T
h
e
u
tili
za
tio
n
o
f
th
is
p
ar
ticu
lar
ap
p
r
o
ac
h
r
esu
lted
in
a
n
im
p
r
o
v
ed
tr
ad
e
-
o
f
f
b
etwe
en
ac
cu
r
ac
y
an
d
p
er
f
o
r
m
an
ce
.
T
h
e
th
in
-
p
late
s
p
lin
e
(
T
PS
)
++
tr
an
s
f
o
r
m
atio
n
f
o
r
tex
t
co
r
r
ec
tio
n
,
k
n
o
wn
as
T
PS
++
,
was
f
ir
s
t
in
tr
o
d
u
ce
d
b
y
Z
h
en
g
et
a
l
.
[
1
0
]
.
T
h
e
atten
tio
n
tech
n
iq
u
e
is
em
p
lo
y
ed
b
y
T
PS
++
to
e
n
h
an
ce
th
e
p
r
ec
is
io
n
a
n
d
ad
ap
tab
ilit
y
o
f
tex
t
c
o
r
r
ec
tio
n
.
T
h
e
T
PS
++
s
y
s
tem
em
p
lo
y
e
d
a
s
im
u
ltan
eo
u
s
ass
ess
m
en
t
o
f
atten
tio
n
s
co
r
es
an
d
f
o
r
e
g
r
o
u
n
d
co
n
tr
o
l
p
o
in
ts
to
en
h
an
ce
th
e
r
ea
d
ab
ilit
y
an
d
n
atu
r
aln
ess
o
f
tex
t
r
ep
air
s
.
T
h
e
s
h
ar
in
g
o
f
th
e
r
ec
o
g
n
izer
'
s
f
ea
tu
r
e
b
ac
k
b
o
n
e
r
esu
lts
in
a
d
ec
r
ea
s
e
in
b
o
th
th
e
in
f
er
en
ce
tim
e
an
d
th
e
p
ar
am
eter
o
v
er
h
ea
d
.
T
h
e
g
r
a
p
h
-
b
ased
m
o
d
elin
g
a
p
p
r
o
ac
h
was
in
tr
o
d
u
ce
d
b
y
Yan
et
a
l
.
[
1
1
]
as
a
m
eth
o
d
f
o
r
ac
q
u
ir
in
g
b
asic
r
ep
r
esen
tatio
n
s
o
f
tex
t
g
r
ap
h
ics
f
r
o
m
s
ce
n
es.
T
o
tr
ain
th
ese
r
ep
r
esen
tatio
n
s
,
th
e
r
ese
ar
ch
er
s
d
ev
elo
p
e
d
weig
h
ted
ag
g
r
eg
ato
r
s
an
d
p
o
o
lin
g
tech
n
iq
u
es.
T
h
e
in
p
u
t
r
ep
r
esen
tatio
n
s
u
n
d
er
g
o
a
tr
a
n
s
f
o
r
m
atio
n
p
r
o
ce
s
s
u
s
in
g
g
r
a
p
h
c
o
n
v
o
lu
tio
n
al
n
etwo
r
k
s
,
r
esu
ltin
g
i
n
th
e
g
en
er
a
tio
n
o
f
m
o
r
e
in
tr
icate
v
is
u
al
t
ex
t
r
ep
r
esen
tatio
n
s
.
T
h
e
f
o
llo
win
g
wo
r
k
p
r
o
p
o
s
es
a
s
y
s
tem
atic
ap
p
r
o
ac
h
to
ad
d
r
ess
in
g
m
is
alig
n
m
en
t
p
r
o
b
lem
s
in
th
e
f
ield
o
f
tex
t
r
ec
o
g
n
itio
n
.
T
h
e
p
r
o
p
o
s
ed
te
ch
n
iq
u
e,
r
ef
e
r
r
ed
to
as
p
r
im
itiv
e
r
ep
r
esen
tatio
n
lear
n
in
g
n
etwo
r
k
with
2
D
atten
tio
n
(
PR
E
N2
D
)
,
is
an
en
co
d
er
-
d
ec
o
d
er
m
o
d
el
th
at
u
t
ilizes
a
2
D
atten
tio
n
m
ec
h
an
i
s
m
an
d
v
is
u
al
tex
t
r
ep
r
esen
tatio
n
s
.
T
h
e
tech
n
iq
u
e
em
p
lo
y
ed
i
n
th
is
a
p
p
r
o
ac
h
u
tili
ze
s
ch
ar
ac
ter
-
by
-
c
h
ar
ac
t
er
id
en
tific
atio
n
to
d
ec
r
ea
s
e
th
e
s
p
ee
d
o
f
p
r
o
ce
s
s
in
g
.
T
h
e
d
ec
o
u
p
led
atten
tio
n
n
etwo
r
k
was
in
tr
o
d
u
ce
d
b
y
W
an
g
et
a
l
.
[
1
2
]
to
ad
d
r
ess
th
e
ch
allen
g
es
o
f
alig
n
m
en
t
an
d
h
is
to
r
ical
d
ec
o
d
in
g
in
STR
.
T
h
e
d
ee
p
alig
n
m
en
t
n
etwo
r
k
c
o
n
s
is
ts
o
f
th
r
ee
p
r
im
a
r
y
c
o
m
p
o
n
en
ts
:
a
f
ea
tu
r
e
e
n
co
d
e
r
,
a
d
ec
o
u
p
le
d
tex
t
d
ec
o
d
er
,
an
d
a
c
o
n
v
o
lu
tio
n
al
alig
n
m
e
n
t
m
o
d
u
le.
T
h
e
d
etac
h
m
e
n
t
alig
n
m
en
t
n
etwo
r
k
en
h
a
n
ce
s
th
e
ac
cu
r
ac
y
an
d
f
le
x
ib
ilit
y
o
f
t
ex
t
r
ec
o
g
n
itio
n
b
y
is
o
latin
g
th
e
alig
n
m
e
n
t
p
r
o
ce
d
u
r
e.
T
h
e
ex
p
er
im
en
ts
co
n
d
u
cted
o
n
te
x
t
-
lik
e
s
o
u
n
d
p
atter
n
s
r
ev
ea
led
th
at
th
e
m
eth
o
d
e
n
co
u
n
ter
ed
d
if
f
icu
lti
es
in
ac
cu
r
ately
alig
n
i
n
g
th
e
tex
t.
Dee
lak
a
et
a
l
.
[
1
3
]
d
e
v
elo
p
ed
a
n
ew
m
o
d
el
ar
ch
itectu
r
e
th
at
in
co
r
p
o
r
ate
d
v
ar
io
u
s
v
is
u
al
f
ea
tu
r
e
en
c
o
d
i
n
g
an
d
f
ea
t
u
r
e
p
r
o
jectio
n
tec
h
n
iq
u
es.
T
h
e
m
o
d
el
p
r
o
d
u
ce
d
a
p
r
ed
eter
m
in
ed
s
et
o
f
item
lab
els
b
y
co
n
s
id
er
in
g
th
e
r
estricte
d
ch
ar
ac
ter
co
u
n
t
in
th
e
tr
ain
in
g
im
ag
es.
Ho
wev
er
,
th
e
s
y
s
tem
was
n
o
t
ca
p
ab
le
o
f
ac
cu
r
ately
f
o
r
ec
asti
n
g
th
e
p
o
s
itio
n
s
o
f
th
e
item
s
.
Fed
er
ate
d
lear
n
in
g
s
y
s
tem
s
aim
to
m
i
n
im
ize
p
ar
am
eter
s
p
ac
es
an
d
co
m
p
u
tatio
n
al
co
m
p
le
x
ity
to
ac
h
iev
e
ef
f
icien
t
tr
ain
in
g
an
d
r
ea
l
-
tim
e
in
f
er
e
n
ce
.
T
h
e
m
o
d
el
u
tili
ze
d
a
f
ea
tu
r
e
lo
ca
lizatio
n
u
n
it
an
d
an
en
co
d
er
th
at
r
elied
o
n
g
eo
m
etr
ic
s
h
ap
es
to
p
r
ed
ict
g
r
o
u
n
d
-
tr
u
th
lab
el
s
eq
u
en
ce
s
.
T
h
e
m
o
d
el
ass
u
m
ed
th
at
th
e
in
p
u
t
p
h
o
to
s
wer
e
ar
r
an
g
e
d
h
o
r
izo
n
tally
an
d
co
n
tain
ed
o
n
ly
o
n
e
r
o
w
o
f
tex
t.
T
h
e
tech
n
iq
u
e
is
s
p
ec
if
ically
d
esig
n
ed
to
h
a
n
d
le
n
u
m
er
ical
d
ata.
Ho
wev
e
r
,
t
h
e
u
s
e
o
f
u
n
e
x
p
ec
ted
o
r
ir
r
eg
u
lar
lan
g
u
ag
e
ca
n
s
ig
n
if
i
ca
n
tly
im
p
ac
t
th
e
ef
f
ec
tiv
en
ess
an
d
ef
f
icien
c
y
o
f
th
e
tech
n
i
q
u
e.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
[
1
4
]
aim
s
to
ac
h
iev
e
two
m
ain
o
b
jectiv
es:
en
h
an
cin
g
th
e
m
o
d
el'
s
s
en
s
itiv
ity
to
laten
t
f
ea
tu
r
es
an
d
ex
p
ed
itin
g
e
n
d
-
to
-
en
d
s
eq
u
en
ce
lear
n
in
g
f
o
r
Per
s
ian
d
ig
it
id
e
n
tific
atio
n
.
T
h
e
in
co
r
p
o
r
atio
n
o
f
a
co
n
v
o
l
u
tio
n
al
-
b
ased
m
o
d
el
th
at
co
m
b
in
es
th
e
ex
citatio
n
g
ate
with
s
q
u
ee
zin
g
en
ab
les th
e
ac
h
iev
e
m
en
t o
f
th
is
o
b
jectiv
e.
Fo
r
STR
,
o
r
v
is
u
al
co
llab
o
r
ati
o
n
an
d
d
u
al
-
s
tr
ea
m
f
u
s
io
n
(
V
OL
T
E
R
)
,
it is
s
tr
o
n
g
ly
ad
v
is
ed
to
em
p
lo
y
d
u
al
-
s
tr
ea
m
f
u
s
io
n
a
n
d
v
is
u
al
au
g
m
en
tatio
n
a
p
p
r
o
ac
h
es.
T
o
o
v
er
c
o
m
e
v
is
u
al
co
n
s
tr
ain
ts
an
d
en
h
an
ce
p
r
ed
ictiv
e
ca
p
a
b
ilit
ies,
th
e
f
ir
s
t
s
tep
is
to
d
ev
elo
p
a
m
u
l
ti
-
s
tag
e
lo
ca
l
-
g
lo
b
al
c
o
llab
o
r
atio
n
v
is
u
al
m
o
d
el
(
L
GC
-
VM
)
[
1
5
]
.
I
n
teg
r
atin
g
lo
ca
l
an
d
g
lo
b
al
elem
en
ts
at
v
ar
io
u
s
s
ca
les
is
th
is
p
ar
ad
i
g
m
'
s
m
ain
g
o
al.
A
v
is
io
n
-
lan
g
u
a
g
e
co
n
tr
asti
v
e
(
VL
C
)
m
o
d
u
le
is
o
u
r
s
y
s
tem
'
s
s
ec
o
n
d
f
ea
tu
r
e.
B
y
m
ak
in
g
it
p
o
s
s
ib
le
to
co
m
p
ar
e
th
e
r
ep
r
esen
tatio
n
s
o
f
b
o
th
l
an
g
u
ag
es,
th
is
m
o
d
u
le
aim
s
to
f
ac
ilit
ate
s
u
cc
es
s
f
u
l
lin
k
s
b
etwe
en
v
is
io
n
an
d
lan
g
u
ag
e.
Acc
u
r
ately
alig
n
i
n
g
th
e
f
ea
tu
r
e
s
p
ac
es
o
f
th
e
lan
g
u
ag
e
-
m
o
d
el
(
L
M)
a
n
d
v
is
io
n
-
m
o
d
el
(
VM
)
is
th
e
m
ain
g
o
al
.
I
n
a
d
d
itio
n
,
we
p
r
o
p
o
s
e
th
e
c
r
ea
tio
n
o
f
a
d
u
al
-
s
tr
ea
m
f
ea
tu
r
e
e
n
h
an
ce
m
e
n
t
(
DSFE)
m
o
d
u
le
to
s
o
lv
e
th
e
p
r
o
b
lem
o
f
s
y
n
ch
r
o
n
izin
g
s
ev
er
al
m
o
d
alities
an
d
o
f
f
er
a
s
m
o
o
th
er
in
teg
r
atio
n
.
Facilitatin
g
o
n
e
-
way
co
m
m
u
n
icatio
n
b
etwe
en
v
er
b
a
l a
n
d
v
is
u
al
elem
en
ts
is
th
e
ai
m
o
f
th
is
m
o
d
u
le.
T
h
e
ap
p
r
o
ac
h
f
o
r
te
x
t
r
ec
o
g
n
i
tio
n
is
r
ef
er
r
ed
to
as
p
r
o
to
ty
p
e
-
b
ased
u
n
s
u
p
er
v
is
ed
d
o
m
ain
ad
ap
tatio
n
(P
r
o
to
UDA
)
[
1
6
]
–
[
1
8
]
.
T
h
e
cl
ass
p
r
o
to
ty
p
es
ar
e
co
m
p
u
ted
u
s
in
g
th
e
s
o
u
r
ce
,
tar
g
et,
an
d
m
i
x
ed
(
s
o
u
r
ce
-
tar
g
et
)
d
o
m
ain
s
in
th
is
ap
p
r
o
ac
h
.
T
h
e
Pro
to
UDA
tech
n
iq
u
e
u
tili
ze
s
p
s
eu
d
o
-
lab
els
to
e
x
tr
ac
t
ch
a
r
a
cter
f
ea
tu
r
es
wh
ile
s
im
u
ltan
eo
u
s
ly
o
f
f
er
i
n
g
wo
r
d
-
lev
el
m
o
n
it
o
r
in
g
.
A
d
d
itio
n
ally
,
we
p
r
o
v
i
d
e
two
co
m
p
lem
en
tar
y
p
ar
alle
l
m
o
d
u
les
f
o
r
alig
n
m
e
n
t
at
b
o
th
th
e
i
n
s
tan
ce
an
d
class
lev
els.
T
h
e
p
u
r
p
o
s
e
o
f
th
ese
m
o
d
u
les
is
to
f
ac
ilit
ate
th
e
tr
an
s
f
er
o
f
d
ata
f
r
o
m
s
o
u
r
ce
d
o
m
ain
to
d
esti
n
atio
n
d
o
m
ai
n
,
u
tili
zin
g
s
p
ec
if
ic
ch
ar
ac
ter
f
ea
tu
r
es a
s
cr
iter
ia.
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
831
-
8
4
0
834
3.
M
E
T
H
O
D
T
h
e
p
r
o
ce
s
s
o
f
e
x
tr
ac
tin
g
an
d
f
u
s
in
g
en
h
a
n
ce
d
f
ea
t
u
r
es
b
eg
i
n
s
with
f
ee
d
in
g
an
i
n
p
u
t
p
ictu
r
e
in
to
th
e
d
ef
o
r
m
a
b
le
f
ea
tu
r
e
e
x
tr
ac
to
r
n
etwo
r
k
(
DFEN
)
m
o
d
u
le.
T
h
e
f
ea
tu
r
e
with
1
2
8
ch
a
n
n
els
is
p
r
o
d
u
ce
d
b
y
co
n
ca
ten
atin
g
th
e
f
u
s
ed
en
h
a
n
ce
d
f
ea
tu
r
es
af
te
r
th
ey
h
av
e
b
ee
n
u
p
s
am
p
le
d
to
1
/4
o
f
th
e
o
r
ig
in
al
im
ag
e'
s
s
ize.
Nex
t,
to
ex
tr
ac
t
r
ef
in
em
e
n
t
a
n
d
g
et
a
f
ea
tu
r
e
,
b
y
im
p
lem
en
ti
n
g
th
e
p
r
ec
is
e
DFR
M.
an
d
a
r
e
ad
d
ed
to
g
eth
er
elem
en
t
b
y
elem
e
n
t
to
g
et
th
e
u
s
ag
e.
A
p
r
ed
ictio
n
h
ea
d
is
f
ed
with
to
f
o
r
ec
ast
tex
t
an
d
n
o
n
-
te
x
t
s
co
r
e
m
ap
s
.
W
ith
in
p
u
t,
a
s
ec
o
n
d
p
r
e
d
ictio
n
h
ea
d
cr
ea
tes
th
e
th
r
esh
o
l
d
m
a
p
.
Ultim
ately
,
th
e
s
co
r
e
m
ap
an
d
t
h
r
esh
o
ld
m
a
p
u
s
e
d
if
f
er
en
tiab
le
b
i
n
ar
izatio
n
to
co
m
p
u
te
th
e
ap
p
r
o
x
im
at
e
b
in
ar
y
m
a
p
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
co
n
n
ec
te
d
to
th
e
n
etwo
r
k
d
u
r
in
g
th
e
tr
ain
in
g
p
h
ase
to
im
p
r
o
v
e
th
e
f
ea
t
u
r
e
r
ep
r
esen
tatio
n
s
.
E
v
er
y
n
etwo
r
k
n
o
d
e
is
tr
ain
e
d
f
r
o
m
s
tar
t to
f
in
is
h
.
Fig
u
r
e
1
s
h
o
ws th
e
p
r
o
p
o
s
ed
ADFAR
N
ar
ch
itectu
r
e
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
ADFAR
N
ar
ch
itectu
r
e
3
.
1
.
Def
o
rma
ble f
ea
t
ure
ex
t
ra
ct
o
r
net
wo
r
k
m
o
du
le
T
h
is
m
o
d
el
em
p
l
o
y
s
a
d
ef
o
r
m
ab
le
co
n
v
o
l
u
tio
n
al
n
etw
o
r
k
to
e
x
tr
ac
t
en
h
an
ce
d
f
e
atu
r
es
as
1
,
2
,
3
,
4
,
an
d
5
r
ep
r
esen
ts
v
ar
io
u
s
f
ea
t
u
r
e
m
ap
s
wh
er
ein
th
e
r
eso
lu
ti
o
n
s
r
ec
o
r
d
e
d
as
g
iv
en
as
1
/2
,
1
/4
,
1
/8
,
1
/
1
6
,
an
d
1
/3
2
f
o
r
t
h
e
in
p
u
t
s
ize
o
f
th
e
im
a
g
e
with
t
h
e
c
o
r
r
esp
o
n
d
in
g
ch
an
n
els
as
g
iv
e
n
b
y
6
4
,
2
5
6
,
5
1
2
,
1
0
2
4
,
an
d
2
0
4
8
.
T
h
e
m
o
d
el
p
r
o
v
id
es
an
alter
i
n
g
f
ield
f
o
r
th
e
m
o
d
el
wh
ich
b
en
ef
its
th
e
tex
t
in
s
tan
ce
s
f
o
r
v
ar
ied
asp
ec
ts
an
d
s
ca
les.
T
h
e
co
n
v
o
l
u
tio
n
s
ar
e
ap
p
lied
in
al
l
th
e
th
r
ee
s
tag
es.
T
h
e
en
h
an
c
ed
f
ea
tu
r
es
ar
e
th
e
n
f
u
r
th
er
f
u
s
ed
b
y
u
p
-
s
am
p
lin
g
th
e
s
u
m
elem
en
t
-
wis
e.
Fu
r
th
e
r
,
th
e
f
u
s
ed
e
n
h
an
ce
d
f
ea
t
u
r
e
s
o
f
1
/4
,
1
/
8
,
1
/
1
6
,
an
d
1
/3
2
r
eso
lu
tio
n
ar
e
g
e
n
er
a
ted
with
1
2
8
ch
an
n
els.
3
.
2
.
Dee
p f
e
a
t
ure
a
ug
m
ent
a
t
io
n m
o
du
le
R
o
b
u
s
t
C
NN
s
ar
e
u
s
ed
in
s
ce
n
e
tex
t
id
en
tific
atio
n
alg
o
r
ith
m
s
to
ex
tr
ac
t
im
p
r
o
v
ed
f
ea
tu
r
es a
n
d
b
o
o
s
t
o
v
er
all
p
er
f
o
r
m
an
ce
.
Ho
wev
er
,
wh
e
n
cr
ea
tin
g
f
ea
tu
r
e
m
ap
s
o
f
v
ar
i
o
u
s
s
izes
u
s
in
g
b
asic
s
am
p
lin
g
o
r
co
n
v
o
l
u
tio
n
al
a
p
p
r
o
ac
h
es,
th
e
tex
tu
r
es
a
n
d
b
o
r
d
er
s
o
f
tex
t
in
s
tan
ce
s
ar
e
co
m
p
r
o
m
is
ed
.
T
h
is
in
s
ig
h
t
lead
s
t
o
th
e
d
ev
elo
p
m
e
n
t
o
f
a
lig
h
t
weig
h
t,
p
lu
g
g
a
b
le
m
o
d
u
le
f
o
r
DFAM
en
h
an
ce
m
en
t.
e
n
h
an
ce
s
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
Ho
wev
e
r
,
g
iv
e
n
with
=
{
1
,
2
,
3
,
4
,
5
]
,
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
cu
s
es
o
n
th
e
p
r
ed
ictio
n
o
f
th
e
r
ec
o
n
s
tr
u
cted
im
a
g
e
f
o
r
th
e
r
e
co
n
s
tr
u
ctio
n
o
f
th
e
in
p
u
t im
ag
e
is
g
iv
en
b
y
(
1
)
.
=
(
)
=
1
,
2
,
3
,
4
,
5
(
1
)
T
o
f
ee
d
th
e
in
p
u
t
t
o
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
c
o
n
d
u
ctin
g
th
e
ab
latio
n
ex
p
er
im
en
ts
th
e
in
p
u
ts
f
ed
ar
e
1
,
2
,
an
d
3
.
T
h
e
f
ea
tu
r
es
d
er
iv
e
d
f
r
o
m
th
e
i
n
p
u
t
2
an
d
3
ar
e
u
p
s
am
p
le
d
to
1
th
r
o
u
g
h
li
n
ea
r
in
ter
p
o
latio
n
.
T
h
ese
ar
e
co
n
c
aten
ated
an
d
p
r
o
ce
s
s
ed
th
r
o
u
g
h
a
co
n
v
o
lu
tio
n
b
lo
ck
f
o
llo
wed
b
y
an
elem
e
n
t
s
u
m
o
f
1
,
to
s
am
p
le
f
ea
tu
r
e
m
ap
s
th
at
ar
e
f
ix
ed
to
th
e
o
r
ig
in
a
l size
w
ith
in
th
e
in
p
u
t im
ag
e,
a
d
ec
o
n
v
o
l
u
tio
n
al
lay
er
is
m
o
d
if
ied
.
T
h
e
ex
p
ec
ted
o
u
tco
m
es
ar
e
p
r
o
d
u
ce
d
u
s
in
g
d
if
f
er
en
t
3
×
3
co
n
v
o
lu
tio
n
al
lay
er
s
.
I
n
o
r
d
er
to
en
a
b
le
tex
t
d
etec
tio
n
ac
r
o
s
s
s
ce
n
er
ies
an
d
tr
af
f
ic
p
an
els,
th
e
n
etwo
r
k
le
ar
n
s
an
d
ac
q
u
ir
es
co
m
p
r
eh
e
n
s
iv
e
in
f
o
r
m
atio
n
o
n
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
o
f
tex
ts
.
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
A
d
a
p
tive
d
efo
r
ma
b
le
fea
tu
r
e
a
u
g
men
ta
tio
n
a
n
d
r
efin
eme
n
t n
etw
o
r
k
fo
r
s
ce
n
e
text
… (
R
a
tn
a
ma
la
S
.
P
a
til
)
835
3
.
3
.
Dee
p f
e
a
t
ure
re
f
inem
en
t
m
o
du
le
I
n
g
en
er
al,
it
is
a
co
m
p
lex
task
to
class
if
y
p
ix
els
in
an
n
o
tatio
n
s
th
at
a
r
e
f
ar
f
r
o
m
th
e
b
o
u
n
d
a
r
y
ac
cu
r
ately
.
Ou
r
s
tu
d
y
an
d
te
s
tin
g
s
u
g
g
est
th
at
en
h
an
ce
d
f
ea
tu
r
e
f
u
s
io
n
m
i
g
h
t
lead
to
co
n
f
u
s
io
n
b
etwe
en
b
ac
k
g
r
o
u
n
d
an
d
b
o
r
d
e
r
d
ata.
C
u
r
r
en
t
s
ce
n
e
tex
t
d
etec
to
r
s
d
is
r
eg
ar
d
th
e
s
ig
n
if
ica
n
ce
o
f
tex
t
b
o
r
d
e
r
s
an
d
alwa
y
s
co
n
s
id
er
ev
er
y
p
ix
el
in
a
p
r
o
p
o
s
al
id
en
tically
.
Desp
ite
m
ak
in
g
u
p
a
r
elativ
ely
m
in
o
r
p
o
r
tio
n
o
f
th
e
im
ag
e,
th
e
p
ix
els
in
tex
t b
o
r
d
e
r
s
ar
e
cr
u
cial
f
o
r
te
x
t
lo
ca
lizatio
n
.
Fo
r
a
cc
u
r
ate
lo
ca
lizatio
n
,
we
th
u
s
s
u
g
g
est
an
FR
E
th
at
s
p
ec
if
ically
u
s
es
tex
t
r
ef
in
em
e
n
t.
T
h
e
co
n
v
o
lu
ti
o
n
al
n
etwo
r
k
s
in
o
r
th
o
g
o
n
al
d
ir
ec
tio
n
s
1
×
3
3
×
1
in
th
r
ee
d
ilatio
n
s
wh
ich
ca
p
tu
r
e
tex
t r
ef
in
em
e
n
t.
T
h
r
o
u
g
h
el
em
en
t
-
wis
e
s
u
m
,
co
n
cr
etely
,
a
n
elem
en
t
-
wis
e
lis
t
i
s
g
en
er
ated
as
a
f
ea
tu
r
e
,
th
e
co
n
tain
s
in
f
o
r
m
atio
n
th
at
co
m
b
in
es
an
d
,
in
th
is
o
n
e
b
r
an
ch
u
s
es
a
b
o
u
n
d
ar
y
m
ap
p
r
ed
ictio
n
h
ea
d
.
T
o
ac
q
u
ir
e
a
f
ea
tu
r
e
f
o
r
tex
t
-
b
ased
f
ea
t
u
r
e
im
p
r
o
v
em
en
t,
th
e
o
th
er
b
r
an
c
h
g
o
es
th
r
o
u
g
h
a
3
×
3
an
d
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
;
th
e
b
o
u
n
d
ar
y
m
ap
i
n
d
icate
s
th
e
b
o
u
n
d
ar
y
/n
o
n
-
b
o
u
n
d
ar
y
class
if
icatio
n
p
r
o
ce
s
s
.
3
.
4
.
O
pti
m
ized
la
bel g
ener
a
t
io
n a
nd
lo
s
s
f
un
ct
io
n
E
v
er
y
te
x
t
o
cc
u
r
r
en
ce
is
d
esig
n
ated
as
a
p
o
ly
g
o
n
in
th
e
s
co
r
e
m
ap
,
th
r
esh
o
ld
m
a
p
,
an
d
esti
m
ated
b
in
ar
y
m
ap
,
wh
ich
ar
e
all
th
e
s
am
e.
Dif
f
er
en
t
d
atasets
ar
e
u
s
ed
to
d
if
f
er
en
tiate
th
e
v
er
tex
es.
E
ac
h
p
ix
el
(
x
,
y
)
in
th
e
b
in
ar
y
m
ap
is
d
o
wn
s
i
ze
d
to
a
p
ix
el
wh
o
s
e
v
al
u
e
is
s
u
m
m
ar
ized
to
0
,
an
d
th
e
s
h
o
r
test
d
is
tan
ce
is
ca
lcu
lated
,
.
T
h
e
m
ap
p
in
g
d
is
tan
ce
f
o
r
ea
ch
tex
t
is
f
o
r
m
u
lat
ed
as
s
h
o
wn
in
(
2
)
.
T
h
e
d
is
tan
ce
is
m
ap
p
ed
f
r
o
m
ea
c
h
tex
t d
is
tan
ce
wh
ich
is
ev
alu
ated
as g
iv
en
in
(
3
)
.
T
h
is
is
ev
alu
ated
as g
iv
en
in
(
4
)
.
=
{
,
}
;
,
∈
(
2
)
=
{
1
,
<
2
0
(
3
)
=
+
1
+
2
(
+
)
+
3
(
4
)
Her
e
,
,
,
an
d
d
ep
icts
b
o
r
d
er
m
ap
s
,
b
in
ar
y
m
ap
s
,
s
co
r
e
m
ap
s
,
th
r
esh
o
ld
m
a
p
s
,
an
d
r
ec
o
n
s
tr
u
cted
im
ag
es.
T
h
e
p
ar
am
eter
s
ar
e
s
et
to
2
,
0
.
2
,
a
n
d
0
.
0
2
.
T
h
e
b
in
a
r
y
c
r
o
s
s
en
tr
o
p
y
lo
s
s
v
alu
e
is
u
s
ed
to
r
ep
r
esen
t
th
e
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
.
f
o
r
,
1
lo
s
s
f
o
r
an
d
d
i
ce
lo
s
s
as
.
Alg
o
r
ith
m
1
s
h
o
ws
th
e
en
h
an
ce
d
b
o
u
n
d
ar
y
-
en
h
an
ce
d
STR (
ADF
R
N)
alg
o
r
ith
m
.
Alg
o
r
ith
m
1.
E
n
h
a
n
ce
d
b
o
u
n
d
ar
y
-
en
h
an
ce
d
STR
(
ADF
A
R
N
)
I
n
p
u
t:
An
in
p
u
t im
a
g
e
Step
1
:
DFEN
:
i)
Feed
th
e
in
p
u
t im
a
g
e
in
to
t
h
e
DFEN
m
o
d
u
le.
ii)
Ap
p
ly
d
e
f
o
r
m
a
b
le
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
s
to
e
x
tr
ac
t
en
h
a
n
ce
d
f
ea
t
u
r
es
C
1
,
C
2
,
C
3
,
C
4
,
an
d
C
5
wh
er
e:
–
C
1
h
as r
eso
lu
tio
n
1
/
2
–
C
2
h
as r
eso
lu
tio
n
1
/
4
–
C
3
h
as r
eso
lu
tio
n
1
/
8
–
C
4
h
as r
eso
lu
tio
n
1
/1
6
–
C
5
h
as r
eso
lu
tio
n
1
/
3
2
E
ac
h
f
ea
tu
r
e
m
ap
h
as c
h
an
n
el
s
6
4
,
2
5
6
,
5
1
2
,
1
0
2
4
,
an
d
2
0
4
8
r
esp
ec
tiv
ely
.
iii)
Fu
s
e
th
e
en
h
an
ce
d
f
ea
tu
r
es
b
y
u
p
s
am
p
lin
g
an
d
s
u
m
m
in
g
elem
en
t
-
wis
e,
r
esu
ltin
g
in
a
f
u
s
ed
d
ef
o
r
m
a
b
le
f
ea
tu
r
e
fe
with
1
2
8
ch
an
n
els.
Step
2
:
FR
E
:
i)
I
m
p
lem
en
t th
e
FR
E
:
–
E
x
tr
ac
t
r
ef
in
em
e
n
t
to
g
et
fe
x
f
ea
tu
r
e
u
s
in
g
c
o
n
v
o
lu
tio
n
al
n
etwo
r
k
s
in
o
r
th
o
g
o
n
al
d
ir
ec
tio
n
s
c
on
v
1
×
3
c
on
v
3
×
1
in
th
r
ee
d
ilatio
n
s
.
–
C
o
m
b
in
e
fe
an
d
fe
x
elem
en
t
-
wis
e
t
o
g
et
fe
us
ag
e
ii)
Use a
p
r
ed
ictio
n
h
ea
d
-
o
n
fe
us
a
g
e
to
f
o
r
ec
ast tex
t a
n
d
n
o
n
-
te
x
t sco
r
e
m
ap
s
.
iii)
Use a
s
ec
o
n
d
p
r
ed
ictio
n
h
ea
d
-
on
fe
to
cr
ea
te
th
e
th
r
esh
o
ld
m
a
p
.
Step
3
:
Dif
f
er
en
tiab
le
b
in
a
r
izatio
n
:
u
s
e
d
if
f
er
en
tiab
le
b
in
ar
izatio
n
to
ca
lcu
late
th
e
esti
m
ated
b
in
ar
y
m
a
p
b
ased
o
n
th
e
s
co
r
e
m
ap
an
d
t
h
r
esh
o
ld
m
ap
.
Step
4
:
DFA:
i)
Fo
r
ea
ch
f
ea
tu
r
e
m
ap
E
k
with
k
=
{
1
,
2
,
3
,
4
,
5
}
:
p
r
ed
ict
r
ec
o
n
s
tr
u
cte
d
im
ag
e
=
(
)
.
ii)
Up
s
am
p
le
f
ea
tu
r
es d
er
iv
e
d
f
r
o
m
E
2
an
d
E
3
ar
e
u
p
s
am
p
led
to
E
1
th
r
o
u
g
h
lin
ea
r
in
ter
p
o
latio
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
831
-
8
4
0
836
iii)
C
o
n
ca
ten
ate
an
d
p
r
o
ce
s
s
th
ese
f
ea
tu
r
es
th
r
o
u
g
h
a
co
n
v
o
lu
t
io
n
b
lo
ck
,
f
o
llo
wed
b
y
a
n
elem
en
t
s
u
m
o
f
E
1
.
iv
)
Sam
p
le
f
ea
tu
r
e
m
ap
s
th
at
ar
e
f
ix
ed
to
th
e
o
r
ig
in
al
in
p
u
t
i
m
ag
e
s
ize
s
h
o
u
ld
b
e
s
u
b
jecte
d
to
a
d
ec
o
n
v
o
lu
tio
n
al
lay
e
r
.
v)
Gen
er
ate
p
r
ed
icted
r
esu
lts
th
r
o
u
g
h
v
ar
io
u
s
c
on
v
3
×
3
co
n
v
o
lu
tio
n
al
la
y
er
s
.
Step
5
:
Op
tim
ized
lab
el
g
en
e
r
atio
n
:
i)
L
ab
el
ea
ch
tex
t
o
cc
u
r
r
en
ce
a
s
a
p
o
l
y
g
o
n
f
o
r
th
e
s
co
r
e
m
ap
,
th
r
esh
o
ld
m
ap
,
an
d
esti
m
ated
b
in
ar
y
m
a
p
.
ii)
C
o
m
p
u
te
th
e
s
h
o
r
test
d
is
tan
ce
F
x
,
y
f
r
o
m
ea
ch
p
ix
el
x
,
y
an
d
s
h
r
in
k
it
in
th
e
b
in
ar
y
m
ap
to
a
p
ix
el
v
alu
e
o
f
0
i
f
F
x
,
y
<2
,
o
th
er
wi
s
e
to
1
.
Step
6
:
Op
tim
ized
lo
s
s
f
u
n
ctio
n
: c
alcu
late
th
e
lo
s
s
as:
N
=
N
U
+
α
1
N
V
+
α
2
(
N
D
+
N
PS
)
+
α
3
N
l
o
s
s
i)
W
h
er
e
N
U
r
ep
r
esen
ts
th
e
s
co
r
e
m
ap
lo
s
s
,
N
V
r
ep
r
esen
ts
th
e
th
r
esh
o
ld
m
ap
lo
s
s
,
N
D
r
ep
r
esen
ts
th
e
b
in
ar
y
m
ap
lo
s
s
,
N
PS
r
ep
r
es
en
ts
th
e
r
ec
o
n
s
tr
u
cte
d
im
ag
e
lo
s
s
,
an
d
N
l
o
s
s
r
ep
r
esen
ts
th
e
b
o
u
n
d
ar
y
m
ap
lo
s
s
.
ii)
Set p
ar
am
eter
s
to
α1
=2
,
α2
=0
.
2
,
a
n
d
α3
=0
.
0
2
.
iii)
Use c
r
o
s
s
-
en
tr
o
p
y
lo
s
s
f
o
r
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
v
alu
e
N
U
,
lo
s
s
N
1
f
o
r
N
V
,
an
d
d
ice
l
o
s
s
f
o
r
N
D
.
Step
7
:
T
r
ain
in
g
:
c
o
n
n
ec
t
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
t
h
e
n
etwo
r
k
a
n
d
tr
ain
ev
e
r
y
n
etwo
r
k
n
o
d
e
f
r
o
m
s
tar
t
to
f
in
is
h
to
im
p
r
o
v
e
f
ea
t
u
r
e
r
e
p
r
e
s
en
tatio
n
s
.
Ou
tp
u
t:
ap
p
r
o
x
im
ate
b
in
ar
y
m
ap
in
d
ic
atin
g
d
etec
ted
an
d
r
ec
o
g
n
ized
tex
t.
4.
P
E
RF
O
RM
A
NCE
E
VA
L
U
AT
I
O
N
T
h
e
ass
ess
m
en
t
m
etr
ics
u
tili
ze
d
f
o
r
tex
t
d
etec
tio
n
e
n
co
m
p
ass
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
.
T
h
e
r
atio
o
f
r
ec
o
g
n
ized
tex
t
r
eg
io
n
s
to
all
tex
t
r
eg
io
n
s
is
m
ea
s
u
r
ed
b
y
th
e
r
ec
all
m
etr
ic.
T
h
e
F1
-
s
co
r
e
,
s
o
m
etim
es
r
ef
er
r
ed
to
as
F1
-
s
co
r
e
,
is
a
s
tatis
tic
th
at
u
s
es
h
ar
m
o
n
ic
av
er
ag
e
to
co
m
b
in
e
r
ec
all
an
d
ac
cu
r
ac
y
.
I
t
is
f
r
eq
u
e
n
tly
u
s
ed
to
ass
es
s
h
o
w
well
d
etec
tio
n
alg
o
r
ith
m
s
wo
r
k
.
On
e
cr
u
cial
cr
iter
io
n
f
o
r
ass
es
s
in
g
a
m
o
d
el's
p
er
f
o
r
m
an
ce
is
its
co
m
p
u
tatio
n
al
co
m
p
lex
ity
.
I
t
tak
es
in
to
ac
co
u
n
t
elem
en
ts
lik
e
in
f
er
en
ce
tim
e,
co
m
p
u
tatio
n
al
c
o
m
p
lex
ity
,
a
n
d
p
ar
am
eter
co
u
n
t.
T
h
e
r
o
b
u
s
tn
ess
m
etr
ic,
wh
ich
is
s
ee
n
t
o
b
e
o
f
th
e
u
tm
o
s
t
r
elev
an
ce
,
is
f
r
eq
u
en
tly
u
s
ed
t
o
ass
ess
a
m
o
d
el'
s
p
er
f
o
r
m
an
c
e.
T
h
e
ca
p
ac
ity
o
f
m
o
d
el
to
p
er
f
o
r
m
co
n
s
is
ten
tly
ac
r
o
s
s
m
an
y
d
atasets
an
d
co
n
t
ex
ts
is
r
ef
er
r
ed
to
as m
o
d
el
s
tab
ilit
y
th
e
f
o
r
m
at
o
f
ta
b
les an
d
g
r
ap
h
s
.
4
.
1
.
Da
t
a
s
et
det
a
ils
A
lar
g
e
-
s
ca
le
d
ataset
ca
lled
C
OC
O
-
T
ex
t
was
cr
ea
ted
to
im
p
r
o
v
e
tex
t
id
en
tific
atio
n
a
n
d
d
e
tectio
n
in
n
atu
r
al
p
h
o
to
s
.
I
t
a
d
d
s
m
o
r
e
t
h
an
6
3
,
6
8
6
p
h
o
to
s
with
m
o
r
e
th
an
1
7
3
,
5
8
9
tex
t
in
s
tan
ce
s
to
th
e
C
OC
O
d
ataset.
B
o
u
n
d
in
g
b
o
x
es,
tr
an
s
cr
ip
tio
n
s
,
an
d
ch
ar
ac
ter
is
tics
lik
e
l
an
g
u
ag
e
an
d
r
ea
d
ab
ilit
y
ar
e
ad
d
ed
t
o
ea
ch
tex
t
in
s
tan
ce
.
T
h
e
d
ataset
is
p
e
r
f
ec
t
f
o
r
cr
ea
tin
g
a
n
d
ev
alu
atin
g
r
eliab
le
tex
t
d
etec
tio
n
an
d
id
en
tific
atio
n
alg
o
r
ith
m
s
b
ec
au
s
e
o
f
th
e
v
ar
iety
o
f
tex
t
ap
p
ea
r
an
ce
s
,
i
n
tr
icate
b
ac
k
g
r
o
u
n
d
s
,
an
d
m
u
ltil
in
g
u
al
co
n
ten
t.
W
id
ely
u
s
ed
f
o
r
b
en
ch
m
a
r
k
in
g
,
C
OC
O
-
T
ex
t
h
elp
s
p
u
s
h
th
e
b
o
u
n
d
ar
ies
o
f
s
ce
n
e
u
n
d
er
s
tan
d
i
n
g
b
y
in
co
r
p
o
r
atin
g
tex
t
u
al
in
f
o
r
m
at
io
n
,
o
f
f
er
in
g
a
co
m
p
r
eh
e
n
s
iv
e
r
eso
u
r
ce
f
o
r
r
esear
ch
er
s
an
d
p
r
ac
titi
o
n
er
s
aim
in
g
to
en
h
an
ce
tex
t a
n
aly
s
is
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
.
4
.
2
.
Resul
t
s
A
co
m
p
ar
is
o
n
o
f
d
if
f
e
r
en
t
ap
p
r
o
ac
h
es
b
ased
o
n
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
is
s
h
o
wn
i
n
T
ab
le
1
.
T
h
e
p
r
o
p
o
s
ed
s
eg
m
en
tatio
n
(
PS
)
m
eth
o
d
o
u
tp
e
r
f
o
r
m
s
all
o
th
er
m
eth
o
d
s
with
th
e
h
ig
h
est p
r
ec
is
io
n
o
f
9
6
.
8
9
%,
r
ec
all
o
f
9
6
.
7
6
%,
an
d
an
F
1
-
s
co
r
e
o
f
9
6
.
5
%.
No
tab
ly
,
th
e
en
s
em
b
le
s
eg
m
en
tatio
n
(
ES
)
m
eth
o
d
also
d
em
o
n
s
tr
ates
s
tr
o
n
g
p
er
f
o
r
m
a
n
ce
,
ac
h
iev
in
g
a
p
r
ec
is
io
n
o
f
9
4
.
2
8
%,
r
ec
all
o
f
9
3
.
8
4
%,
a
n
d
an
F1
-
s
co
r
e
o
f
9
4
.
0
5
%,
i
n
d
icatin
g
its
ef
f
ec
tiv
en
ess
in
th
e
g
iv
en
c
o
n
tex
t.
T
ab
le
1
.
R
esu
lts
M
e
t
h
o
d
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
M
a
n
j
a
r
i
e
t
a
l
.
[
1
9
]
8
1
.
3
6
7
9
.
8
4
8
0
.
5
9
P
r
a
b
u
a
n
d
S
u
n
d
a
r
[
2
0
]
8
2
.
5
7
8
0
.
6
5
8
1
.
5
9
La
r
b
i
[
2
1
]
7
6
.
8
9
7
7
.
9
8
7
7
.
4
3
Ta
r
r
i
d
e
e
t
a
l
.
[
2
2
]
7
5
.
4
1
7
6
.
3
2
7
5
.
8
6
B
h
a
t
t
e
t
a
l
.
[
2
3
]
8
9
.
6
3
8
7
.
8
1
8
8
.
7
1
V
i
sh
w
a
k
a
r
m
a
e
t
a
l
.
[
2
4
]
9
1
.
3
7
8
6
.
2
9
8
8
.
7
5
ES
[
2
5
]
9
4
.
2
8
9
3
.
8
4
9
4
.
0
5
PS
9
6
.
8
9
9
6
.
7
6
9
6
.
5
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
A
d
a
p
tive
d
efo
r
ma
b
le
fea
tu
r
e
a
u
g
men
ta
tio
n
a
n
d
r
efin
eme
n
t n
etw
o
r
k
fo
r
s
ce
n
e
text
… (
R
a
tn
a
ma
la
S
.
P
a
til
)
837
Fig
u
r
e
2
in
d
icate
s
th
at
th
e
PS
m
eth
o
d
o
l
o
g
ies
ac
h
iev
e
th
e
h
i
g
h
est
p
r
ec
is
io
n
,
ar
o
u
n
d
9
4
%,
s
u
g
g
esti
n
g
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
in
co
r
r
e
ctly
id
en
tify
in
g
r
elev
a
n
t
in
s
tan
ce
s
co
m
p
ar
ed
to
o
th
e
r
m
eth
o
d
s
.
T
h
e
r
esear
ch
in
[
2
3
]
,
[
2
4
]
also
d
em
o
n
s
tr
ate
h
ig
h
p
r
ec
is
io
n
,
ar
o
u
n
d
9
0
%
an
d
91%
r
esp
ec
tiv
ely
,
s
h
o
win
g
s
tr
o
n
g
p
e
r
f
o
r
m
an
ce
b
u
t
s
lig
h
tly
lo
wer
th
an
E
S
an
d
PS
.
Pra
b
u
an
d
Su
n
d
ar
[
2
0
]
ac
h
iev
ed
a
p
r
ec
is
io
n
o
f
a
p
p
r
o
x
i
m
ately
8
3
%,
w
h
ich
is
m
o
d
er
ate
b
u
t
s
till
s
ig
n
if
ic
an
tly
h
ig
h
er
th
a
n
th
e
r
em
ain
in
g
m
eth
o
d
s
.
T
h
e
r
esear
c
h
in
[
1
9
]
,
[
2
1
]
s
h
o
w
p
r
ec
is
io
n
v
alu
es
o
f
ap
p
r
o
x
im
a
tely
8
1
%
an
d
7
7
%
r
esp
ec
tiv
ely
,
in
d
icatin
g
r
ea
s
o
n
ab
le
b
u
t
lo
wer
p
er
f
o
r
m
an
ce
.
T
ar
r
id
e
et
a
l
.
[
2
2
]
p
r
esen
t
t
h
e
lo
west
p
r
ec
is
io
n
v
alu
e
at
ar
o
u
n
d
7
5
%,
s
u
g
g
esti
n
g
r
o
o
m
f
o
r
im
p
r
o
v
em
e
n
t.
T
h
is
an
aly
s
is
h
ig
h
lig
h
ts
t
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
PS
an
d
E
S
m
et
h
o
d
s
in
ac
h
ie
v
in
g
h
ig
h
p
r
ec
is
io
n
in
tex
t
d
etec
tio
n
task
s
o
n
th
e
C
OC
O
-
T
ex
t d
ataset.
Fig
u
r
e
2
.
Pre
cisi
o
n
m
ea
s
u
r
e
Fig
u
r
e
3
d
ep
icts
th
e
r
ec
all
(
%
)
v
alu
es
f
o
r
v
ar
i
o
u
s
m
eth
o
d
o
l
o
g
ies
ap
p
lied
to
th
e
C
OC
O
-
T
ex
t
d
ataset.
T
h
e
m
eth
o
d
s
co
m
p
ar
e
d
ar
e
f
r
o
m
s
tu
d
ies
in
[
1
9
]
–
[
2
5
]
,
r
esp
ec
tiv
ely
.
T
h
e
PS
an
d
E
S
m
et
h
o
d
o
lo
g
ies
ac
h
iev
e
th
e
h
ig
h
est
r
ec
all
r
ates,
b
o
th
a
r
o
u
n
d
9
4
%,
in
d
icatin
g
th
eir
s
u
p
er
io
r
ab
ilit
y
t
o
id
e
n
tify
r
elev
an
t
in
s
tan
ce
s
.
T
h
e
r
esear
ch
in
[
2
3
]
,
[
2
4
]
also
d
em
o
n
s
tr
ate
s
tr
o
n
g
r
ec
all
v
a
lu
es
at
ap
p
r
o
x
im
ately
8
8
%,
s
h
o
win
g
ef
f
ec
tiv
e
p
er
f
o
r
m
an
ce
.
T
h
e
r
esear
ch
in
[
1
9
]
,
[
2
0
]
ac
h
iev
e
m
o
d
er
ate
r
ec
all
r
ates
o
f
ar
o
u
n
d
8
1
%,
w
h
ile
th
e
r
esear
ch
i
n
[
2
1
]
,
[
2
2
]
s
h
o
w
lo
wer
r
ec
all
r
ates
at
ap
p
r
o
x
im
ately
7
7
%
an
d
7
6
%
r
esp
ec
tiv
ely
.
T
h
is
an
al
y
s
is
h
ig
h
lig
h
ts
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
PS
an
d
E
S
m
eth
o
d
s
in
ac
h
iev
in
g
h
i
g
h
r
ec
all
in
te
x
t
d
etec
tio
n
task
s
o
n
th
e
C
OC
O
-
T
ex
t
d
ataset,
o
u
tp
er
f
o
r
m
in
g
o
th
er
m
eth
o
d
o
l
o
g
ies in
ter
m
s
o
f
id
e
n
tify
in
g
m
a
x
im
u
m
n
u
m
b
er
o
f
r
elev
an
t in
s
tan
ce
s
.
Fig
u
r
e
3
.
R
ec
all
m
ea
s
u
r
e
Fig
u
r
e
4
p
r
esen
ts
a
co
m
p
ar
ativ
e
an
aly
s
is
o
f
v
ar
io
u
s
m
eth
o
d
s
b
ased
o
n
th
ei
r
F1
-
s
co
r
es.
T
h
e
m
eth
o
d
s
ev
alu
ated
in
cl
u
d
e
th
o
s
e
p
r
o
p
o
s
ed
in
[
1
9
]
–
[
2
5
]
,
r
esp
ec
tiv
e
ly
.
T
h
e
h
ig
h
est
F1
-
s
co
r
e
is
a
ch
iev
ed
b
y
th
e
PS
m
eth
o
d
with
9
6
.
5
%,
f
o
llo
wed
clo
s
ely
b
y
th
e
E
S
m
et
h
o
d
wi
th
9
4
.
0
5
%.
T
h
e
r
esear
ch
in
[
2
3
]
,
[
2
4
]
also
s
h
o
w
s
tr
o
n
g
p
er
f
o
r
m
an
ce
s
with
F1
-
s
co
r
es
o
f
8
8
.
7
5
%
an
d
8
8
.
7
1
%,
r
esp
ec
tiv
ely
.
T
h
e
m
eth
o
d
s
in
[
1
9
]
,
[
2
0
]
y
iel
d
8
1
,
3
6
8
2
,
5
7
7
6
,
8
9
7
5
,
4
1
8
9
,
6
3
9
1
,
3
7
9
4
,
2
8
9
6
,
8
9
0
20
40
60
80
1
0
0
1
2
0
M
an
j
ar
i
et
al
.
[
1
9
]
Pr
a
b
u
an
d
S
u
n
d
a
r
[2
0
]
L
ar
b
i
[2
1
]
T
ar
r
i
d
e et
al
.
[
2
2
]
B
h
at
t
et
al
.
[
2
3
]
Vi
s
h
w
ak
ar
m
a
et
al
.
[2
4
]
E
S
[
2
5
]
PS
v
al
ue
M
e
t
ho
do
l
o
g
y
P
r
e
c
i
si
o
n
(
%)
7
9
,
8
4
8
0
,
6
5
7
7
,
9
8
7
6
,
3
2
8
7
,
8
1
8
6
,
2
9
9
3
,
8
4
9
6
,
7
6
0
20
40
60
80
1
0
0
1
2
0
M
a
n
j
a
r
i
et
al
.
[
1
9
]
P
r
a
b
u
a
n
d
S
u
n
d
a
r
[2
0
]
L
a
r
b
i
[2
1
]
T
a
r
r
i
d
e et
al
.
[
2
2
]
B
h
a
t
t
et
al
.
[
2
3
]
Vi
s
h
w
a
k
a
r
m
a
et
al
.
[2
4
]
E
S
[
2
5
]
PS
v
al
ue
M
e
t
ho
do
l
o
g
y
R
e
c
al
l
(
%)
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
831
-
8
4
0
838
m
o
d
er
ate
p
er
f
o
r
m
an
ce
s
,
with
F1
-
s
co
r
es
o
f
8
1
.
5
9
%
a
n
d
8
0
.
5
9
%.
I
n
co
n
tr
ast,
th
e
r
esear
ch
in
[
2
1
]
,
[
2
2
]
h
a
v
e
th
e
lo
west
F1
-
s
co
r
e
s
at
7
7
.
4
3
%
an
d
7
5
.
8
6
%,
r
esp
ec
tiv
ely
.
T
h
is
a
n
aly
s
is
h
ig
h
lig
h
ts
t
h
at
th
e
PS
an
d
E
S
m
eth
o
d
s
s
ig
n
if
ican
tly
o
u
t
p
er
f
o
r
m
th
e
o
th
er
s
,
in
d
icatin
g
th
e
ir
s
u
p
er
io
r
ef
f
icac
y
in
t
h
e
ev
a
lu
ated
co
n
tex
t.
T
h
e
r
esu
lts
s
u
g
g
est
th
at
wh
ile
s
e
v
er
al
m
eth
o
d
s
s
h
o
w
co
m
p
eti
tiv
e
p
er
f
o
r
m
an
ce
,
th
e
r
e
is
a
clea
r
d
is
tin
ctio
n
in
ef
f
ec
tiv
en
ess
am
o
n
g
th
e
to
p
-
p
er
f
o
r
m
in
g
an
d
lo
wer
-
p
er
f
o
r
m
i
n
g
m
eth
o
d
s
.
Fig
u
r
e
4
.
F1
-
s
co
r
e
m
ea
s
u
r
e
c
o
m
p
ar
is
o
n
o
f
ex
is
tin
g
s
tate
-
of
-
a
r
t te
ch
n
iq
u
es with
PS
4
.
3
.
Co
m
pa
riso
n a
na
ly
s
is
T
h
e
ADF
A
R
N
ap
p
r
o
ac
h
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
s
t
h
e
cu
r
r
e
n
t
s
tate
-
of
-
th
e
-
ar
t
tec
h
n
iq
u
e,
E
S,
ac
co
r
d
in
g
to
p
er
f
o
r
m
a
n
ce
m
e
asu
r
es.
I
n
p
ar
ticu
lar
,
a
g
r
ea
te
r
r
ate
o
f
c
o
r
r
ec
tly
id
en
tifie
d
tex
t
in
s
tan
ce
s
was
s
h
o
wn
b
y
a
2
.
7
7
%
im
p
r
o
v
em
en
t
in
th
e
p
r
ec
is
io
n
s
tatis
tic.
T
h
e
m
o
d
el'
s
im
p
r
o
v
e
d
ca
p
ac
i
ty
to
r
ec
o
g
n
ize
all
p
er
tin
en
t
tex
t
in
s
tan
ce
s
was
r
ef
lecte
d
in
th
e
r
ec
all
m
etr
ic,
w
h
ich
ex
p
er
ie
n
ce
d
an
e
v
en
b
i
g
g
er
im
p
r
o
v
em
en
t
o
f
3
.
1
1
%.
Ad
d
itio
n
ally
,
th
e
F1
-
s
co
r
e
—
wh
ich
s
tr
ik
es
a
co
m
p
r
o
m
is
e
b
etwe
en
r
ec
all
an
d
p
r
ec
is
io
n
—
r
o
s
e
b
y
2
.
6
0
%,
in
d
icatin
g
a
co
m
p
r
eh
e
n
s
iv
e
im
p
r
o
v
em
en
t
in
th
e
tex
t
r
ec
o
g
n
itio
n
s
y
s
tem
'
s
o
v
er
all
p
er
f
o
r
m
an
ce
.
T
h
ese
en
h
an
ce
m
e
n
ts
h
ig
h
lig
h
t
h
o
w
well
th
e
s
u
g
g
ested
ADF
A
R
N
ap
p
r
o
ac
h
h
an
d
les
th
e
ch
al
len
g
es
o
f
STR
an
d
p
r
o
d
u
ce
s
b
etter
o
u
tc
o
m
es th
an
o
th
er
ap
p
r
o
ac
h
es.
T
h
e
c
o
m
p
a
r
is
o
n
an
aly
s
is
is
d
is
p
lay
ed
in
T
ab
le
2
.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
an
aly
s
is
M
e
t
r
i
c
ES
PS
I
mp
r
o
v
i
z
a
t
i
o
n
i
n
%
P
r
e
c
i
s
i
o
n
(
%)
9
4
.
2
8
9
6
.
8
9
2
.
7
6
8
3
4
9
5
9
7
R
e
c
a
l
l
(
%)
9
3
.
8
4
9
6
.
7
6
3
.
1
1
1
6
7
9
4
5
4
F1
-
sc
o
r
e
(
%)
9
4
.
0
5
9
6
.
5
2
.
6
0
4
9
9
7
3
4
2
5.
CO
NCLU
SI
O
N
T
h
e
ADF
A
R
N
m
eth
o
d
o
lo
g
y
p
r
esen
ts
a
s
ig
n
if
ican
t
ad
v
an
ce
m
en
t
in
th
e
f
ield
o
f
STR
.
B
y
in
teg
r
atin
g
d
ef
o
r
m
a
b
le
co
n
v
o
lu
tio
n
al
n
et
wo
r
k
s
f
o
r
d
ef
o
r
m
ab
le
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
a
n
o
v
el
FR
E
,
ADF
A
R
N
ef
f
ec
tiv
ely
ad
d
r
ess
es th
e
ch
allen
g
es p
o
s
ed
b
y
v
ar
iatio
n
s
in
tex
t a
p
p
ea
r
a
n
ce
,
o
r
ien
tatio
n
,
an
d
b
ac
k
g
r
o
u
n
d
co
m
p
lex
ity
.
T
h
e
co
m
p
r
eh
e
n
s
iv
e
p
r
o
ce
s
s
o
f
D
FEN,
DFAM,
DFR
M,
an
d
th
e
u
s
e
o
f
d
if
f
e
r
en
tiab
le
b
i
n
ar
i
za
tio
n
en
h
a
n
ce
s
th
e
p
r
ec
is
io
n
an
d
ac
c
u
r
ac
y
o
f
tex
t
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
in
n
atu
r
al
s
ce
n
es.
T
h
e
ad
ap
tiv
e
n
atu
r
e
o
f
th
e
m
o
d
el
allo
ws
f
o
r
r
o
b
u
s
t
h
an
d
lin
g
o
f
lo
w
-
r
eso
lu
tio
n
an
d
p
ar
tially
o
cc
lu
d
ed
tex
t,
m
a
k
in
g
it
h
i
g
h
ly
v
er
s
atile.
T
h
e
in
co
r
p
o
r
atio
n
o
f
an
in
te
g
r
at
ed
m
o
d
u
le
f
u
r
t
h
er
im
p
r
o
v
es
th
e
alig
n
m
en
t
o
f
v
is
u
al
an
d
tex
tu
al
f
ea
tu
r
es.
Per
f
o
r
m
an
ce
e
v
alu
atio
n
s
o
n
th
e
C
OC
O
-
T
ex
t
d
ataset
d
em
o
n
s
tr
ate
th
at
ADF
A
R
N
s
ig
n
if
i
ca
n
tly
o
u
tp
e
r
f
o
r
m
s
ex
is
tin
g
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
,
ac
h
iev
in
g
h
ig
h
e
r
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
s
.
T
h
is
r
esear
ch
estab
lis
h
e
s
ADF
A
R
N
as
a
r
o
b
u
s
t
an
d
ef
f
icien
t
s
o
lu
tio
n
f
o
r
r
ea
l
-
wo
r
ld
tex
t
r
ec
o
g
n
itio
n
ap
p
licatio
n
s
,
p
av
in
g
t
h
e
way
f
o
r
f
u
r
th
er
a
d
v
an
ce
m
en
ts
in
th
is
d
o
m
ain
.
ACK
NO
WL
E
DG
M
E
N
T
S
We
wo
u
ld
lik
e
t
o
ex
p
r
ess
o
u
r
s
in
ce
r
e
g
r
atitu
d
e
to
o
u
r
g
u
i
d
e
f
o
r
h
is
c
o
n
s
tan
t
d
i
r
ec
tio
n
,
p
r
icele
s
s
in
s
ig
h
ts
,
an
d
s
u
p
p
o
r
t d
u
r
in
g
th
e
r
esear
ch
p
r
o
ce
s
s
.
8
0
,
5
9
8
1
,
5
9
7
7
,
4
3
7
5
,
8
6
8
8
,
7
1
8
8
,
7
5
9
4
,
0
5
9
6
,
5
0
20
40
60
80
1
0
0
1
2
0
M
a
n
j
a
r
i
et
al
.
[
1
9
]
Pr
a
b
u
an
d
S
u
n
d
a
r
[2
0
]
L
a
r
b
i
[2
1
]
T
a
r
r
i
d
e et
al
.
[
2
2
]
B
h
a
t
t
et
al
.
[
2
3
]
Vi
s
h
w
a
k
a
r
m
a
et
al
.
[2
4
]
E
S
[
2
5
]
PS
v
al
ue
M
e
t
ho
do
l
o
g
y
F1
-
sc
o
r
e
(
%)
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
A
d
a
p
tive
d
efo
r
ma
b
le
fea
tu
r
e
a
u
g
men
ta
tio
n
a
n
d
r
efin
eme
n
t n
etw
o
r
k
fo
r
s
ce
n
e
text
… (
R
a
tn
a
ma
la
S
.
P
a
til
)
839
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
a
u
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
e
d
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
R
atn
am
ala
S
.
Patil
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Gee
ta
Han
ji
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
ak
esh
Hu
d
u
d
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
le
in
[
I
E
E
E
T
r
an
s
ac
tio
n
s
o
n
Geo
s
cien
ce
an
d
R
em
o
te
Sen
s
in
g
]
at
h
ttp
://d
o
i.o
r
g
/
1
0
.
1
1
0
9
/TG
R
S.2
0
2
4
.
3
4
0
4
6
0
5
,
r
ef
er
en
ce
[
3
].
RE
F
E
R
E
NC
E
S
[
1
]
M
.
A
g
r
a
w
a
l
,
A
.
S
.
J
a
l
a
l
,
a
n
d
H
.
S
h
a
r
ma
,
“
A
d
e
e
p
l
e
a
r
n
i
n
g
b
a
se
d
st
r
a
t
e
g
i
e
s
f
o
r
s
c
e
n
e
-
t
e
x
t
VQA
s
y
st
e
m
u
sed
i
n
i
n
d
u
s
t
r
i
a
l
a
p
p
l
i
c
a
t
i
o
n
s:
a
c
r
i
t
i
c
a
l
a
n
a
l
y
s
i
s
,
”
i
n
2
0
2
4
1
st
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
n
o
v
a
t
i
v
e
S
u
s
t
a
i
n
a
b
l
e
T
e
c
h
n
o
l
o
g
i
e
s
f
o
r
E
n
e
r
g
y
,
Me
c
h
a
t
r
o
n
i
c
s
,
a
n
d
S
m
a
r
t
S
y
s
t
e
m
s
(
I
S
T
EM
S
)
,
D
e
h
r
a
d
u
n
,
I
n
d
i
a
:
I
EEE,
A
p
r
.
2
0
2
4
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
TEM
S
6
0
1
8
1
.
2
0
2
4
.
1
0
5
6
0
1
2
2
.
[
2
]
Y
.
Y
a
n
,
N
.
C
o
o
p
e
r
,
O
.
C
h
a
p
a
r
r
o
,
K
.
M
o
r
a
n
,
a
n
d
D
.
P
o
sh
y
v
a
n
y
k
,
“
S
e
m
a
n
t
i
c
G
U
I
sce
n
e
l
e
a
r
n
i
n
g
a
n
d
v
i
d
e
o
a
l
i
g
n
men
t
f
o
r
d
e
t
e
c
t
i
n
g
d
u
p
l
i
c
a
t
e
v
i
d
e
o
-
b
a
s
e
d
b
u
g
r
e
p
o
r
t
s,”
i
n
I
EEE
/
A
C
M
4
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
ren
c
e
o
n
S
o
f
t
w
a
r
e
En
g
i
n
e
e
r
i
n
g
,
N
e
w
Y
o
r
k
,
N
Y
,
U
S
A
:
A
C
M
,
A
p
r
.
2
0
2
4
,
p
p
.
1
–
13
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
9
7
5
0
3
.
3
6
3
9
1
6
3
.
[
3
]
F
.
W
a
n
g
,
X
.
Zh
u
,
X
.
Li
u
,
Y
.
Z
h
a
n
g
,
a
n
d
Y
.
L
i
,
“
S
c
e
n
e
g
r
a
p
h
-
a
w
a
r
e
h
i
e
r
a
r
c
h
i
c
a
l
f
u
s
i
o
n
n
e
t
w
o
r
k
f
o
r
r
e
m
o
t
e
s
e
n
s
i
n
g
i
m
a
g
e
r
e
t
r
i
e
v
a
l
w
i
t
h
t
e
x
t
f
e
e
d
b
a
c
k
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
G
e
o
s
c
i
e
n
c
e
a
n
d
Re
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
6
2
,
p
p
.
1
–
1
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TG
R
S
.
2
0
2
4
.
3
4
0
4
6
0
5
.
[
4
]
T.
d
o
,
T.
Tr
a
n
,
T.
N
g
u
y
e
n
,
D
.
-
D
.
L
e
,
a
n
d
T.
D
.
N
g
o
,
“
S
i
g
n
b
o
a
r
d
T
e
x
t
:
t
e
x
t
d
e
t
e
c
t
i
o
n
a
n
d
r
e
c
o
g
n
i
t
i
o
n
i
n
i
n
-
t
h
e
-
w
i
l
d
si
g
n
b
o
a
r
d
i
ma
g
e
s,
”
I
E
EE
A
c
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
6
2
9
4
2
–
6
2
9
5
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
3
9
5
3
7
4
.
[
5
]
X
.
Y
a
n
g
,
Z.
Q
i
a
o
,
J.
W
e
i
,
D
.
Y
a
n
g
,
a
n
d
Y
.
Zh
o
u
,
“
M
a
s
k
e
d
a
n
d
p
e
r
m
u
t
e
d
i
m
p
l
i
c
i
t
c
o
n
t
e
x
t
l
e
a
r
n
i
n
g
f
o
r
sc
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
EEE
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
L
e
t
t
e
rs
,
v
o
l
.
3
1
,
p
p
.
9
6
4
–
9
6
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
LSP
.
2
0
2
4
.
3
3
8
1
8
9
3
.
[
6
]
J.
Le
e
,
S
.
P
a
r
k
,
J.
B
a
e
k
,
S
.
J.
O
h
,
S
.
K
i
m
,
a
n
d
H
.
L
e
e
,
“
O
n
r
e
c
o
g
n
i
z
i
n
g
t
e
x
t
s
o
f
a
r
b
i
t
r
a
r
y
sh
a
p
e
s
w
i
t
h
2
D
se
l
f
-
a
t
t
e
n
t
i
o
n
,
”
i
n
2
0
2
0
I
EEE/
C
VF
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
W
o
rk
sh
o
p
s
(
C
VPR
W)
,
S
e
a
t
t
l
e
,
W
A
,
U
S
A
:
I
EEE,
Ju
n
.
2
0
2
0
,
p
p
.
2
3
2
6
–
2
3
3
5
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
W
5
0
4
9
8
.
2
0
2
0
.
0
0
2
8
1
.
[
7
]
Z.
C
h
e
n
g
,
Y
.
X
u
,
F
.
B
a
i
,
Y
.
N
i
u
,
S
.
P
u
,
a
n
d
S
.
Z
h
o
u
,
“
A
O
N
:
t
o
w
a
r
d
s
a
r
b
i
t
r
a
r
i
l
y
-
o
r
i
e
n
t
e
d
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
1
8
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
S
a
l
t
La
k
e
C
i
t
y
,
U
T,
U
S
A
:
I
EEE,
J
u
n
.
2
0
1
8
,
p
p
.
5
5
7
1
–
5
5
7
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
8
.
0
0
5
8
4
.
[
8
]
X
.
Y
u
e
,
Z.
K
u
a
n
g
,
C
.
Li
n
,
H
.
S
u
n
,
a
n
d
W
.
Z
h
a
n
g
,
“
R
o
b
u
s
t
S
c
a
n
n
e
r
:
d
y
n
a
mi
c
a
l
l
y
e
n
h
a
n
c
i
n
g
p
o
s
i
t
i
o
n
a
l
c
l
u
e
s
f
o
r
r
o
b
u
st
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
–
E
C
C
V
2
0
2
0
(
E
C
C
V
2
0
2
0
)
,
G
l
a
sg
o
w
,
U
K
:
S
p
r
i
n
g
e
r
,
C
h
a
m
,
2
0
2
0
,
p
p
.
1
3
5
–
1
5
1
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
030
-
5
8
5
2
9
-
7
_
9
.
[
9
]
Y
.
D
u
e
t
a
l
.
,
“
S
V
TR
:
sc
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
a
s
i
n
g
l
e
v
i
s
u
a
l
m
o
d
e
l
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
T
h
i
rt
y
-
Fi
rs
t
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
C
a
l
i
f
o
r
n
i
a
:
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
r
e
n
c
e
s
o
n
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
O
r
g
a
n
i
z
a
t
i
o
n
,
Ju
l
.
2
0
2
2
,
p
p
.
8
8
4
–
890
,
d
o
i
:
1
0
.
2
4
9
6
3
/
i
j
c
a
i
.
2
0
2
2
/
1
2
4
.
[
1
0
]
T.
Zh
e
n
g
,
Z.
C
h
e
n
,
J.
B
a
i
,
H
.
X
i
e
,
a
n
d
Y
.
-
G
.
Ji
a
n
g
,
“
TPS
+
+
:
a
t
t
e
n
t
i
o
n
-
e
n
h
a
n
c
e
d
t
h
i
n
-
p
l
a
t
e
sp
l
i
n
e
f
o
r
s
c
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
T
h
i
r
t
y
-
S
e
c
o
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
i
n
t
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
A
u
g
.
2
0
2
3
,
p
p
.
1
7
7
7
–
1
7
8
5
,
d
o
i
:
1
0
.
2
4
9
6
3
/
i
j
c
a
i
.
2
0
2
3
/
1
9
7
.
[
1
1
]
R
.
Y
a
n
,
L
.
P
e
n
g
,
S
.
X
i
a
o
,
a
n
d
G
.
Y
a
o
,
“
P
r
i
mi
t
i
v
e
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
f
o
r
s
c
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
2
1
I
EEE
/
C
VF
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
(
C
VP
R)
,
N
a
sh
v
i
l
l
e
,
TN
,
U
S
A
:
I
EEE,
J
u
n
.
2
0
2
1
,
p
p
.
2
8
4
–
2
9
3
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
6
4
3
7
.
2
0
2
1
.
0
0
0
3
5
.
[
1
2
]
T.
W
a
n
g
e
t
a
l
.
,
“
D
e
c
o
u
p
l
e
d
a
t
t
e
n
t
i
o
n
n
e
t
w
o
r
k
f
o
r
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
AA
AI
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
4
,
n
o
.
7
,
p
p
.
1
2
2
1
6
–
1
2
2
2
4
,
A
p
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
4
i
0
7
.
6
9
0
3
.
[
1
3
]
P
.
N
.
D
e
e
l
a
k
a
,
D
.
R
.
J
a
y
a
k
o
d
i
,
a
n
d
D
.
Y
.
S
i
l
v
a
,
“
G
e
o
me
t
r
i
c
p
e
r
c
e
p
t
i
o
n
b
a
se
d
e
f
f
i
c
i
e
n
t
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
”
a
r
Xi
v
:
2
3
0
2
.
0
3
8
7
3
,
2
0
2
3
[
1
4
]
A
.
A
.
A
.
A
l
s
h
a
w
i
,
J.
Ta
n
h
a
,
a
n
d
M
.
A
.
B
a
l
a
f
a
r
,
“
A
n
a
t
t
e
n
t
i
o
n
-
b
a
s
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
s
c
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
8
1
2
3
–
8
1
3
4
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
3
5
2
7
4
8
.
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
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
831
-
8
4
0
840
[
1
5
]
J.
-
N
.
L
i
,
X
.
-
Q
.
L
i
u
,
X
.
Lu
o
,
a
n
d
X
.
-
S
.
X
u
,
“
V
O
LT
ER
:
v
i
su
a
l
c
o
l
l
a
b
o
r
a
t
i
o
n
a
n
d
d
u
a
l
-
st
r
e
a
m
f
u
s
i
o
n
f
o
r
sc
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
2
6
,
p
p
.
6
4
3
7
–
6
4
4
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
T
M
M
.
2
0
2
4
.
3
3
5
0
9
1
6
.
[
1
6
]
X.
-
Q
.
Li
u
,
X
.
-
Y
.
D
i
n
g
,
X
.
Lu
o
,
a
n
d
X
.
-
S
.
X
u
,
“
P
r
o
t
o
U
D
A
:
p
r
o
t
o
t
y
p
e
-
b
a
s
e
d
u
n
su
p
e
r
v
i
s
e
d
a
d
a
p
t
a
t
i
o
n
f
o
r
c
r
o
ss
-
d
o
ma
i
n
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
3
6
,
n
o
.
1
2
,
p
p
.
9
0
9
6
–
9
1
0
8
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
2
3
.
3
3
4
4
7
6
1
.
[
1
7
]
P
.
P
u
j
a
r
,
A
.
K
u
m
a
r
,
a
n
d
V
.
K
u
mar
,
“
Ef
f
i
c
i
e
n
t
p
l
a
n
t
l
e
a
f
d
e
t
e
c
t
i
o
n
t
h
r
o
u
g
h
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
b
a
se
d
o
n
c
o
r
n
l
e
a
f
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
I
AE
S
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
1
1
3
9
–
1
1
4
8
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
1
.
p
p
1
1
3
9
-
1
1
4
8
.
[
1
8
]
S
.
H
.
S
r
e
e
d
h
a
r
a
,
V
.
K
u
mar,
a
n
d
S
.
S
a
l
ma
,
“
Ef
f
i
c
i
e
n
t
b
i
g
d
a
t
a
c
l
u
s
t
e
r
i
n
g
u
si
n
g
a
d
h
o
c
f
u
z
z
y
C
me
a
n
s
a
n
d
a
u
t
o
-
e
n
c
o
d
e
r
C
N
N
,
”
i
n
I
n
v
e
n
t
i
v
e
C
o
m
p
u
t
a
t
i
o
n
a
n
d
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
i
e
s
,
N
e
w
D
e
l
h
i
,
I
n
d
i
a
:
S
p
r
i
n
g
e
r
,
S
i
n
g
a
p
o
r
e
,
2
0
2
3
,
p
p
.
3
5
3
–
3
6
8
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
19
-
7
4
0
2
-
1
_
2
5
.
[
1
9
]
K
.
M
a
n
j
a
r
i
,
M
.
V
e
r
m
a
,
G
.
S
i
n
g
a
l
,
a
n
d
S
.
N
a
m
a
su
d
r
a
,
“
Q
EST:
q
u
a
n
t
i
z
e
d
a
n
d
e
f
f
i
c
i
e
n
t
sce
n
e
t
e
x
t
d
e
t
e
c
t
o
r
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,”
AC
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
Asi
a
n
a
n
d
L
o
w
-
R
e
so
u
r
c
e
L
a
n
g
u
a
g
e
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ss
i
n
g
,
v
o
l
.
2
2
,
n
o
.
5
,
p
p
.
1
–
1
8
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
2
6
2
1
7
.
[
2
0
]
S
.
P
r
a
b
u
a
n
d
K
.
J
.
A
.
S
u
n
d
a
r
,
“
En
h
a
n
c
e
d
a
t
t
e
n
t
i
o
n
-
b
a
s
e
d
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
f
r
a
mew
o
r
k
f
o
r
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
n
t
e
l
l
i
g
e
n
t
A
u
t
o
m
a
t
i
o
n
& S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
3
5
,
n
o
.
2
,
p
p
.
2
0
7
1
–
2
0
8
6
,
2
0
2
3
,
d
o
i
:
1
0
.
3
2
6
0
4
/
i
a
sc
.
2
0
2
3
.
0
2
9
1
0
5
.
[
2
1
]
G
.
La
r
b
i
,
“
Tw
o
-
st
e
p
t
e
x
t
d
e
t
e
c
t
i
o
n
f
r
a
mew
o
r
k
i
n
n
a
t
u
r
a
l
s
c
e
n
e
s
b
a
s
e
d
o
n
P
seu
d
o
-
Ze
r
n
i
k
e
m
o
me
n
t
s
a
n
d
C
N
N
,
”
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
.
7
,
p
p
.
1
0
5
9
5
–
1
0
6
1
6
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
0
4
2
-
022
-
1
3
6
9
0
-
6.
[
2
2
]
S
.
T
a
r
r
i
d
e
e
t
a
l
.
,
“
La
r
g
e
-
s
c
a
l
e
g
e
n
e
a
l
o
g
i
c
a
l
i
n
f
o
r
m
a
t
i
o
n
e
x
t
r
a
c
t
i
o
n
f
r
o
m
h
a
n
d
w
r
i
t
t
e
n
Q
u
e
b
e
c
p
a
r
i
s
h
r
e
c
o
r
d
s
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
n
D
o
c
u
m
e
n
t
A
n
a
l
y
s
i
s
a
n
d
R
e
c
o
g
n
i
t
i
o
n
,
v
o
l
.
2
6
,
n
o
.
3
,
p
p
.
2
5
5
–
2
7
2
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
0
0
3
2
-
0
2
3
-
0
0
4
2
7
-
w.
[
2
3
]
R
.
B
h
a
t
t
,
A
.
R
a
i
,
S
.
C
h
a
n
d
a
,
a
n
d
N
.
C
.
K
r
i
s
h
n
a
n
,
“
P
h
o
(
S
C
)
-
C
TC
—
a
h
y
b
r
i
d
a
p
p
r
o
a
c
h
t
o
w
a
r
d
s z
e
r
o
-
sh
o
t
w
o
r
d
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
n
D
o
c
u
m
e
n
t
An
a
l
y
si
s
a
n
d
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
2
6
,
n
o
.
1
,
p
p
.
5
1
–
6
3
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
0
3
2
-
0
2
2
-
0
0
4
0
7
-
6.
[
2
4
]
D
.
K
.
V
i
sh
w
a
k
a
r
ma,
P
.
M
e
e
l
,
A
.
Y
a
d
a
v
,
a
n
d
K
.
S
i
n
g
h
,
“
A
f
r
a
mew
o
r
k
o
f
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
o
n
w
e
b
p
l
a
t
f
o
r
m
u
si
n
g
C
o
n
v
N
e
t
,
”
S
o
c
i
a
l
N
e
t
w
o
r
k
A
n
a
l
y
si
s
a
n
d
M
i
n
i
n
g
,
v
o
l
.
1
3
,
n
o
.
1
,
Ja
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
3
2
7
8
-
023
-
0
1
0
2
6
-
7.
[
2
5
]
T.
G
e
n
g
,
“
Tr
a
n
sf
o
r
m
i
n
g
s
c
e
n
e
t
e
x
t
d
e
t
e
c
t
i
o
n
a
n
d
r
e
c
o
g
n
i
t
i
o
n
:
a
m
u
l
t
i
-
s
c
a
l
e
e
n
d
-
to
-
e
n
d
a
p
p
r
o
a
c
h
w
i
t
h
t
r
a
n
sf
o
r
mer
f
r
a
mew
o
r
k
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
4
0
5
8
2
–
4
0
5
9
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
3
7
5
4
9
7
.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
Ra
tn
a
m
a
l
a
S
.
Pa
til
re
c
e
iv
e
d
h
e
r
b
a
c
h
e
lo
r’s
d
e
g
re
e
i
n
El
e
c
tro
n
ics
a
n
d
Co
m
m
u
n
ica
ti
o
n
E
n
g
in
e
e
rin
g
fr
o
m
th
e
Visv
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
Un
i
v
e
rsity
,
Be
lg
a
u
m
,
In
d
ia
in
2
0
1
4
a
n
d
m
a
ste
r
d
e
g
r
e
e
in
Di
g
it
a
l
C
o
m
m
u
n
ica
ti
o
n
a
n
d
Ne
two
rk
i
n
g
fro
m
sa
m
e
u
n
i
v
e
rsity
i
n
2
0
1
6
.
S
h
e
is cu
rre
n
tl
y
p
u
rs
u
in
g
h
e
r
P
h
.
D.
d
e
g
re
e
fro
m
th
e
sa
m
e
u
n
iv
e
rsity
.
S
h
e
is
p
re
se
n
tl
y
wo
rk
in
g
a
s
a
ss
istan
t
p
r
o
fe
ss
o
r
in
De
p
a
rtme
n
ts
o
f
El
e
c
tr
o
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
En
g
i
n
e
e
rin
g
S
h
a
rn
b
a
sv
a
Un
i
v
e
rsity
Ka
lab
u
ra
g
i
,
Ka
rn
a
tak
a
,
In
d
ia.
He
r
p
rima
ry
a
re
a
o
f
in
tere
st
is
ima
g
e
p
r
o
c
e
ss
in
g
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
p
a
tt
e
r
n
re
c
o
g
n
i
ti
o
n
.
Sh
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
ra
tn
a
m
a
la_
1
2
@re
d
iffma
il
.
c
o
m
.
G
e
e
ta
H
a
n
ji
wo
rk
in
g
p
re
se
n
tl
y
a
s
p
r
o
fe
ss
o
r
in
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ics
a
n
d
Co
m
m
u
n
ica
ti
o
n
E
n
g
i
n
e
e
rin
g
,
P
o
o
jy
a
D
o
d
d
a
p
p
a
A
p
p
a
C
o
ll
e
g
e
o
f
En
g
in
e
e
rin
g
,
Ka
lab
u
ra
g
i
.
S
h
e
h
a
s
1
8
y
e
a
rs
o
f
tea
c
h
in
g
a
n
d
1
0
y
e
a
rs
o
f
re
se
a
rc
h
e
x
p
e
rien
c
e
,
a
n
d
c
o
m
p
lete
d
h
e
r
B.
E
.
,
M
.
Tec
h
.
,
a
n
d
P
h
.
D
.
in
El
e
c
tro
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
E
n
g
in
e
e
rin
g
.
He
r
re
se
a
rc
h
a
re
a
in
c
lu
d
e
s
d
ig
i
tal
ima
g
e
p
ro
c
e
ss
in
g
a
n
d
p
a
t
tern
re
c
o
g
n
it
i
o
n
.
S
h
e
p
u
b
li
s
h
e
d
m
o
re
t
h
a
n
55
re
se
a
rc
h
p
a
p
e
rs
in
a
b
o
v
e
m
e
n
ti
o
n
e
d
a
re
a
s
.
S
h
e
h
a
s
3
0
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
a
n
d
1
8
y
e
a
rs
o
f
re
se
a
rc
h
e
x
p
e
rien
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
g
e
e
tan
jalip
a
ti
l1
2
3
@
g
m
a
il
.
c
o
m
or
g
e
e
tah
a
n
j
i@p
d
a
e
n
g
g
.
c
o
m
Dr
.
Ra
k
e
sh
H
u
d
u
d
h
a
v
e
e
a
rn
e
d
a
n
e
n
g
in
e
e
rin
g
d
e
g
re
e
i
n
El
e
c
tr
o
n
ics
a
n
d
Co
m
m
u
n
ica
ti
o
n
fr
o
m
S
DM
Co
l
l
e
g
e
o
f
En
g
in
e
e
rin
g
i
n
D
h
a
rwa
d
,
a
ffil
iate
d
with
Visv
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
Un
iv
e
rsit
y
,
Be
lag
a
v
i,
i
n
2
0
1
1
.
F
o
ll
o
we
d
b
y
a
n
M
.
Tec
h
.
d
e
g
re
e
fro
m
P
o
o
jy
a
Do
d
d
a
p
p
a
Ap
p
a
Co
l
leg
e
o
f
En
g
in
e
e
rin
g
i
n
Ka
lab
u
ra
g
i,
a
lso
a
f
fil
iate
d
wit
h
Visv
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
U
n
iv
e
rsit
y
,
Be
lag
a
v
i,
in
2
0
1
3
,
a
n
d
c
u
lmin
a
ti
n
g
wit
h
a
P
h
.
D.
i
n
Im
a
g
e
P
ro
c
e
ss
in
g
fr
o
m
S
ri
S
a
t
y
a
S
a
i
Un
iv
e
rsit
y
o
f
Tec
h
n
o
l
o
g
y
a
n
d
M
e
d
ica
l
S
c
ien
c
e
s
in
S
e
h
o
re
i
n
2
0
1
9
,
c
u
rre
n
tl
y
se
rv
e
s
a
s
a
n
a
ss
istan
t
p
ro
fe
ss
o
r
a
t
P
o
o
jy
a
D
o
d
d
a
p
p
a
Ap
p
a
Co
ll
e
g
e
o
f
En
g
i
n
e
e
rin
g
in
Ka
lab
u
ra
g
i,
Ka
r
n
a
tak
a
.
Wi
t
h
o
v
e
r
fiv
e
y
e
a
rs
o
f
e
x
p
e
rien
c
e
,
a
n
d
c
u
rre
n
tl
y
g
u
i
d
in
g
re
se
a
rc
h
sc
h
o
lar
u
n
d
e
r
V
isv
e
sv
a
ra
y
a
Tec
h
n
o
lo
g
ica
l
Un
i
v
e
rsity
i
n
Be
lag
a
v
i
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
rh
u
d
e
d
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.