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
.
4
,
A
u
g
u
s
t
20
26
,
p
p
.
3
5
3
7
~
3
5
4
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
5
3
7
-
3
5
4
5
3537
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
E
-
TE
XTL
O
C
:
e
ff
icient
te
x
t
lo
ca
liza
tion a
nd e
x
trac
tion in
real
-
wo
rld video
s
cenes
Da
y
a
na
nd
a
K
o
da
la
J
a
y
a
ra
m
1,
2
,
P
utt
eg
o
wda
Dev
eg
o
wda
1
1
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
A
T
M
E
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
a
f
f
i
l
i
a
t
e
d
t
o
V
i
s
v
e
s
v
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
,
B
e
l
a
g
a
v
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
V
i
d
y
a
v
a
r
d
h
a
k
a
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
(
A
u
t
o
n
o
m
o
u
s
I
n
st
i
t
u
t
e
)
a
f
f
i
l
i
a
t
e
d
t
o
V
i
sv
e
sv
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
,
B
e
l
a
g
a
v
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
Sep
6
,
2
0
2
5
R
ev
is
ed
May
22
,
2
0
2
6
Acc
ep
ted
J
u
l
9
,
2
0
2
6
Lo
c
a
li
z
a
ti
o
n
o
f
a
n
a
c
c
u
ra
te
tex
t
is
q
u
it
e
a
c
o
m
p
l
ica
ted
issu
e
,
e
sp
e
c
i
a
ll
y
wh
e
n
a
tt
e
m
p
ti
n
g
to
e
x
trac
t
fro
m
a
c
o
m
p
lex
v
id
e
o
.
It
is
m
a
in
ly
d
u
e
to
th
e
li
m
it
a
ti
o
n
o
f
re
so
u
rc
e
s,
d
y
n
a
m
ic
b
a
c
k
g
ro
u
n
d
,
a
n
d
tex
t
v
a
riab
il
it
y
.
Th
e
re
is
v
a
rio
u
s
a
rti
ficia
l
in
telli
g
e
n
c
e
b
a
se
d
m
e
th
o
d
s
a
d
o
p
ti
n
g
m
a
c
h
i
n
e
lea
rn
in
g
f
o
r
a
d
d
re
ss
in
g
su
c
h
issu
e
s
e
n
c
o
u
n
t
e
r
lo
we
r
p
o
siti
o
n
a
l
a
c
c
u
ra
c
y
a
n
d
in
c
u
r
m
a
x
imiz
e
d
c
o
m
p
u
tati
o
n
a
l
c
o
st.
Th
e
re
fo
re
,
t
h
e
p
ro
p
o
se
d
sy
ste
m
in
tro
d
u
c
e
s
e
fficie
n
t
tex
t
l
o
c
a
li
z
a
ti
o
n
a
n
d
e
x
trac
ti
o
n
fo
r
c
o
m
p
re
h
e
n
siv
e
p
o
siti
o
n
a
l
a
c
c
u
ra
c
y
in
c
o
m
p
lex
v
id
e
o
s
(E
-
TE
XTLOC).
Diffe
re
n
t
fro
m
c
o
n
v
e
n
ti
o
n
a
l
a
p
p
ro
a
c
h
e
s,
E
-
TE
XTLOC
fa
c
il
it
a
tes
sa
m
p
li
n
g
o
f
v
i
d
e
o
fra
m
e
s
wh
il
e
M
o
b
i
leN
e
tV2
is d
e
p
l
o
y
e
d
t
o
wa
rd
s fas
ter l
o
c
a
li
z
a
ti
o
n
o
f
tex
t
.
Th
e
o
u
tco
m
e
o
f
re
c
o
g
n
ize
d
tex
t
is f
u
rt
h
e
r
re
fin
e
d
b
y
a
v
e
rifi
e
r
m
o
d
u
le,
w
h
ich
p
r
o
v
i
d
e
s a
se
lf
-
su
p
e
rv
ise
d
re
sp
o
n
se
.
As
se
ss
e
d
o
n
t
h
e
Yo
u
Tu
b
e
v
id
e
o
d
a
tas
e
t,
th
e
p
ro
p
o
se
d
m
o
d
e
l
a
c
c
o
m
p
li
sh
e
s
9
8
.
2
%
a
c
c
u
ra
c
y
with
2
9
.
6
m
s
to
wa
r
d
s
g
e
n
e
ra
ti
n
g
a
n
a
ly
ti
c
a
l
o
u
tco
m
e
s.
I
t
m
e
a
n
s
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
a
c
c
o
m
p
l
ish
e
s
6
-
1
0
%
a
c
c
u
ra
c
y
e
n
h
a
n
c
e
m
e
n
t
with
a
4
0
-
5
0
%
re
d
u
c
ti
o
n
o
f
sp
e
e
d
i
n
c
o
n
tr
a
st
to
th
e
e
x
isti
n
g
sy
ste
m
.
Th
e
imp
li
c
a
ti
o
n
s
o
f
t
h
e
p
ro
p
o
se
d
stu
d
y
c
a
n
b
e
sta
ted
to
wa
rd
s
su
rv
e
i
ll
a
n
c
e
sy
ste
m
,
a
u
to
n
o
m
o
u
s
v
e
h
icle
s,
a
n
d
a
ss
isti
v
e
d
e
v
ice
s
th
a
t
wo
rk
s i
n
re
a
l
-
ti
m
e
.
K
ey
w
o
r
d
s
:
Acc
u
r
ac
y
Ma
ch
in
e
lear
n
in
g
T
ex
t d
etec
tio
n
T
ex
t lo
ca
lizatio
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
:
Day
an
an
d
a
Ko
d
ala
J
ay
ar
am
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
AT
ME
C
o
lleg
e
o
f
E
n
g
i
n
ee
r
in
g
a
f
f
iliated
to
Vis
v
esv
ar
ay
a
T
ec
h
n
o
lo
g
ical
U
n
iv
er
s
ity
B
an
n
u
r
R
d
,
My
s
u
r
u
,
Kar
n
ata
k
a
–
5
7
0
0
2
8
,
I
n
d
ia
E
m
ail:
d
ay
an
an
d
a.
k
em
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
ter
m
tex
t
lo
ca
lizatio
n
r
ef
er
s
to
th
e
m
ec
h
an
is
m
o
f
d
ete
r
m
in
in
g
th
e
p
o
s
itio
n
o
f
tex
tu
al
co
n
ten
t
p
r
esen
t
with
in
an
y
f
o
r
m
o
f
m
u
ltime
d
ia
o
b
ject.
T
h
er
e
ar
e
tw
o
o
th
er
eq
u
iv
alen
t
ter
m
s
k
n
o
wn
as
tex
t
d
etec
tio
n
an
d
tex
t
r
ec
o
g
n
itio
n
,
w
h
ich
ar
e
clo
s
ely
co
n
n
ec
ted
to
tex
t
lo
c
aliza
tio
n
,
wh
er
e
t
h
e
f
o
r
m
er
o
n
e
r
elate
s
to
f
in
d
in
g
th
e
tex
t
elem
e
n
t
with
in
an
i
m
ag
e
o
r
v
id
eo
f
r
am
e
,
wh
ile
th
e
latter
r
elate
s
to
th
e
m
eth
o
d
o
f
d
ec
o
d
i
n
g
th
e
co
n
tex
t
o
f
ac
tu
al
tex
t
f
r
o
m
a
lo
ca
lized
r
eg
io
n
[
1
]
.
T
h
e
s
ig
n
if
ican
ce
o
f
tex
t
lo
ca
lizatio
n
is
th
at
it
ac
ts
as
a
f
o
u
n
d
atio
n
f
o
r
tex
t
r
ec
o
g
n
iti
o
n
,
wh
ile
th
ey
ar
e
q
u
ite
ess
en
tial
in
v
ar
io
u
s
r
ea
l
-
wo
r
ld
ap
p
licatio
n
s
,
v
iz.
,
ass
is
tiv
e
tech
n
o
lo
g
y
,
au
g
m
e
n
ted
r
ea
lity
,
d
o
cu
m
en
t
d
ig
itizatio
n
,
an
d
a
u
to
n
o
m
o
u
s
v
e
h
icles
[
2
]
–
[
4
]
.
T
ex
t
lo
ca
lizatio
n
also
ass
is
ts
in
ef
f
icien
t
p
r
o
ce
s
s
in
g
as
th
e
s
y
s
tem
d
o
esn
’
t
r
e
q
u
ir
e
e
x
ec
u
tin
g
o
p
tical
c
h
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
o
n
th
e
co
m
p
lete
im
ag
e,
as
lo
ca
lizatio
n
r
eg
io
n
s
ar
e
o
n
ly
s
u
b
jecte
d
to
f
u
r
th
er
p
r
o
ce
s
s
in
g
.
T
h
er
e
ar
e
v
ar
io
u
s
r
esear
ch
-
b
ased
ap
p
r
o
ac
h
es
to
war
d
s
tex
t
lo
ca
lizatio
n
,
v
iz.
,
tex
tu
r
e
-
b
ased
,
s
em
an
tic
s
eg
m
en
tatio
n
,
an
d
r
e
g
io
n
p
r
o
p
o
s
al
m
eth
o
d
[
5
]
–
[
7
]
.
I
r
r
esp
e
ctiv
e
o
f
v
ar
io
u
s
r
esear
c
h
-
b
ase
d
o
n
g
o
in
g
s
o
lu
tio
n
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
3
7
-
3
5
4
5
3538
th
er
e
ar
e
v
ar
io
u
s
o
n
g
o
in
g
r
esear
ch
ch
allen
g
es,
v
iz.
,
tex
t
v
a
r
iab
ilit
y
,
v
is
u
al
v
ar
iatio
n
,
g
en
e
r
aliza
tio
n
,
r
eso
u
r
ce
lim
itatio
n
,
an
d
lay
o
u
t c
o
m
p
lex
ity
[
8
]
.
Ma
ch
in
e
lear
n
in
g
-
b
ased
m
eth
o
d
s
h
av
e
ev
o
lv
e
d
as
o
n
e
o
f
th
e
p
r
o
m
is
in
g
s
o
lu
tio
n
to
war
d
s
tex
t
lo
ca
lizatio
n
[
9
]
.
I
t
is
d
ep
lo
y
ed
f
o
r
au
to
m
ated
f
ea
tu
r
e
le
ar
n
in
g
wh
er
e
h
ier
ar
ch
ical
f
e
atu
r
es
ar
e
lear
n
ed
au
to
n
o
m
o
u
s
ly
f
r
o
m
d
ata.
Var
io
u
s
m
ac
h
in
e
lear
n
in
g
m
o
d
el
s
ar
e
u
s
ed
s
p
ec
if
ically
to
war
d
s
lo
ca
lizatio
n
an
d
r
ec
o
g
n
itio
n
,
v
iz.
,
ch
ar
ac
ter
r
e
g
io
n
awa
r
e
n
ess
f
o
r
tex
t
d
etec
tio
n
(
C
R
AFT)
,
ef
f
icien
t
an
d
ac
cu
r
ate
s
ce
n
e
tex
t
d
etec
to
r
(
E
AST)
,
a
n
d
co
n
n
e
ctio
n
is
t
tex
t
p
r
o
p
o
s
al
n
etwo
r
k
(
C
T
PN)
.
T
h
ese
m
o
d
els
ar
e
q
u
ite
ca
p
ab
le
o
f
id
en
tify
in
g
b
o
th
n
o
n
-
s
tan
d
ar
d
an
d
n
o
n
-
lin
ea
r
lay
o
u
ts
o
f
tex
t
r
eg
io
n
s
.
Var
io
u
s
lig
h
tweig
h
t
m
o
d
els,
lik
e
v
ar
ian
ts
o
f
M
o
b
ileNet,
f
ac
ilit
ate
r
ea
l
-
tim
e
d
ep
l
o
y
m
en
t
o
n
r
e
s
o
u
r
ce
-
co
n
s
tr
ain
e
d
d
e
v
ices
an
d
th
er
eb
y
o
p
tim
ize
laten
cy
.
Ad
o
p
tio
n
o
f
m
ac
h
in
e
lear
n
in
g
also
p
r
o
v
id
es
th
e
s
y
s
tem
to
u
n
d
er
g
o
lear
n
in
g
o
p
er
at
io
n
s
with
in
v
ar
ian
t
r
ep
r
esen
tatio
n
s
,
wh
ich
ca
n
g
e
n
er
alize
q
u
ite
well
o
n
d
iv
er
s
e
co
n
d
itio
n
s
.
Var
io
u
s
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
(
e.
g
.
,
v
is
io
n
tr
an
s
f
o
r
m
er
s
)
p
r
o
v
id
e
m
o
d
ellin
g
o
f
g
lo
b
al
co
n
tex
t
an
d
o
f
f
er
ex
p
licit
atten
tio
n
s
.
T
h
is
ap
p
r
o
ac
h
p
o
ten
tially
ass
is
ts
in
em
p
h
asizin
g
th
e
r
elev
an
t
r
eg
io
n
o
f
te
x
t
f
o
r
an
y
g
iv
e
n
co
n
d
itio
n
o
f
s
ce
n
es.
Alth
o
u
g
h
m
ac
h
in
e
lear
n
i
n
g
a
n
d
d
ee
p
lear
n
in
g
o
f
f
e
r
s
ig
n
if
ican
t
co
n
tr
ib
u
tio
n
s
to
war
d
s
s
o
lv
in
g
tex
t
lo
ca
lizatio
n
p
r
o
b
lem
s
,
th
er
e
ar
e
v
ar
io
u
s
p
o
ten
tial
s
etb
ac
k
s
to
o
th
at
ar
e
co
n
s
is
ten
tly
b
ein
g
a
p
ar
t
o
f
o
n
g
o
in
g
in
v
esti
g
atio
n
.
T
h
e
p
r
im
a
r
y
b
ig
g
e
r
ch
allen
g
e
in
th
e
d
ep
lo
y
m
en
t
o
f
m
ac
h
i
n
e
lear
n
in
g
ap
p
r
o
ac
h
es
is
r
ea
l
-
wo
r
ld
v
a
r
iab
ilit
y
,
wh
ich
is
ch
a
r
ac
ter
ized
b
y
a
d
y
n
am
ic
en
v
ir
o
n
m
en
t
with
u
n
s
ee
n
lay
o
u
ts
,
s
cr
ip
ts
,
a
n
d
f
o
n
ts
.
Usu
ally
,
m
o
d
els
a
r
e
tr
ain
ed
o
n
a
clea
n
er
v
e
r
s
io
n
o
f
th
e
d
ataset,
wh
ich
f
ails
wh
e
n
s
u
b
jecte
d
to
d
if
f
er
e
n
t
u
n
f
o
r
e
s
ee
n
en
v
ir
o
n
m
en
ts
.
Var
io
u
s
d
y
n
am
ic
c
o
n
d
itio
n
s
l
ik
e
b
ac
k
g
r
o
u
n
d
n
o
is
e,
m
o
tio
n
b
lu
r
,
o
cc
lu
s
io
n
,
an
d
lig
h
tin
g
ca
n
ea
s
ily
d
eg
r
ad
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el.
Var
io
u
s
h
ig
h
-
p
er
f
o
r
m
a
n
ce
m
o
d
el
o
f
m
ac
h
in
e
lear
n
in
g
(
e.
g
.
,
tr
an
s
f
o
r
m
er
)
d
em
an
d
s
s
u
p
p
o
r
t
o
f
g
r
ap
h
ical
o
r
ten
s
o
r
p
r
o
ce
s
s
in
g
u
n
it
.
T
h
is
n
o
t
o
n
ly
c
o
n
s
u
m
es
ex
te
n
s
iv
e
r
eso
u
r
ce
s
b
u
t
also
d
eg
r
ad
es th
e
b
alan
ce
am
o
n
g
a
cc
u
r
ac
y
,
s
p
ee
d
,
an
d
p
o
wer
d
e
m
an
d
s
.
Var
io
u
s
r
elate
d
wo
r
k
s
h
av
e
b
ee
n
s
tu
d
ied
to
u
n
d
er
s
tan
d
th
e
ad
o
p
tio
n
o
f
m
ac
h
in
e
lear
n
in
g
to
war
d
s
tex
t
lo
ca
lizatio
n
.
E
x
is
tin
g
s
y
s
tem
s
h
av
e
b
ee
n
d
o
m
in
a
n
tly
witn
ess
ed
to
ad
o
p
t
co
n
v
o
lu
t
io
n
n
eu
r
al
n
etwo
r
k
(
C
NN)
to
war
d
s
lo
ca
lizatio
n
o
f
tex
t
co
n
ten
ts
b
y
in
teg
r
atin
g
C
NN
with
v
ar
io
u
s
o
t
h
er
d
ee
p
lear
n
in
g
m
o
d
els,
e.
g
.
,
tr
a
n
s
f
o
r
m
e
r
an
d
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
[
1
0
]
–
[
1
4
]
.
A
lth
o
u
g
h
th
ese
C
NN
-
b
ased
ap
p
r
o
ac
h
es
ar
e
k
n
o
wn
to
ad
d
r
ess
is
s
u
es p
er
tain
in
g
to
th
e
in
teg
r
atio
n
o
f
te
x
t in
to
th
e
d
etec
tio
n
o
f
o
b
jects,
s
h
ap
es o
f
ir
r
eg
u
lar
tex
t,
an
d
r
ec
o
g
n
itio
n
o
f
b
ilin
g
u
al
tex
t,
th
ey
d
em
an
d
m
ass
iv
e
an
n
o
t
ated
d
ata
an
d
ar
e
h
ig
h
ly
ch
a
llen
g
in
g
to
h
an
d
le
o
cc
lu
d
ed
in
s
tan
ce
s
o
f
tex
t.
T
h
er
e
ar
e
v
ar
io
u
s
r
esear
ch
m
o
d
e
ls
d
esig
n
ed
u
s
in
g
h
is
to
g
r
am
o
f
o
r
ien
ted
g
r
ad
ien
t
s
(
HOG)
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
to
war
d
s
s
ce
n
e
tex
t
d
etec
tio
n
[
1
5
]
–
[
1
9
]
.
T
h
ese
ap
p
r
o
ac
h
es
ar
e
k
n
o
wn
to
ac
co
m
p
lis
h
a
d
ec
en
t
ac
cu
r
ac
y
o
f
r
ec
o
g
n
itio
n
th
at
i
s
f
o
u
n
d
s
u
itab
le
f
o
r
n
o
n
-
E
n
g
li
s
h
tex
tu
al
co
n
ten
ts
with
s
im
p
lifie
d
f
ea
tu
r
e
ex
tr
ac
t
io
n
.
Ho
wev
er
,
HOG
s
ch
em
es
with
SVM
ar
e
also
witn
ess
ed
with
s
h
o
r
tco
m
in
g
s
th
at
ar
e
r
ep
o
r
ted
to
lim
it
th
e
p
er
f
o
r
m
a
n
ce
in
co
n
tr
ast
to
m
o
d
er
n
tim
es
o
f
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
es.
HOG
h
as
also
b
ee
n
in
teg
r
ated
with
an
o
th
er
e
f
f
icien
t
m
ac
h
in
e
l
ea
r
n
in
g
al
g
o
r
ith
m
o
f
r
an
d
o
m
f
o
r
est
to
war
d
s
ac
co
m
p
lis
h
in
g
b
etter
p
er
f
o
r
m
an
ce
o
f
tex
t
l
o
ca
lizatio
n
[
2
0
]
–
[
2
6
]
.
T
h
ese
s
tu
d
ies
p
er
f
o
r
m
th
e
ex
t
r
ac
tio
n
o
f
u
n
iq
u
e
f
ea
t
u
r
es wh
er
e
HOG
is
u
tili
ze
d
f
o
r
ex
tr
ac
tin
g
g
r
ad
ien
t
-
o
r
ien
ted
f
ea
tu
r
es,
wh
ile
th
ey
ar
e
also
in
teg
r
ated
with
v
ar
io
u
s
ty
p
es
o
f
m
o
m
e
n
ts
.
Ho
wev
er
,
s
u
ch
ap
p
r
o
ac
h
es
ar
e
n
o
t
f
o
u
n
d
to
with
s
tan
d
co
m
p
lex
v
ar
iatio
n
i
n
s
ce
n
e
tex
t.
T
h
e
r
esear
ch
p
r
o
b
lem
s
id
en
tif
ied
ar
e
i)
m
ajo
r
ity
o
f
t
h
e
ex
is
tin
g
s
y
s
tem
is
f
o
u
n
d
to
wo
r
k
in
g
f
in
e
with
th
eir
d
ataset
with
s
ta
tic
v
id
eo
s
ce
n
e
b
u
t n
o
t m
u
ch
ev
i
d
en
ce
i
s
p
u
t f
o
r
war
d
ed
to
ad
d
r
ess
in
ac
cu
r
ate
lo
ca
lizatio
n
o
f
tex
t
f
o
r
co
m
p
lex
s
ce
n
es
o
f
v
id
eo
,
ii)
alm
o
s
t
all
th
e
ap
p
r
o
ac
h
es
g
iv
es
m
o
r
e
i
m
p
o
r
tan
ce
to
s
o
lv
e
d
etec
tio
n
-
b
ased
p
r
o
b
lem
b
u
t
n
o
t
m
u
ch
e
m
p
h
asis
is
to
war
d
s
in
v
esti
g
atin
g
l
o
ca
tio
n
-
b
ased
in
f
o
r
m
atio
n
v
iz
.
b
o
u
n
d
in
g
co
o
r
d
in
ates,
s
ize,
a
n
d
o
r
ien
tatio
n
,
iii)
ir
r
esp
ec
tiv
e
o
f
ex
h
ib
its
o
f
s
o
m
e
h
ig
h
er
ac
cu
r
ac
y
s
co
r
es
,
m
ajo
r
ity
o
f
ex
is
tin
g
m
o
d
els
ad
o
p
ts
m
ac
h
i
n
e
lear
n
in
g
a
p
p
r
o
ac
h
es
th
at
will
d
ef
in
itel
y
r
esu
lt
in
h
i
g
h
er
co
m
p
u
tatio
n
al
co
s
t
an
d
in
c
r
ea
s
ed
in
f
er
en
ce
tim
e,
an
d
i
v
)
co
n
v
e
n
tio
n
al
f
r
a
m
ewo
r
k
th
at
ar
e
tr
ain
ed
o
n
p
ar
ticu
lar
d
ataset
d
o
esn
’
t
g
en
er
alize
o
p
tim
ally
to
m
u
ltime
d
i
a
d
ataset
with
an
in
clu
s
io
n
o
f
d
if
f
er
en
t
lan
g
u
ag
e,
m
o
tio
n
,
a
n
d
f
o
n
ts
.
T
h
e
p
r
o
b
l
em
o
f
ac
c
u
r
ate
lo
ca
lizatio
n
o
f
tex
t
is
ad
d
r
ess
ed
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
b
y
p
r
esen
tin
g
a
u
n
iq
u
e
f
ea
tu
r
e
ex
tr
ac
tio
n
s
ch
em
e
th
at
ad
ap
tiv
ely
p
er
f
o
r
m
s
tex
t
lo
ca
lizatio
n
ca
p
ab
le
o
f
f
in
e
-
tu
n
in
g
its
elf
to
d
if
f
e
r
en
t
t
y
p
es
o
f
b
ac
k
g
r
o
u
n
d
o
f
v
id
eo
.
T
h
e
is
s
u
e
p
er
tain
i
n
g
t
o
p
o
s
itio
n
al
in
f
o
r
m
atio
n
is
ad
d
r
ess
ed
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
b
y
co
n
s
id
er
in
g
a
b
o
u
n
d
in
g
b
o
x
f
o
r
m
u
lated
d
y
n
am
ica
lly
,
f
ac
ilit
atin
g
th
e
m
o
d
el
with
en
r
ich
ed
s
p
atial
co
n
tex
t
ap
ar
t
f
r
o
m
p
e
r
f
o
r
m
in
g
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
.
T
h
e
p
r
o
b
lem
o
f
ex
ten
s
iv
e
co
m
p
u
tatio
n
al
co
s
t
an
d
in
cr
ea
s
ed
in
f
er
e
n
ce
tim
e
is
co
n
tr
o
lled
b
y
th
e
p
r
o
p
o
s
ed
s
y
s
tem
b
y
ad
o
p
tin
g
an
o
p
tim
ized
a
r
ch
itectu
r
e
o
f
a
n
eu
r
al
n
etwo
r
k
,
m
a
k
in
g
it su
it
ab
le
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
T
h
e
r
esear
ch
aim
o
f
th
e
p
r
o
p
o
s
ed
s
tu
d
y
m
o
d
el
is
to
war
d
s
in
tr
o
d
u
cin
g
an
e
f
f
icien
t
tex
t
l
o
ca
lizatio
n
f
r
am
ewo
r
k
g
u
id
e
d
b
y
an
in
te
llig
en
t
v
er
if
ier
f
o
r
co
m
p
lex
f
o
r
m
s
o
f
v
id
e
o
co
n
ten
t.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
ca
lled
as
E
-
T
E
XT
L
OC
,
wh
ic
h
will
m
ea
n
a
n
ef
f
icien
t
tex
t
lo
ca
lizatio
n
an
d
e
x
tr
ac
tio
n
f
o
r
co
m
p
r
eh
e
n
s
iv
e
p
o
s
itio
n
al
ac
cu
r
ac
y
in
co
m
p
lex
v
id
eo
s
.
T
h
e
v
alu
e
-
a
d
d
ed
co
n
tr
ib
u
tio
n
o
f
E
-
T
E
XT
L
O
C
ar
e
as
f
o
llo
ws
:
i
)
E
-
T
E
XT
L
OC
p
r
esen
ts
a
f
r
am
ewo
r
k
f
o
r
ac
cu
r
ate
lo
ca
lizatio
n
o
f
tex
t
b
y
ad
o
p
tin
g
co
n
to
u
r
-
b
ased
r
eg
io
n
p
r
o
p
o
s
al
tech
n
iq
u
e
f
o
r
h
ig
h
li
g
h
tin
g
a
r
ea
s
en
r
ich
e
d
with
te
x
t
with
in
cr
ea
s
ed
s
p
atial
s
ali
en
cy
,
ii)
th
e
m
o
d
el
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
E
-
TEX
TL
OC
:
efficien
t te
xt
lo
c
a
liz
a
tio
n
a
n
d
ex
tr
a
ctio
n
in
r
ea
l
-
w
o
r
ld
vid
eo
… (
Da
ya
n
a
n
d
a
K
o
d
a
la
Ja
ya
r
a
m
)
3539
im
p
lem
en
ts
d
ee
p
co
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
(
DC
NN)
f
o
r
v
er
if
icatio
n
o
f
ca
n
d
id
ate
r
eg
io
n
r
ap
id
ly
o
p
tim
izin
g
th
e
ac
cu
r
ac
y
p
er
f
o
r
m
an
ce
with
m
in
im
al
o
v
e
r
h
ea
d
,
iii)
t
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
ta
r
g
ets
to
r
ed
u
ce
th
e
in
f
e
r
en
ce
tim
e
f
o
r
ea
c
h
v
id
eo
f
r
am
e
a
d
o
p
tin
g
o
p
tim
ized
f
ea
tu
r
e
s
elec
tio
n
f
o
r
f
ac
ilit
atin
g
E
-
T
E
X
T
L
OC
in
r
ea
l
-
tim
e
en
v
ir
o
n
m
en
t,
a
n
d
iv
)
th
e
m
o
d
el
im
p
lem
en
ts
a
n
o
v
el
an
d
y
et
s
im
p
lifie
d
f
ee
d
b
ac
k
m
ec
h
an
is
m
u
s
in
g
v
er
if
ier
f
o
r
ass
es
s
in
g
o
u
tco
m
e
o
f
OC
R
f
o
llo
wed
b
y
in
co
r
p
o
r
atin
g
m
o
d
el
lear
n
in
g
f
o
r
o
p
tim
ized
ac
c
u
r
a
cy
p
er
f
o
r
m
an
ce
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
d
if
f
er
e
n
t f
r
o
m
c
o
n
v
e
n
tio
n
al
lo
ca
lizat
io
n
m
eth
o
d
s
s
u
ch
as E
AST
an
d
C
R
AFT,
wh
ich
ess
en
tially
d
ep
en
d
u
p
o
n
d
en
s
e
p
i
x
el
-
lev
el
f
o
r
ec
asti
n
g
an
d
d
em
an
d
p
o
s
t
-
p
r
o
ce
s
s
in
g
th
at
is
ac
tu
ally
co
m
p
u
tatio
n
ally
in
ten
s
iv
e.
O
n
th
e
co
n
tr
a
r
y
,
a
lig
h
tweig
h
t
co
n
to
u
r
-
b
ased
r
eg
io
n
p
r
o
p
o
s
al
m
eth
o
d
h
as
b
ee
n
p
r
esen
ted
b
y
E
-
T
E
XT
L
OC
f
o
r
m
in
im
izin
g
th
e
s
ea
r
ch
s
p
ac
e
b
ef
o
r
e
v
alid
atin
g
it
b
y
d
ee
p
l
ea
r
n
in
g
.
Similar
ly
,
tr
an
s
f
o
r
m
er
-
b
ased
m
eth
o
d
s
ar
e
g
o
o
d
i
n
m
o
d
ellin
g
co
n
tex
t
u
al
d
ep
en
d
e
n
cies
an
d
y
et
th
ey
n
ee
d
an
ex
te
n
s
iv
ely
lar
g
e
lab
elled
d
ataset
an
d
n
ee
d
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
.
On
th
e
co
n
tr
ar
y
,
E
-
T
E
XT
L
OC
f
o
cu
s
es
o
n
co
m
p
u
tatio
n
al
e
f
f
icien
cy
u
s
in
g
v
er
if
ier
m
o
d
ellin
g
o
f
C
NN
an
d
f
ee
d
b
ac
k
m
eth
o
d
s
u
s
in
g
OC
R
.
I
n
s
h
o
r
t,
a
s
elf
-
s
u
p
er
v
is
ed
v
e
r
if
ied
m
o
d
ellin
g
h
as
b
ee
n
p
r
esen
ted
b
y
E
-
T
E
XT
L
OC
,
wh
ich
r
e
f
in
es
th
e
lo
ca
lizatio
n
r
esu
lt
u
s
in
g
th
e
co
n
f
i
d
en
ce
s
co
r
e
o
f
an
OC
R
.
T
h
e
n
ex
t
s
ec
tio
n
p
r
esen
ts
a
d
is
cu
s
s
io
n
o
f
th
e
ad
o
p
ted
m
eth
o
d
o
lo
g
y
t
o
ac
co
m
p
lis
h
th
e
s
tu
d
y
o
b
jectiv
es.
2.
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
o
f
E
-
T
E
XT
L
OC
ad
o
p
ts
a
s
im
p
lifie
d
m
ath
em
atica
l
m
o
d
ellin
g
tar
g
etin
g
tex
t
lo
ca
lizatio
n
,
wh
ich
f
u
r
th
e
r
co
n
s
is
ts
o
f
b
o
th
d
etec
tio
n
as
we
ll
as
r
ec
o
g
n
itio
n
o
f
tex
t
r
esid
i
n
g
with
in
t
h
e
v
id
e
o
f
r
am
es.
Fig
u
r
e
1
h
ig
h
lig
h
ts
th
e
ad
o
p
ted
s
tu
d
y
ar
ch
itectu
r
e
co
n
s
is
tin
g
o
f
v
ar
i
o
u
s
o
p
e
r
atio
n
m
o
d
u
les
to
ac
co
m
p
lis
h
th
e
s
tu
d
y
g
o
al.
T
h
e
p
r
im
e
m
o
d
u
le
in
v
o
l
v
ed
in
ar
ch
itectu
r
e
is
r
esp
o
n
s
ib
le
f
o
r
th
e
g
en
er
atio
n
o
f
r
eg
io
n
as
p
ar
t
o
f
ac
co
m
p
lis
h
in
g
th
e
g
o
al
o
f
lo
ca
lizatio
n
wh
il
e
th
e
D
C
NN
m
o
d
u
le
a
n
d
th
e
OC
R
m
o
d
u
le
ca
r
r
y
o
u
t
th
e
task
o
f
d
etec
tio
n
a
n
d
r
ec
o
g
n
itio
n
.
T
h
e
o
p
tim
al
s
tr
u
ctu
r
e
an
d
in
teg
r
atio
n
o
f
ea
ch
m
o
d
u
le
ar
e
m
ath
em
atica
lly
f
o
r
m
u
lated
.
Fig
u
r
e
1
.
Ar
c
h
itectu
r
e
o
f
E
-
T
E
XT
L
OC
2
.
1
.
Co
rner
re
s
po
ns
e
f
ea
t
ur
e
m
a
p r
eg
io
n pro
po
s
a
l
T
h
is
m
o
d
u
le
is
r
esp
o
n
s
ib
le
f
o
r
g
en
er
atin
g
th
e
ca
n
d
id
ate
r
eg
io
n
with
a
h
ig
h
p
o
s
s
ib
ilit
y
o
f
th
e
p
r
esen
ce
o
f
te
x
tu
al
co
n
te
n
t
in
th
e
v
id
eo
f
r
am
es.
T
h
e
n
o
tio
n
o
f
co
r
n
er
r
esp
o
n
s
e
f
ea
tu
r
e
m
a
p
(
C
R
FM
)
is
b
ase
d
o
n
th
e
co
m
p
u
tatio
n
o
f
lo
ca
l
v
ar
iatio
n
ass
o
ciate
d
with
in
ten
s
ity
,
f
o
llo
wed
b
y
th
e
d
etec
tio
n
o
f
s
tr
u
ctu
r
es
with
co
n
to
u
r
s
.
C
o
n
s
id
er
ad
o
p
tin
g
a
g
r
ad
ien
t
ed
g
e
d
etec
to
r
as
(
I
x
,
I
y
)
=(
∂I
/∂x
,
∂I
/∂y
)
wh
er
e
I
(
x
,
y
)
r
ep
r
esen
ts
g
r
ay
s
ca
le
in
ten
s
ity
at
lo
ca
tio
n
(
x
,
y
)
in
th
e
v
i
d
eo
f
r
am
e.
T
h
e
m
o
d
el
d
ef
in
es
cu
r
v
atu
r
e
-
b
ased
in
ter
est
p
o
in
t
d
etec
to
r
co
n
t
o
u
r
r
esp
o
n
s
e
R
(
x
,
y
)
as (
1
)
.
(
,
)
=
de
t
(
[
2
2
]
∗
)
−
.
(
[
2
2
]
)
2
(
1
)
I
n
th
e
(
1
)
,
th
e
s
tr
u
ctu
r
e
in
s
id
e
th
e
d
eter
m
in
an
t
d
et
f
u
n
ctio
n
i
s
a
s
tr
u
ctu
r
e
ten
s
o
r
ass
o
ciate
d
with
ea
ch
p
ix
el,
wh
ile
G
σ
r
ep
r
esen
ts
a
Gau
s
s
ian
k
er
n
el
th
at
is
u
s
ed
f
o
r
s
m
o
o
t
h
en
in
g
(
alo
n
g
with
s
tan
d
ar
d
d
ev
iatio
n
)
,
an
d
th
e
v
ar
iab
le
k
is
an
em
p
ir
ical
co
n
s
tan
t
with
v
alu
e
[
0
.
0
4
,
0
.
0
6
]
.
A
clo
s
er
lo
o
k
in
to
th
e
f
ir
s
t
co
m
p
o
n
en
t
o
f
(
1
)
will
b
e
r
ep
r
esen
ted
in
t
h
e
f
o
r
m
o
f
{
2
2
-
(
I
x
I
y
)
2
}
w
h
ile
th
e
s
ec
o
n
d
c
o
m
p
o
n
en
t o
f
(
1
)
will
r
e
p
r
esen
t
(
2
+
2
)
.
Fu
r
th
er
,
tr
a
ce
(
)
r
ep
r
esen
ts
th
e
s
u
m
m
atio
n
o
f
all
d
iag
o
n
al
ele
m
en
ts
p
r
esen
t
with
in
th
e
s
tr
u
c
tu
r
e
ten
s
o
r
.
C
R
FM
also
d
ef
in
es
C
(
x
,
y
)
as a
b
in
a
r
y
co
n
to
u
r
m
a
p
th
at
is
r
ep
r
esen
ted
as (
2
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
3
7
-
3
5
4
5
3540
(
,
)
=
{
1
(
,
)
>
0
ℎ
(
2
)
Acc
o
r
d
in
g
to
th
e
(
2
)
,
th
e
v
al
u
atio
n
o
f
th
e
b
in
ar
y
c
o
n
to
u
r
m
ap
c
o
m
p
letely
d
ep
e
n
d
s
u
p
o
n
co
n
to
u
r
r
esp
o
n
s
e
R
(
x
,
y
)
,
wh
er
e
T
c
r
e
p
r
esen
ts
its
cu
t
-
o
f
f
v
alu
e
f
o
r
d
eter
m
in
in
g
th
e
p
o
ten
tial
c
o
n
to
u
r
r
esp
o
n
s
es.
T
h
e
m
o
d
el
th
en
ap
p
lies
m
o
r
p
h
o
lo
g
ical
o
p
er
atio
n
s
s
u
b
jecte
d
to
a
b
in
ar
y
co
n
to
u
r
m
ap
C
(
x
,
y
)
th
at
ca
n
n
o
w
b
e
r
ep
r
esen
ted
as
’
(
,
)
=
(
(
(
,
)
)
)
.
T
h
e
s
y
s
te
m
th
en
u
s
es
lab
ellin
g
o
f
all
co
n
n
ec
ted
co
m
p
o
n
en
ts
to
ac
q
u
ir
e
th
e
co
n
to
u
r
s
.
(
)
−
1
f
r
o
m
co
n
n
ec
ted
r
eg
io
n
s
o
f
C
’(
x
,
y
)
.
Fin
ally
,
a
s
p
ec
if
ic
r
eg
io
n
with
lo
ca
tio
n
in
f
o
r
m
atio
n
with
a
v
alid
asp
ec
t r
atio
is
r
etain
ed
b
y
th
e
f
ilter
in
g
s
tep
as (
3
)
.
∈
=
1
∈
[
1
,
2
]
,
∈
[
,
]
(
3
)
I
n
th
e
(
3
)
,
th
e
v
a
r
iab
le
B
i
r
ep
r
esen
ts
a
b
o
u
n
d
in
g
b
o
x
s
p
ec
if
ic
to
co
n
t
o
u
r
r
eg
io
n
R
i,
wh
ile
S
r
ep
r
esen
ts
a
s
et
o
f
all
ca
n
d
id
ate
r
eg
io
n
s
o
f
tex
t
i.e
.
,
=
{
}
.
T
h
e
v
ar
iab
le
D
i
r
ep
r
esen
ts
(
w
i
/
h
i
)
,
wh
er
e
w
i
an
d
h
i
r
ep
r
esen
t
th
e
wid
th
an
d
h
ei
g
h
t
o
f
th
e
b
o
u
n
d
in
g
b
o
x
an
d
ar
e
a
is
r
ep
r
esen
ted
as
(
w
i
.
h
i
)
.
T
h
e
v
alid
r
an
g
e
o
f
ar
ea
is
f
u
r
th
er
d
esig
n
ated
as
(
,
)
.
T
h
e
n
o
v
elty
o
f
th
is
p
a
r
t
o
f
th
e
im
p
lem
en
tatio
n
is
ass
o
ciate
d
with
th
e
in
tr
o
d
u
ctio
n
o
f
l
o
w
-
lev
el
f
ea
tu
r
e
ex
tr
ac
ti
o
n
alo
n
g
with
th
e
m
ec
h
an
is
m
o
f
r
eg
io
n
p
r
o
p
o
s
al.
T
h
is
m
eth
o
d
n
o
t
o
n
ly
c
o
n
tr
ib
u
tes
to
war
d
s
h
ig
h
ly
g
r
a
n
u
lar
an
d
lig
h
tw
eig
h
t
tex
t
lo
ca
lizatio
n
b
u
t
a
ls
o
to
war
d
s
f
aster
id
en
tific
atio
n
o
f
p
o
ten
tial c
an
d
id
ate
r
eg
io
n
s
.
2
.
2
.
Dee
p c
o
nv
o
lutio
n neura
l net
wo
rk
v
er
if
ier
T
h
e
k
e
y
p
u
r
p
o
s
e
o
f
th
is
m
o
d
u
le
is
to
war
d
s
e
n
r
ich
in
g
th
e
o
u
tco
m
e
o
f
th
e
p
r
io
r
C
R
FM
m
o
d
u
le
b
y
r
em
o
v
in
g
all
th
e
f
alse
p
o
s
itiv
es
wh
er
e
r
eg
i
o
n
s
m
ay
b
e
d
et
ec
ted
with
en
r
ic
h
ed
c
o
n
to
u
r
s
,
b
u
t
th
e
y
lack
th
e
p
o
s
s
ess
io
n
o
f
an
y
tex
tu
al
co
n
t
en
t.
As
th
e
C
R
FM
m
o
d
u
le
ca
n
n
o
t
af
f
o
r
d
to
m
is
s
an
y
tex
t,
h
e
n
ce
DC
C
N
v
er
if
ier
en
s
u
r
es
th
at
th
e
tex
t
r
eg
io
n
with
a
h
ig
h
er
co
n
f
id
e
n
ce
r
ate
s
h
o
u
ld
b
e
p
ass
ed
o
n
f
o
r
its
co
n
s
ec
u
tiv
e
s
tag
es
o
f
r
ec
o
g
n
itio
n
.
E
-
T
E
XT
L
OC
u
s
es
th
e
D
C
C
N
v
er
if
ier
in
o
r
d
er
to
m
in
im
ize
th
e
p
o
s
s
ib
ilit
i
es
o
f
o
u
tlier
s
f
r
o
m
C
R
FM,
wh
ich
is
em
p
ir
ically
r
ep
r
esen
ted
as
(
4
)
.
=
×
×
3
[
0
,
1
]
(
4
)
I
n
th
e
(
4
)
,
r
ep
r
esen
ts
a
b
in
ar
y
class
if
ier
th
at
tak
es
th
e
v
alu
e
o
f
eith
er
1
o
r
0
b
ased
o
n
wh
et
h
er
th
e
r
eg
io
n
x
is
f
o
u
n
d
to
h
a
v
e
t
ex
tu
al
co
n
ten
ts
o
r
o
th
er
wis
e,
r
esp
ec
tiv
ely
,
w
h
ile
th
e
s
u
f
f
ix
θ
is
a
n
etwo
r
k
p
ar
am
eter
.
T
h
e
m
o
d
el
c
o
n
s
id
er
s
a
lig
h
tweig
h
t
m
o
d
el
t
h
at
co
n
s
is
ts
o
f
co
n
v
o
lu
tio
n
lay
er
s
f
o
llo
wed
b
y
b
atch
n
o
r
m
aliza
tio
n
,
ac
tiv
atio
n
o
f
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
,
an
d
a
lay
er
with
a
f
u
lly
co
n
n
ec
ted
n
etwo
r
k
.
A
d
ataset
=
{
(
,
)
}
is
u
s
ed
to
tr
ain
th
e
n
etwo
r
k
,
wh
e
r
e
an
d
r
ep
r
esen
t
a
cr
o
p
p
e
d
p
atc
h
o
f
a
n
im
ag
e
an
d
lab
el
(
0
,
1
)
,
r
esp
ec
tiv
ely
.
T
h
e
s
y
s
tem
co
m
p
u
tes b
i
n
ar
y
c
r
o
s
s
-
en
tr
o
p
y
lo
s
s
as
(
5
)
.
(
)
=
−
1
∑
[
1
+
2
]
=
1
(
5
)
I
n
th
e
(
5
)
,
it
ca
n
b
e
n
o
ted
th
at
co
m
p
u
tatio
n
o
f
lo
s
s
f
u
n
cti
o
n
(
)
d
ep
en
d
s
u
p
o
n
two
en
titi
es,
i.e
.
,
1
an
d
2
,
wh
ich
r
e
p
r
esen
t
.
(
)
an
d
(
1
−
)
.
(
1
−
(
)
)
r
esp
ec
tiv
ely
.
Fin
ally
,
th
e
in
f
er
en
ce
is
ca
r
r
ied
o
u
t
with
̂
th
at
u
n
d
e
r
tak
es
th
e
v
al
u
e
o
f
eith
er
1
o
r
0
f
o
r
(
)
>
o
r
o
t
h
er
wis
e,
r
esp
ec
tiv
ely
;
wh
er
e
th
e
attr
i
b
u
te
T
p
r
e
p
r
ese
n
ts
a
b
in
ar
y
c
u
t
-
o
f
f
o
f
class
if
icatio
n
co
n
f
id
en
ce
.
I
t
s
h
o
u
ld
b
e
n
o
ted
th
at
th
e
o
n
l
y
r
eg
io
n
with
a
d
etec
ted
b
o
u
n
d
i
n
g
b
o
x
as a
p
ar
t o
f
S
h
as with
v
alu
e
o
f
̂
as u
n
ity
as f
o
r
war
d
ed
to
th
e
n
ex
t stag
e
o
f
o
p
e
r
atio
n
.
2
.
3
.
Rec
o
g
nitio
n
m
o
du
le
wit
h
o
ptic
a
l c
ha
ra
ct
er
re
co
g
nitio
n
T
h
e
p
u
r
p
o
s
e
o
f
th
is
m
o
d
u
le
is
to
tr
an
s
f
o
r
m
th
e
c
o
n
f
ir
m
e
d
r
e
g
io
n
o
f
te
x
t
in
to
a
s
p
ec
if
ic
f
o
r
m
o
f
tex
t
th
at
is
m
ac
h
in
e
-
r
ea
d
ab
le.
R
esizin
g
is
ca
r
r
ied
o
u
t f
o
r
t
h
e
r
e
g
io
n
p
ass
ed
v
ia
th
e
DC
NN
f
ilter
an
d
th
e
n
s
u
b
jecte
d
to
a
p
r
etr
ain
e
d
OC
R
p
r
o
ce
s
s
in
g
u
n
it.
I
n
o
r
d
er
to
ca
r
r
y
o
u
t
tr
an
s
cr
ip
tio
n
o
f
tex
tu
al
c
o
n
te
n
ts
,
th
e
r
eg
io
n
s
o
f
r
etain
ed
im
ag
e
ar
e
f
o
r
war
d
ed
to
a
p
r
o
ce
s
s
in
g
u
n
it
wh
ich
ex
ec
u
tes
r
ec
o
g
n
itio
n
o
p
e
r
atio
n
a
s
Ф
×
×
3
→
Σ
∗
,
wh
er
e
th
e
n
o
tatio
n
Σ
*
r
ep
r
esen
ts
a
s
tr
in
g
as
s
o
ciate
d
with
an
alp
h
ab
et
Σ
.
C
o
n
s
id
er
in
g
∈
b
e
an
in
p
u
t
r
eg
io
n
,
t
h
e
r
ec
o
g
n
is
ed
tex
t t
is
r
ep
r
esen
ted
as (
6
)
.
=
Ф
(
x
)
∈
Σ
∗
(
6
)
I
n
(
6
)
in
f
er
s
th
at
wh
e
n
t
≠
∅
th
an
th
e
id
en
tifie
d
r
eg
io
n
is
co
n
f
ir
m
ed
to
p
o
s
s
ess
tex
tu
al
co
n
ten
t.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
u
s
es
T
ess
er
ac
t
OC
T
f
o
r
im
p
lem
e
n
tin
g
Ф
.
E
-
T
E
XT
L
OC
also
im
p
le
m
en
ts
non
-
m
ax
im
u
m
s
u
p
p
r
ess
io
n
(
NM
S).
I
t
is
d
o
n
e
to
co
n
t
r
o
l
t
h
e
p
o
s
s
ib
le
o
cc
u
r
r
en
ce
s
o
f
r
ed
u
n
d
a
n
t
b
o
u
n
d
in
g
b
o
x
es.
T
h
e
co
m
p
u
ter
v
is
io
n
m
etr
ic
(
C
VM
)
u
s
ed
f
o
r
th
is
p
u
r
p
o
s
e
is
as
in
(
7
)
.
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
E
-
TEX
TL
OC
:
efficien
t te
xt
lo
c
a
liz
a
tio
n
a
n
d
ex
tr
a
ctio
n
in
r
ea
l
-
w
o
r
ld
vid
eo
… (
Da
ya
n
a
n
d
a
K
o
d
a
la
Ja
ya
r
a
m
)
3541
(
,
)
=
|
∩
|
|
∪
|
(
7
)
I
n
th
e
(
7
)
,
th
e
co
m
p
u
tatio
n
o
f
C
VM
i
s
p
er
f
o
r
m
ed
b
y
im
p
le
m
en
tin
g
in
ter
s
ec
tio
n
o
v
e
r
u
n
i
o
n
,
wh
ich
co
m
p
u
tes
th
e
d
e
g
r
ee
o
f
o
v
e
r
la
p
p
in
g
o
f
p
r
ed
icted
an
d
g
r
o
u
n
d
tr
u
th
s
co
r
es.
T
h
e
b
o
u
n
d
in
g
b
o
x
with
C
VM
m
o
r
e
th
an
cu
t
-
o
f
f
τ
is
s
u
b
jecte
d
to
s
u
p
p
r
ess
io
n
,
wh
ile
th
e
b
o
u
n
d
i
n
g
b
o
x
with
m
ax
im
u
m
a
r
ea
o
r
co
n
f
id
en
ce
o
f
OC
R
is
r
etain
ed
.
T
h
e
v
al
u
e
o
f
cu
t
-
o
f
f
τ
ca
n
b
e
r
etain
ed
as 0
.
2
-
0
.
6
.
3.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
T
h
e
ass
ess
m
en
t
o
f
th
e
E
-
T
E
XT
L
OC
is
p
er
f
o
r
m
ed
o
n
a
h
ig
h
ly
co
n
tr
o
lled
s
im
u
latio
n
e
n
v
ir
o
n
m
en
t
wh
er
e
th
e
s
cr
ip
tin
g
is
ca
r
r
ied
o
u
t
in
Py
th
o
n
.
T
h
e
m
o
d
el
is
ass
es
s
ed
with
a
d
atase
t
co
n
s
tr
u
cted
u
s
in
g
Yo
u
T
u
b
e
v
id
eo
s
.
T
h
e
d
ataset
co
n
s
is
ts
o
f
an
.
m
p
4
v
id
eo
f
ile
with
2
5
f
r
am
es
p
er
s
ec
o
n
d
with
3
5
4
k
b
p
s
as
d
ata
r
ate.
T
h
e
v
id
eo
p
r
o
ce
s
s
in
g
,
e
x
tr
ac
tio
n
o
f
f
ea
tu
r
es,
an
d
f
o
r
m
atio
n
o
f
a
b
o
u
n
d
in
g
b
o
x
,
is
ca
r
r
ied
o
u
t
u
s
in
g
Op
en
C
V,
wh
ile
d
ee
p
lear
n
in
g
m
o
d
u
les
ar
e
im
p
lem
en
ted
u
s
in
g
Py
T
o
r
ch
,
wh
er
e
Mo
b
ileNetV2
is
u
s
ed
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
th
e
DC
C
N
clas
s
if
ier
is
u
s
ed
f
o
r
tr
ain
in
g
.
OC
R
is
im
p
lem
en
ted
u
s
in
g
Py
tess
er
ac
t
f
o
r
ac
q
u
ir
i
n
g
r
ea
d
ab
le
tex
t,
wh
ile
m
ath
em
a
tical
o
p
er
atio
n
is
p
er
f
o
r
m
e
d
b
y
Nu
m
Py
.
T
h
e
im
p
lem
e
n
tat
io
n
o
f
t
h
e
b
aselin
e
m
ac
h
in
e
lear
n
i
n
g
m
o
d
el
is
p
e
r
f
o
r
m
e
d
u
s
in
g
s
cik
it
lear
n
in
g
.
T
h
e
h
a
r
d
war
e
c
o
n
f
ig
u
r
atio
n
s
ar
e
as
f
o
llo
ws:
th
e
s
y
s
tem
is
im
p
lem
en
ted
o
n
a
n
o
r
m
al
I
n
tel
i5
p
r
o
ce
s
s
o
r
with
3
2
GB
R
AM
an
d
NVI
DI
A
GE
FOR
C
E
GT
X
in
s
talled
o
n
a
W
in
d
o
ws
1
1
m
ac
h
in
e.
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
is
co
m
p
a
r
ed
with
ex
is
tin
g
a
p
p
r
o
ac
h
es,
as
th
er
e
ar
e
v
ar
io
u
s
r
ep
o
r
ts
s
tatin
g
th
e
u
s
ag
e
o
f
m
ac
h
in
e
lear
n
i
n
g
an
d
OC
R
-
b
ased
m
eth
o
d
o
lo
g
ies,
wh
ich
ar
e
n
ea
r
ly
eq
u
iv
alen
t to
E
-
T
E
XT
L
OC
[
2
7
]
,
[
2
8
]
.
3
.
1
.
Acc
o
m
pli
s
hed r
esu
lt
T
ab
le
1
h
ig
h
lig
h
ts
th
e
ac
c
o
m
p
lis
h
ed
s
tu
d
y
o
u
tc
o
m
es
u
s
in
g
v
ar
io
u
s
lib
r
a
r
ies
an
d
s
o
f
twar
e
to
o
ls
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
is
co
m
p
a
r
ed
with
th
r
ee
b
aselin
e
m
o
d
els
o
f
T
in
y
C
NN,
HOG
in
teg
r
ated
wi
th
SVM,
an
d
HOG
in
teg
r
ated
with
r
a
n
d
o
m
f
o
r
est
.
I
t
is
to
b
e
n
o
ted
th
at
T
in
y
C
NN
is
a
s
lig
h
tly
lo
wer
v
er
s
io
n
o
f
th
e
ad
o
p
ted
d
ee
p
lear
n
in
g
m
o
d
el,
wh
ile
t
h
e
ex
t
r
ac
tio
n
o
f
s
h
ap
e
-
o
r
ien
te
d
f
ea
t
u
r
es
is
ca
r
r
ied
o
u
t
b
y
HOG
in
te
g
r
ated
SVM
m
o
d
el
th
at
ca
n
d
if
f
er
en
tiate
b
o
th
tex
tu
al
an
d
n
o
n
-
te
x
tu
al
r
eg
io
n
s
.
Fu
r
th
er
,
a
s
im
ilar
f
ea
tu
r
e
is
a
ls
o
u
s
ed
b
y
HOG
wh
en
in
teg
r
ated
with
r
a
n
d
o
m
f
o
r
est;
h
o
we
v
er
,
th
e
r
eg
io
n
s
a
r
e
class
if
ied
v
ia
a
c
o
n
s
en
s
u
s
m
eth
o
d
p
ar
ticip
ated
in
b
y
m
u
ltip
le
d
ec
is
io
n
tr
ee
s
.
T
h
e
ac
co
m
p
lis
h
ed
o
u
tco
m
e
is
an
aly
ze
d
u
s
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e
,
an
d
av
er
ag
e
in
f
er
e
n
ce
tim
e
p
er
f
r
a
m
e.
T
ab
le
1
.
Acc
o
m
p
lis
h
ed
b
e
n
ch
m
ar
k
ed
o
u
tco
m
e
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
A
v
g
.
i
n
f
e
r
e
n
c
e
t
i
m
e
p
e
r
f
r
a
m
e
(
ms)
E
-
TEX
TLO
C
9
8
.
2
9
5
.
4
9
4
.
1
9
4
.
7
2
9
.
6
Ti
n
y
C
N
N
9
2
.
3
8
7
.
6
8
4
.
5
8
6
.
0
5
2
.
3
H
O
G
+
S
V
M
8
9
.
7
8
1
.
4
7
7
.
8
7
9
.
5
3
8
.
5
H
O
G
+
r
a
n
d
o
m f
o
r
e
st
8
8
.
1
8
2
.
1
7
3
.
2
7
7
.
4
4
0
.
1
T
h
e
s
tu
d
y
o
u
tco
m
e
s
h
o
ws
th
at
E
-
T
E
XT
L
OC
ac
co
m
p
lis
h
es
9
8
.
2
%
ac
cu
r
ac
y
,
wh
ich
in
f
er
s
its
ap
p
r
o
p
r
iaten
ess
with
r
esp
ec
t
to
all
d
ec
is
io
n
s
to
war
d
s
cla
s
s
if
icatio
n
o
n
all
v
id
eo
f
r
am
es.
T
h
e
p
r
ec
is
io
n
an
d
r
ec
all
r
ate
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
e
f
o
u
n
d
to
b
e
9
5
.
4
%
an
d
9
4
.
1
%
r
esp
ec
tiv
ely
,
e
x
h
i
b
itin
g
a
s
ig
n
if
ican
t
r
ed
u
ctio
n
o
f
f
alse
p
o
s
itiv
es
as
well
as
f
alse
n
eg
ativ
es.
Fu
r
t
h
er
,
th
e
p
r
o
p
o
s
ed
s
y
s
tem
is
n
o
ted
to
ac
c
o
m
p
lis
h
9
4
.
7
%
o
f
F1
-
s
co
r
e
u
n
d
e
r
d
i
f
f
er
en
t
c
o
n
d
itio
n
s
o
f
v
id
e
o
f
o
r
lo
ca
lizin
g
th
e
p
o
s
itio
n
o
f
tex
t.
Fin
ally
,
th
e
in
f
er
en
ce
tim
e
f
o
r
E
-
T
E
XT
L
OC
is
n
o
ted
to
b
e
ju
s
t
2
9
.
6
m
s
,
p
r
o
v
in
g
it
as
th
e
f
astes
t
o
p
er
atio
n
s
u
itab
le
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
3
.
2
.
Dis
cus
s
io
n
B
ased
o
n
th
e
ac
co
m
p
lis
h
ed
n
u
m
er
ical
o
u
tco
m
e
,
th
e
a
n
aly
s
is
h
as
b
ee
n
ex
te
n
s
iv
ely
ca
r
r
ied
o
u
t
to
f
in
d
th
at
th
e
r
a
n
g
e
o
f
th
e
ac
c
u
r
ac
y
m
etr
ic
r
esid
es
b
etwe
en
9
7
-
9
9
%
wh
ile
th
e
en
h
an
ce
m
e
n
t
in
t
h
e
av
er
a
g
e
r
ate
o
f
p
r
ec
is
io
n
an
d
r
ec
all
is
b
etwe
en
1
5
a
n
d
1
2
%
r
esp
ec
tiv
ely
.
S
im
ilar
ly
,
th
e
F1
-
s
co
r
e
is
im
p
r
o
v
ed
to
1
3
-
1
8
%
in
co
n
tr
ast
to
b
aselin
e
m
o
d
els,
wh
ile
p
o
ten
tial
im
p
r
o
v
em
en
t
in
av
er
a
g
e
in
f
er
en
ce
tim
e
is
n
o
ted
to
b
e
4
0
-
5
0
%.
T
h
er
e
ar
e
v
ar
io
u
s
r
atio
n
ales
to
b
e
attr
ib
u
ted
to
th
is
ac
co
m
p
lis
h
ed
o
u
tco
m
e
(
Fig
u
r
e
2
)
.
E
-
T
E
XT
L
OC
u
s
es
p
s
eu
d
o
-
lab
ellin
g
as
well
a
s
in
tellig
en
t
s
am
p
lin
g
,
wh
ich
en
s
u
r
es
th
e
s
elec
tio
n
o
f
o
n
ly
p
o
ten
tial
f
r
am
es
eith
er
d
u
r
in
g
tr
ain
in
g
o
r
d
u
r
in
g
in
f
er
en
ce
g
en
er
atio
n
.
T
h
ese
o
p
er
atio
n
s
lead
to
a
s
ig
n
if
i
ca
n
t
r
ed
u
ctio
n
in
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
as
well
as
n
o
is
e.
E
-
T
E
XT
L
OC
also
in
tr
o
d
u
ce
s
an
ef
f
icien
t
t
ec
h
n
iq
u
e
o
f
r
e
g
io
n
p
r
o
p
o
s
al,
alo
n
g
with
th
e
im
p
lem
en
tatio
n
o
f
f
ea
tu
r
e
e
x
tr
ac
tio
n
ca
r
r
ied
o
u
t
b
y
M
o
b
ileNetV2
,
with
o
u
t
s
ac
r
if
icin
g
an
y
r
ich
n
ess
in
f
ea
t
u
r
es.
Ap
ar
t f
r
o
m
th
is
,
it
is
also
n
o
ted
th
at
c
o
n
v
e
n
tio
n
al
m
o
d
e
ls
u
s
u
ally
tr
ea
t
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
3
7
-
3
5
4
5
3542
p
r
o
b
lem
s
o
f
lo
ca
lizatio
n
in
a
s
ep
ar
ate
way
,
in
co
n
tr
ast
to
p
r
o
b
lem
s
r
elate
d
to
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
.
E
-
T
E
XT
L
OC
in
teg
r
ates
a
s
m
ar
t
an
d
in
tellig
en
t
v
er
if
ier
f
o
r
ass
ess
in
g
th
e
o
u
tco
m
e
o
f
OC
R
,
f
o
llo
wed
b
y
f
ee
d
in
g
it b
ac
k
to
tr
ain
i
n
g
.
T
h
i
s
o
p
er
atio
n
r
esu
lts
in
co
n
s
is
ten
t im
p
r
o
v
em
e
n
t a
lo
n
g
with
s
u
p
er
io
r
r
o
b
u
s
tn
ess
o
f
th
e
m
o
d
el.
T
h
e
d
ep
lo
y
m
en
t
o
f
th
e
co
n
f
i
d
en
ce
s
co
r
es
o
f
OC
R
eith
er
f
o
r
ac
ce
p
tin
g
o
r
r
ejec
tin
g
th
e
p
r
e
d
ictio
n
p
o
ten
tially
ass
is
ts
in
ad
d
r
ess
in
g
th
e
d
is
s
em
in
atio
n
o
f
er
r
o
r
s
wh
ile
in
f
er
en
ce
is
in
t
h
e
p
r
o
ce
s
s
o
f
g
en
er
atio
n
,
as
s
h
o
wn
in
Fig
u
r
e
2
(
a)
,
lea
d
in
g
to
h
ig
h
er
ac
cu
r
ac
y
.
T
h
e
c
o
n
v
e
n
tio
n
al
m
ec
h
a
n
is
m
ad
o
p
ted
i
n
eith
er
th
e
ad
o
p
ted
b
aselin
e
m
o
d
el
o
r
an
y
o
th
er
s
ar
e
also
p
r
o
v
en
n
o
t
to
p
o
s
s
ess
i
n
clu
s
io
n
o
f
co
n
tex
tu
al
lear
n
in
g
as
well
an
d
lack
s
an
y
f
ee
d
b
ac
k
p
r
o
ce
s
s
.
T
h
e
b
a
s
elin
e
m
o
d
els
ar
e
n
o
t
witn
ess
ca
p
ab
le
to
war
d
s
f
ilter
in
g
o
u
t
o
u
tlier
s
ef
f
ec
tiv
ely
,
in
co
n
tr
ast
to
E
-
T
E
XT
L
OC
,
a
s
th
ey
o
f
ten
co
n
s
id
er
ea
ch
r
e
g
io
n
o
r
f
r
am
e
u
n
iq
u
ely
,
r
esu
ltin
g
in
an
in
c
r
ea
s
ed
r
ate
o
f
e
r
r
o
r
s
as
well
as
s
u
b
-
o
p
tim
al
g
en
er
aliza
tio
n
p
er
f
o
r
m
an
ce
wh
en
s
u
b
jecte
d
to
co
m
p
lex
o
r
n
o
is
y
s
ce
n
es
o
f
v
id
e
o
.
B
etter
p
r
ec
is
io
n
an
d
r
ec
all
p
er
f
o
r
m
an
ce
(
Fig
u
r
e
2
(
b
)
)
a
n
d
th
ei
r
b
alan
ce
d
o
p
e
r
atio
n
n
o
ted
in
Fig
u
r
e
2
(
c)
s
h
o
wca
s
e
th
at
tr
ain
in
g
ca
r
r
ie
d
o
u
t
b
y
th
e
v
er
if
ie
r
f
ac
ilit
ates
E
-
T
E
XT
L
OC
f
o
r
co
n
tr
o
llin
g
v
ar
io
u
s
ty
p
es
o
f
ch
allen
g
in
g
co
n
d
itio
n
s
in
r
e
al
-
tim
e,
v
iz.
,
m
o
tio
n
b
lu
r
,
o
cc
lu
s
io
n
,
an
d
illu
m
in
at
io
n
.
B
ased
o
n
t
h
e
o
u
tco
m
e
ac
co
m
p
lis
h
ed
in
th
e
s
tu
d
y
,
it c
an
b
e
in
f
er
r
e
d
th
at
th
e
m
o
d
u
lar
d
esig
n
o
f
E
-
T
E
XT
L
OC
f
ac
ilit
ate
s
it
to
b
e
d
ep
lo
y
ed
o
v
er
an
y
r
eso
u
r
c
e
-
co
n
s
tr
ain
ed
ec
o
s
y
s
tem
ex
h
i
b
itin
g
in
cr
ea
s
in
g
s
ca
lab
ilit
y
(
Fig
u
r
e
2
(
d
)
)
.
Du
e
to
m
in
im
al
in
f
er
en
ce
tim
e,
E
-
T
E
XT
L
OC
en
s
u
r
es
h
ig
h
s
u
itab
ilit
y
to
war
d
s
ass
is
tiv
e
r
ea
d
in
g
d
ev
ices,
au
to
n
o
m
o
u
s
v
eh
icles,
an
d
s
u
r
v
eillan
ce
s
y
s
tem
s
.
T
h
er
e
ar
e
v
ar
io
u
s
p
u
b
licly
av
ailab
le
b
e
n
ch
m
ar
k
d
atasets
,
e.
g
.
,
m
u
lti
-
lin
g
u
al
s
ce
n
e
tex
t
(
ML
T
)
,
co
m
m
o
n
co
n
tex
t
-
tex
t
(
C
OC
O
-
T
ex
t)
,
an
d
I
n
ter
n
ati
o
n
al
C
o
n
f
er
en
ce
o
n
Do
cu
m
en
t
An
aly
s
is
an
d
R
ec
o
g
n
itio
n
(
I
C
DAR),
th
at
ar
e
f
r
eq
u
en
tly
a
d
o
p
te
d
to
war
d
s
ass
es
s
in
g
tex
t
lo
ca
lizatio
n
f
r
o
m
s
ce
n
es
(
s
tatic
im
ag
es).
Su
ch
d
atasets
ar
e
witn
ess
ed
n
o
t
to
s
u
f
f
icien
tly
ex
tr
ac
t
th
e
r
ea
l
-
tim
e
co
n
s
tr
ain
ts
,
f
r
am
e
-
lev
el
r
ed
u
n
d
an
cies,
an
d
tem
p
o
r
al
d
y
n
am
i
cs
co
n
n
ec
ted
with
th
e
lo
ca
lizatio
n
o
f
v
id
eo
-
b
ased
tex
t.
T
h
e
E
-
T
E
XT
L
OC
m
o
d
el
is
e
s
s
en
tially
cr
af
ted
f
o
r
co
n
tin
u
o
u
s
s
tr
ea
m
s
o
f
v
id
eo
in
f
lu
en
ce
d
b
y
m
u
ltip
le
attr
ib
u
tes,
e.
g
.
,
laten
c
y
,
o
cc
lu
s
io
n
,
m
o
tio
n
b
lu
r
,
a
n
d
f
r
am
e
s
a
m
p
lin
g
.
T
h
e
s
tu
d
y
u
s
es
a
cu
r
at
ed
Yo
u
T
u
b
e
v
id
e
o
d
ataset
f
o
r
ca
r
r
y
in
g
o
u
t
t
h
e
ev
alu
atio
n
t
h
at
r
ep
r
esen
ts
a
d
ep
lo
y
m
en
t
ev
e
n
t
in
th
e
r
ea
l
wo
r
ld
.
He
n
ce
,
ass
es
s
m
en
t
o
f
E
-
T
E
XT
L
OC
ad
o
p
tin
g
d
if
f
er
en
t
p
u
b
lic
b
e
n
ch
m
ar
k
s
c
o
u
ld
b
e
ca
r
r
ied
o
u
t
in
f
u
tu
r
e
lin
es
o
f
wo
r
k
f
o
r
ev
alu
atin
g
th
e
g
en
er
a
lizatio
n
ac
r
o
s
s
d
if
f
er
en
t
d
atasets
.
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
2
.
Acc
o
m
p
lis
h
ed
s
tu
d
y
o
u
tco
m
es (
a)
ac
cu
r
ac
y
,
(
b
)
p
r
ec
is
io
n
an
d
r
ec
all,
(
c
)
F1
-
s
co
r
e,
an
d
(
d
)
in
f
e
r
en
ce
tim
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
E
-
TEX
TL
OC
:
efficien
t te
xt
lo
c
a
liz
a
tio
n
a
n
d
ex
tr
a
ctio
n
in
r
ea
l
-
w
o
r
ld
vid
eo
… (
Da
ya
n
a
n
d
a
K
o
d
a
la
Ja
ya
r
a
m
)
3543
3
.
3
.
Abla
t
io
n study
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
f
u
r
th
er
ass
ess
ed
b
y
ad
o
p
tin
g
an
ab
latio
n
s
tu
d
y
,
wh
er
e
th
e
k
ey
m
o
tiv
e
is
to
s
tu
d
y
th
e
co
n
tr
ib
u
tio
n
o
f
in
d
iv
i
d
u
al
co
m
p
o
n
en
ts
to
war
d
s
ac
c
o
m
p
lis
h
in
g
th
e
s
tated
p
er
f
o
r
m
an
ce
i
n
th
e
p
r
i
o
r
s
ec
t
io
n
.
T
a
b
le
2
h
ig
h
lig
h
ts
th
e
n
u
m
er
ical
o
u
tco
m
e
o
f
th
e
a
b
l
atio
n
s
tu
d
y
th
at
is
o
b
tain
ed
b
y
s
elec
tiv
ely
p
e
r
m
it
tin
g
an
d
d
is
ab
lin
g
p
ar
ticu
lar
c
o
m
p
o
n
en
ts
d
u
r
i
n
g
th
e
co
m
p
le
te
o
p
er
atio
n
.
I
n
th
e
f
ir
s
t
co
n
f
ig
u
r
atio
n
o
f
an
ev
alu
atio
n
,
a
s
tu
d
y
is
ca
r
r
ied
o
u
t
b
y
elim
in
atin
g
th
e
C
R
FM
m
o
d
u
le,
wh
ile
th
e
C
NN
v
er
if
ier
ca
n
b
e
d
ir
ec
tly
a
p
p
lie
d
to
a
c
o
m
p
lete
s
tr
ea
m
o
f
v
i
d
eo
.
T
h
e
o
u
tco
m
e
o
f
th
is
co
n
f
ig
u
r
atio
n
is
n
o
ted
with
m
ax
im
ized
ev
en
t
o
f
o
u
t
lier
s
an
d
a
s
u
d
d
en
in
cr
ea
s
e
in
in
f
er
en
ce
d
u
r
atio
n
.
T
h
is
is
m
ain
ly
d
u
e
to
th
e
v
o
lu
m
in
o
u
s
s
ea
r
ch
s
p
ac
e
r
ep
r
esen
tin
g
th
e
s
ig
n
if
ican
ce
o
f
C
R
FM
to
war
d
s
th
e
g
en
er
atio
n
o
f
ca
n
d
id
ate
r
eg
i
o
n
s
ef
f
icien
tly
.
I
n
t
h
e
s
ec
o
n
d
c
o
n
f
ig
u
r
atio
n
o
f
an
ev
alu
ati
o
n
,
th
e
s
tu
d
y
in
v
o
lv
es
r
em
o
v
in
g
t
h
e
DC
C
N
v
er
if
ier
wh
ile
th
e
s
y
s
tem
d
ir
ec
tly
f
o
r
war
d
s
th
e
C
R
FM
-
g
en
er
ated
r
eg
io
n
to
t
h
e
OC
R
co
m
p
o
n
en
t
.
T
h
is
c
o
n
f
ig
u
r
atio
n
r
esu
lts
in
a
n
o
tab
le
d
e
g
r
ad
ati
o
n
in
p
r
ec
is
io
n
as
th
e
m
o
d
el
d
etec
ts
all
n
o
n
-
te
x
t
r
e
g
io
n
s
w
ith
co
n
to
u
r
as
te
x
t
r
eg
io
n
s
in
co
r
r
ec
tly
.
Hen
ce
,
th
i
s
d
is
p
lay
s
th
e
im
p
o
r
tan
ce
o
f
t
h
e
v
er
if
ier
m
o
d
el
to
war
d
s
co
n
tr
o
llin
g
o
u
tlier
s
.
I
n
th
e
th
ir
d
co
n
f
ig
u
r
atio
n
o
f
an
ev
alu
atio
n
,
th
e
s
tu
d
y
m
o
d
el
d
is
ab
les
th
e
f
ee
d
b
ac
k
m
ec
h
an
is
m
co
n
tr
o
lled
b
y
OC
R
wh
ile
p
r
io
r
m
o
d
u
les
o
f
DC
NN
an
d
C
R
F
M
ar
e
r
etai
n
ed
.
T
h
e
o
u
tco
m
e
s
h
o
ws
th
e
co
m
p
ar
ab
le
r
ec
all
s
co
r
e,
wh
ile
th
e
p
r
ec
is
io
n
is
g
r
ad
u
ally
f
o
u
n
d
to
b
e
d
ec
lin
in
g
in
th
e
p
r
esen
ce
o
f
a
co
m
p
lex
s
ce
n
e
o
f
v
id
eo
s
ce
n
e.
T
h
e
an
aly
s
is
h
ig
h
lig
h
ts
th
e
im
p
o
r
tan
ce
o
f
r
ef
in
em
e
n
t
u
s
in
g
f
ee
d
b
ac
k
in
r
etain
in
g
b
e
tter
co
n
s
is
ten
cy
.
T
ab
le
2
.
Nu
m
e
r
ical
o
u
tco
m
es o
f
ab
latio
n
s
tu
d
y
C
o
n
f
i
g
u
r
a
t
i
o
n
I
n
f
e
r
e
n
c
e
t
i
me
(
ms)
P
r
e
c
i
s
i
o
n
(
%)
A
c
c
u
r
a
c
y
(
%)
O
C
R
f
e
e
d
b
a
c
k
D
C
N
N
v
e
r
i
f
i
e
r
C
R
F
M
F
u
l
l
E
-
TEX
TLO
C
2
9
.
6
9
5
.
4
9
8
.
2
W
i
t
h
o
u
t
O
C
R
f
e
e
d
b
a
c
k
3
0
.
4
8
8
.
6
9
5
.
1
W
i
t
h
o
u
t
D
C
N
N
v
e
r
i
f
i
e
r
3
3
.
9
8
2
.
4
9
2
.
8
W
i
t
h
o
u
t
C
R
F
M
6
1
.
8
8
6
.
1
9
1
.
5
4.
CO
NCLU
SI
O
N
T
h
e
p
r
o
p
o
s
ed
s
tu
d
y
o
f
tex
t
lo
ca
lizatio
n
is
p
r
o
v
en
to
s
u
cc
ess
f
u
lly
ad
d
r
ess
th
e
lim
itatio
n
s
ass
o
ciate
d
with
b
aselin
e
m
o
d
els
b
y
co
llab
o
r
ativ
e
u
s
ag
e
o
f
co
n
f
id
en
c
e
-
b
ased
f
ilter
in
g
,
f
ee
d
b
ac
k
u
s
in
g
a
v
er
if
ier
,
an
d
p
s
eu
d
o
-
lab
ellin
g
.
T
h
e
co
n
tr
ib
u
tio
n
to
war
d
s
ad
v
a
n
ce
m
en
t
i
n
E
-
T
E
XT
L
OC
ca
n
b
e
attr
ib
u
ted
to
an
ef
f
ec
tiv
e
s
am
p
lin
g
o
f
v
id
eo
f
r
am
es,
m
i
n
im
izin
g
th
e
d
u
r
atio
n
o
f
in
f
er
en
ce
f
o
r
ea
c
h
f
r
am
e,
m
ax
im
iz
in
g
th
e
r
ich
n
ess
o
f
f
ea
tu
r
e
e
x
tr
ac
tio
n
an
d
im
p
r
o
v
i
n
g
th
e
r
o
b
u
s
tn
ess
o
f
tex
t
lo
ca
l
izatio
n
f
r
o
m
co
m
p
le
x
v
id
eo
.
T
h
e
s
tu
d
y
o
u
tco
m
e
is
in
s
u
p
p
o
r
t
o
f
ad
v
o
ca
tin
g
E
-
T
E
XT
L
OC
d
ep
lo
y
m
en
t
i
n
r
e
al
-
tim
e
u
s
ag
e
as
well
as
in
an
en
v
ir
o
n
m
e
n
t
with
r
ed
u
ce
d
r
eso
u
r
ce
a
v
ailab
ilit
y
.
C
u
m
u
lativ
ely
,
E
-
T
E
XT
L
OC
c
o
n
tr
ib
u
tes
to
a
p
r
ac
tical
an
d
in
n
o
v
ativ
e
a
p
p
r
o
ac
h
to
war
d
s
lo
ca
lizatio
n
o
f
tex
t
f
r
o
m
v
id
eo
co
n
ten
t.
Ho
wev
er
,
th
er
e
ar
e
ce
r
tain
lim
itatio
n
s
ass
o
ciate
d
to
o
.
A
p
o
ten
tial
p
er
f
o
r
m
an
ce
is
ex
h
i
b
ited
b
y
E
-
T
E
XT
L
OC
o
n
E
n
g
lis
h
v
id
eo
tex
t,
wh
ile
th
e
s
tu
d
y
f
in
d
s
th
at
th
e
g
en
er
aliza
tio
n
ab
ilit
y
to
wa
r
d
s
m
u
ltil
in
g
u
al
s
cr
ip
ts
c
o
u
ld
d
e
m
an
d
p
o
ten
tial
f
in
e
-
tu
n
in
g
an
d
ex
tr
a
tr
ain
in
g
.
I
t
is
also
ex
p
ec
ted
th
at
th
e
ad
o
p
ti
o
n
o
f
d
iv
er
s
if
ied
s
ce
n
a
r
io
s
w
ith
f
aster
s
ce
n
e
ch
an
g
es
o
r
e
x
tr
em
ely
cl
u
tter
ed
b
ac
k
g
r
o
u
n
d
s
co
u
ld
e
v
en
tu
ally
ch
allen
g
e
th
e
lo
ca
lizatio
n
ac
c
u
r
ac
y
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
is
in
v
o
lv
ed
.
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
Day
an
an
d
a
Ko
d
ala
J
ay
ar
am
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Pu
tteg
o
wd
a
Dev
eg
o
w
d
a
✓
✓
✓
✓
✓
✓
✓
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
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
3
7
-
3
5
4
5
3544
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
t
h
o
r
d
ec
lar
es
t
h
at
th
e
r
e
ar
e
n
o
k
n
o
wn
co
n
f
licts
o
f
in
ter
est
ass
o
ciate
d
with
th
is
p
u
b
licatio
n
.
T
h
er
e
ar
e
n
o
f
in
a
n
cial
o
r
p
e
r
s
o
n
al
r
elatio
n
s
h
ip
s
th
at
co
u
ld
in
ap
p
r
o
p
r
iately
in
f
l
u
en
ce
o
r
b
ias
th
e
co
n
ten
t
o
f
th
is
wo
r
k
.
I
NF
O
RM
E
D
CO
NS
E
N
T
No
t
ap
p
licab
le.
T
h
is
s
tu
d
y
d
id
n
o
t
in
v
o
l
v
e
h
u
m
an
p
ar
tici
p
an
ts
,
h
u
m
an
d
ata,
o
r
a
n
y
p
er
s
o
n
ally
id
en
tifia
b
le
in
f
o
r
m
atio
n
.
All
d
ata
u
s
ed
wer
e
eith
er
p
u
b
licl
y
av
ailab
le,
f
u
lly
an
o
n
y
m
ize
d
,
o
r
d
er
i
v
ed
f
r
o
m
n
o
n
‑
h
u
m
an
s
o
u
r
ce
s
,
an
d
th
er
e
f
o
r
e
n
o
in
f
o
r
m
ed
co
n
s
en
t w
as
r
eq
u
ir
ed
f
r
o
m
in
d
iv
i
d
u
als.
E
T
H
I
CAL AP
P
RO
V
AL
No
t
ap
p
licab
le.
T
h
is
r
esear
ch
d
id
n
o
t
in
v
o
lv
e
h
u
m
an
s
u
b
jects,
h
u
m
an
b
io
lo
g
ical
m
ater
ials
,
o
r
ex
p
er
im
en
tal
p
r
o
ce
d
u
r
es
o
n
an
im
als.
T
h
e
wo
r
k
was
co
n
d
u
cted
s
o
lely
o
n
c
o
m
p
u
tatio
n
al
m
o
d
els,
p
u
b
licly
av
ailab
le
d
atasets
,
o
r
n
o
n
‑
s
en
s
itiv
e
d
ata
th
at
d
id
n
o
t
r
eq
u
ir
e
i
n
ter
v
en
tio
n
with
liv
in
g
o
r
g
an
is
m
s
.
T
h
er
ef
o
r
e,
eth
ical
ap
p
r
o
v
al
f
r
o
m
an
in
s
titu
tio
n
al
r
ev
iew
b
o
ar
d
o
r
a
n
im
al
eth
ics
c
o
m
m
itte
e
was
n
o
t
n
ec
ess
ar
y
f
o
r
th
is
s
tu
d
y
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
t
h
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
l
e
f
r
o
m
t
h
e
co
r
r
esp
o
n
d
in
g
a
u
th
o
r
,
[
DKJ
]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
G
o
l
l
a
,
B
.
S
u
j
a
t
h
a
,
a
n
d
L
.
S
u
ma
l
a
t
h
a
,
“
TI
E
-
t
e
x
t
i
n
f
o
r
ma
t
i
o
n
e
x
t
r
a
c
t
i
o
n
f
r
o
m
n
a
t
u
r
a
l
sc
e
n
e
i
m
a
g
e
s
u
s
i
n
g
S
V
M
,
”
Me
a
s
u
reme
n
t
:
S
e
n
so
rs
,
v
o
l
.
3
3
,
Ju
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
a
se
n
.
2
0
2
3
.
1
0
1
0
1
8
.
[
2
]
L.
K
e
t
t
l
e
a
n
d
Y
.
-
C
.
Le
e
,
“
A
u
g
m
e
n
t
e
d
r
e
a
l
i
t
y
f
o
r
v
e
h
i
c
l
e
-
d
r
i
v
e
r
c
o
mm
u
n
i
c
a
t
i
o
n
:
a
s
y
st
e
mat
i
c
r
e
v
i
e
w
,
”
S
a
f
e
t
y
,
v
o
l
.
8
,
n
o
.
4
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
s
a
f
e
t
y
8
0
4
0
0
8
4
.
[
3
]
M
.
B
o
n
a
n
n
o
e
t
a
l
.
,
“
A
ssi
s
t
i
v
e
t
e
c
h
n
o
l
o
g
i
e
s
f
o
r
i
n
d
i
v
i
d
u
a
l
s
w
i
t
h
a
d
i
s
a
b
i
l
i
t
y
f
r
o
m
a
n
e
u
r
o
l
o
g
i
c
a
l
c
o
n
d
i
t
i
o
n
:
a
n
a
r
r
a
t
i
v
e
r
e
v
i
e
w
o
n
t
h
e
m
u
l
t
i
m
o
d
a
l
i
n
t
e
g
r
a
t
i
o
n
,
”
H
e
a
l
t
h
c
a
re
,
v
o
l
.
1
3
,
n
o
.
1
3
,
Ju
l
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
h
e
a
l
t
h
c
a
r
e
1
3
1
3
1
5
8
0
.
[
4
]
M
.
P
.
K
a
n
t
i
p
u
d
i
,
S
.
K
u
m
a
r
,
a
n
d
A
.
K
.
Jh
a
,
“
S
c
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
b
i
d
i
r
e
c
t
i
o
n
a
l
LS
TM
a
n
d
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
N
e
u
r
o
sc
i
e
n
c
e
,
v
o
l
.
2
0
2
1
,
n
o
.
1
,
J
a
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
1
/
2
6
7
6
7
8
0
.
[
5
]
J.
Ta
n
g
,
L
.
W
a
n
g
,
J.
H
u
a
n
g
,
A
.
S
h
i
,
a
n
d
L.
X
u
,
“
I
mag
e
sem
a
n
t
i
c
r
e
c
o
g
n
i
t
i
o
n
a
n
d
se
g
me
n
t
a
t
i
o
n
a
l
g
o
r
i
t
h
m
o
f
c
o
l
o
r
i
met
r
i
c
s
e
n
s
o
r
a
r
r
a
y
b
a
se
d
o
n
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
N
e
u
r
o
sc
i
e
n
c
e
,
v
o
l
.
2
0
2
2
,
p
p
.
1
–
1
6
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
2
4
3
9
3
7
1
.
[
6
]
D.
-
L.
Li
,
S
.
-
K
.
Le
e
,
a
n
d
Y
.
-
T
.
Li
u
,
“
P
r
i
n
t
e
d
d
o
c
u
m
e
n
t
l
a
y
o
u
t
a
n
a
l
y
s
i
s
a
n
d
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
s
y
st
e
m
b
a
s
e
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
rt
s
,
v
o
l
.
1
5
,
n
o
.
1
,
J
u
l
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
5
-
0
7
4
3
9
-
y.
[
7
]
K
.
M
o
h
s
e
n
z
a
d
e
g
a
n
,
V
.
Ta
v
a
k
k
o
l
i
,
a
n
d
K
.
K
y
a
m
a
k
y
a
,
“
A
sm
a
r
t
v
i
s
u
a
l
se
n
s
i
n
g
c
o
n
c
e
p
t
i
n
v
o
l
v
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
a
r
o
b
u
s
t
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
u
n
d
e
r
h
a
r
d
r
e
a
l
-
w
o
r
l
d
c
o
n
d
i
t
i
o
n
s,
”
S
e
n
s
o
rs
,
v
o
l
.
2
2
,
n
o
.
1
6
,
A
u
g
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
2
1
6
6
0
2
5
.
[
8
]
V
.
K
a
d
h
a
,
B
.
B
.
D
u
d
d
e
t
i
,
K
.
S
r
i
n
a
d
h
,
S
.
K
.
B
u
d
d
e
p
u
,
L.
Ja
n
j
a
n
a
m,
a
n
d
K
.
M
e
d
h
i
,
“
F
r
o
m
p
i
x
e
l
s
t
o
t
e
x
t
:
a
d
e
e
p
l
e
a
r
n
i
n
g
su
r
v
e
y
o
f
sce
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
,
”
C
o
m
p
u
t
e
rs
a
n
d
E
l
e
c
t
r
i
c
a
l
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
3
5
,
Ju
l
.
2
0
2
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
e
l
e
c
e
n
g
.
2
0
2
6
.
1
1
1
1
3
9
.
[
9
]
Z.
L
i
u
,
R
.
S
o
n
g
,
K
.
L
i
,
a
n
d
Y
.
L
i
,
“
F
r
o
m
d
e
t
e
c
t
i
o
n
t
o
u
n
d
e
r
s
t
a
n
d
i
n
g
:
a
sy
s
t
e
mat
i
c
su
r
v
e
y
o
f
d
e
e
p
l
e
a
r
n
i
n
g
f
o
r
sce
n
e
t
e
x
t
p
r
o
c
e
ss
i
n
g
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
5
,
n
o
.
1
7
,
A
u
g
.
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
5
1
7
9
2
4
7
.
[
1
0
]
B
.
M
.
A
l
b
a
l
a
w
i
,
A
.
T
.
Jam
a
l
,
L.
A
.
A
.
K
h
u
z
a
y
e
m
,
a
n
d
O
.
A
.
A
l
s
a
e
d
i
,
“
A
n
e
n
d
-
to
-
e
n
d
sc
e
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
f
o
r
b
i
l
i
n
g
u
a
l
t
e
x
t
,
”
Bi
g
D
a
t
a
a
n
d
C
o
g
n
i
t
i
v
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
8
,
n
o
.
9
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
b
d
c
c
8
0
9
0
1
1
7
.
[
1
1
]
Q
.
D
i
n
g
,
E
.
Zh
a
n
g
,
Z.
Li
u
,
X
.
Y
a
o
,
a
n
d
G
.
P
a
n
,
“
T
e
x
t
-
g
u
i
d
e
d
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
c
c
u
r
a
c
y
e
n
h
a
n
c
e
m
e
n
t
me
t
h
o
d
b
a
sed
o
n
i
mp
r
o
v
e
d
Y
O
LO
-
w
o
r
l
d
,
”
E
l
e
c
t
r
o
n
i
c
s
,
v
o
l
.
1
4
,
n
o
.
1
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
e
l
e
c
t
r
o
n
i
c
s
1
4
0
1
0
1
3
3
.
[
1
2
]
M
.
S
i
n
t
h
u
j
a
,
C
.
G
.
P
a
d
u
b
i
d
r
i
,
G
.
S
.
J
a
y
a
c
h
a
n
d
r
a
,
M
.
C
.
Te
j
a
,
a
n
d
G
.
S
.
P
.
K
u
mar,
“
Ex
t
r
a
c
t
i
o
n
o
f
t
e
x
t
f
r
o
m
i
m
a
g
e
s
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
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
.
2
3
5
,
p
p
.
7
8
9
–
7
9
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
4
.
0
4
.
0
7
5
.
[
1
3
]
Z.
R
a
i
si
a
n
d
J.
Z
e
l
e
k
,
“
V
i
s
u
a
l
p
l
a
c
e
r
e
c
o
g
n
i
t
i
o
n
f
r
o
m
e
n
d
-
to
-
e
n
d
s
e
ma
n
t
i
c
sce
n
e
t
e
x
t
f
e
a
t
u
r
e
s,”
Fr
o
n
t
i
e
rs
i
n
Ro
b
o
t
i
c
s
a
n
d
AI
,
v
o
l
.
1
1
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
8
9
/
f
r
o
b
t
.
2
0
2
4
.
1
4
2
4
8
8
3
.
[
1
4
]
A
.
F
.
R
i
z
k
y
,
N
.
Y
u
d
i
st
i
r
a
,
a
n
d
E.
S
a
n
t
o
so
,
“
Te
x
t
r
e
c
o
g
n
i
t
i
o
n
o
n
i
ma
g
e
s
u
s
i
n
g
p
r
e
-
t
r
a
i
n
e
d
C
N
N
,
”
F
e
b
.
2
0
2
3
,
a
rX
i
v
:
2
3
0
2
.
0
5
1
0
5
.
[
1
5
]
R
.
N
o
v
i
y
a
n
t
i
a
n
d
E
.
S
i
n
d
u
n
i
n
g
r
u
m,
“
B
r
a
i
l
l
e
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
h
i
s
t
o
g
r
a
m
o
f
o
r
i
e
n
t
e
d
g
r
a
d
i
e
n
t
s
(
H
O
G
)
a
n
d
S
V
M
-
b
a
se
d
i
ma
g
e
p
r
o
c
e
ssi
n
g
,
”
S
y
n
t
a
x
L
i
t
e
r
a
t
e
:
J
u
r
n
a
l
I
l
m
i
a
h
I
n
d
o
n
e
s
i
a
,
v
o
l
.
9
,
n
o
.
8
,
p
p
.
4
4
1
1
–
4
4
1
9
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
3
6
4
1
8
/
s
y
n
t
a
x
-
l
i
t
e
r
a
t
e
.
v
9
i
8
.
1
6
9
5
1
.
[
1
6
]
M
.
M
.
D
.
S
a
v
i
o
,
T
.
D
e
e
p
a
,
A
.
B
o
n
a
s
u
,
a
n
d
T.
S
.
A
n
u
r
a
g
,
“
I
mag
e
p
r
o
c
e
ssi
n
g
f
o
r
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
H
A
A
R
,
H
O
G
,
a
n
d
S
V
M
a
l
g
o
r
i
t
h
ms,
”
J
o
u
rn
a
l
o
f
Ph
y
si
c
s:
C
o
n
f
e
re
n
c
e
S
e
ri
e
s
,
v
o
l
.
1
9
6
4
,
n
o
.
6
,
Ju
l
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
4
2
-
6
5
9
6
/
1
9
6
4
/
6
/
0
6
2
0
2
3
.
[
1
7
]
S
.
S
h
a
r
ma,
L
.
R
a
j
a
,
V
.
B
h
a
t
n
a
g
a
r
,
D
.
S
h
a
r
ma,
S
.
N
.
B
h
a
g
i
r
a
t
h
,
a
n
d
R
.
C
.
P
o
o
n
i
a
,
“
H
y
b
r
i
d
H
O
G
-
S
V
M
e
n
c
r
y
p
t
e
d
f
a
c
e
d
e
t
e
c
t
i
o
n
a
n
d
r
e
c
o
g
n
i
t
i
o
n
m
o
d
e
l
,
”
J
o
u
r
n
a
l
o
f
D
i
scre
t
e
Ma
t
h
e
m
a
t
i
c
a
l
S
c
i
e
n
c
e
s
a
n
d
C
ry
p
t
o
g
ra
p
h
y
,
v
o
l
.
2
5
,
n
o
.
1
,
p
p
.
2
0
5
–
2
1
8
,
Ja
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
8
0
/
0
9
7
2
0
5
2
9
.
2
0
2
1
.
2
0
1
4
1
4
1
.
[
1
8
]
B
.
Li
u
,
F
.
G
a
o
,
P
.
Z
h
a
o
,
Y
.
L
i
,
a
n
d
W
.
L
i
,
“
M
a
r
k
e
r
i
d
e
n
t
i
f
i
c
a
t
i
o
n
o
f
d
i
s
c
w
o
r
k
p
i
e
c
e
b
a
s
e
d
o
n
H
O
G
-
S
V
M
c
l
a
ssi
f
i
e
r
,
”
M
o
b
i
l
e
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
v
o
l
.
2
0
2
2
,
p
p
.
1
–
9
,
J
u
l
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
6
0
5
8
6
0
2
.
[
1
9
]
L.
M
.
F
r
a
n
c
i
s
a
n
d
N
.
S
r
e
e
n
a
t
h
,
“
TE
D
LESS
–
t
e
x
t
d
e
t
e
c
t
i
o
n
u
s
i
n
g
l
e
a
st
-
s
q
u
a
r
e
S
V
M
f
r
o
m
n
a
t
u
r
a
l
sc
e
n
e
,
”
J
o
u
rn
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rsi
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
2
,
n
o
.
3
,
p
p
.
2
8
7
–
2
9
9
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
su
c
i
.
2
0
1
7
.
0
9
.
0
0
1
.
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
E
-
TEX
TL
OC
:
efficien
t te
xt
lo
c
a
liz
a
tio
n
a
n
d
ex
tr
a
ctio
n
in
r
ea
l
-
w
o
r
ld
vid
eo
… (
Da
ya
n
a
n
d
a
K
o
d
a
la
Ja
ya
r
a
m
)
3545
[
2
0
]
A
.
T.
M
a
u
n
g
,
S
.
S
a
l
e
k
i
n
,
a
n
d
M
.
A
.
H
a
q
u
e
,
“
A
h
y
b
r
i
d
a
p
p
r
o
a
c
h
t
o
B
a
n
g
l
a
h
a
n
d
w
r
i
t
t
e
n
O
C
R
:
c
o
m
b
i
n
i
n
g
Y
O
LO
a
n
d
a
n
a
d
v
a
n
c
e
d
C
N
N
,
”
D
i
sc
o
v
e
r A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
5
,
n
o
.
1
,
J
u
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
4
1
6
3
-
0
2
5
-
0
0
2
5
1
-
7.
[
2
1
]
W
.
A
z
i
z
a
h
a
n
d
S
.
A
g
u
s
t
i
n
,
“
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
u
si
n
g
h
i
s
t
o
g
r
a
m
o
f
o
r
i
e
n
t
e
d
g
r
a
d
i
e
n
t
s
a
n
d
m
o
me
n
t
s
w
i
t
h
r
a
n
d
o
m
f
o
r
e
st
c
l
a
ss
i
f
i
c
a
t
i
o
n
f
o
r
B
a
t
i
k
p
a
t
t
e
r
n
d
e
t
e
c
t
i
o
n
,
”
J
u
rn
a
l
S
i
s
f
o
k
o
m
(
S
i
st
e
m
I
n
f
o
rm
a
si
d
a
n
K
o
m
p
u
t
e
r)
,
v
o
l
.
1
4
,
n
o
.
1
,
p
p
.
8
–
1
4
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
3
2
7
3
6
/
s
i
sf
o
k
o
m.
v
1
4
i
1
.
2
2
2
5
.
[
2
2
]
C
.
M
u
k
k
u
a
n
d
M
.
S
a
n
t
h
o
s
h
,
“
Tr
i
-
st
a
g
e
o
f
f
l
i
n
e
Te
l
u
g
u
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
sy
s
t
e
m
b
a
s
e
d
o
n
f
u
si
o
n
o
f
H
O
G
a
n
d
U
L
B
P
,
”
Me
a
su
r
e
m
e
n
t
:
S
e
n
s
o
rs
,
v
o
l
.
3
2
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
mea
se
n
.
2
0
2
4
.
1
0
1
0
5
9
.
[
2
3
]
J.
L
i
u
,
Z
.
L
i
u
,
Q
.
Li
,
W
.
K
o
n
g
,
a
n
d
X
.
L
i
,
“
M
u
l
t
i
-
d
o
ma
i
n
c
o
n
t
r
o
v
e
r
si
a
l
t
e
x
t
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
a
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
st
a
c
k
e
d
e
n
s
e
m
b
l
e
,
”
M
a
t
h
e
m
a
t
i
c
s
,
v
o
l
.
1
3
,
n
o
.
9
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
m
a
t
h
1
3
0
9
1
5
2
9
.
[
2
4
]
E.
P
i
n
t
e
l
a
s
,
A
.
K
o
u
r
s
a
r
i
s,
I
.
E
.
Li
v
i
e
r
i
s,
a
n
d
V
.
Ta
m
p
a
k
a
s,
“
T
e
x
t
N
e
X
:
t
e
x
t
n
e
t
w
o
r
k
o
f
e
x
p
e
r
t
s
f
o
r
r
o
b
u
st
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
—
c
a
s
e
st
u
d
y
o
n
mac
h
i
n
e
-
g
e
n
e
r
a
t
e
d
-
t
e
x
t
d
e
t
e
c
t
i
o
n
,
”
M
a
t
h
e
m
a
t
i
c
s
,
v
o
l
.
1
3
,
n
o
.
1
0
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
3
3
9
0
/
ma
t
h
1
3
1
0
1
5
5
5
.
[
2
5
]
H
.
K
o
h
l
i
,
J.
A
g
a
r
w
a
l
,
a
n
d
M
.
K
u
m
a
r
,
“
A
n
i
m
p
r
o
v
e
d
me
t
h
o
d
f
o
r
t
e
x
t
d
e
t
e
c
t
i
o
n
u
si
n
g
A
d
a
m
o
p
t
i
mi
z
a
t
i
o
n
a
l
g
o
r
i
t
h
m
,
”
G
l
o
b
a
l
T
ra
n
s
i
t
i
o
n
s
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
3
,
n
o
.
1
,
p
p
.
2
3
0
–
2
3
4
,
Ju
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
g
l
t
p
.
2
0
2
2
.
0
3
.
0
2
8
.
[
2
6
]
N
.
Ja
l
a
l
,
A
.
M
e
h
m
o
o
d
,
G
.
S
.
C
h
o
i
,
a
n
d
I
.
A
sh
r
a
f
,
“
A
n
o
v
e
l
i
m
p
r
o
v
e
d
r
a
n
d
o
m
f
o
r
e
st
f
o
r
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
f
e
a
t
u
r
e
r
a
n
k
i
n
g
a
n
d
o
p
t
i
m
a
l
n
u
mb
e
r
o
f
t
r
e
e
s,”
J
o
u
rn
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rs
i
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
4
,
n
o
.
6
,
p
p
.
2
7
3
3
–
2
7
4
2
,
Ju
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
su
c
i
.
2
0
2
2
.
0
3
.
0
1
2
.
[
2
7
]
B
.
A
v
i
n
a
s
h
,
I
.
A
.
S
h
a
i
k
,
a
n
d
R
.
S
y
e
d
,
“
R
e
a
l
-
t
i
me
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
b
a
se
d
o
n
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
(
O
C
R
)
,
”
J
o
u
rn
a
l
o
f
S
c
i
e
n
c
e
&
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
1
–
1
0
,
2
0
2
4
,
d
o
i
:
1
0
.
4
6
2
4
3
/
j
s
t
.
2
0
2
4
.
v
9
.
i
4
.
p
p
1
-
1
0
.
[
2
8
]
S
.
Μ
.
P
a
t
i
l
,
V
.
S
.
M
a
l
e
m
a
t
h
,
S
.
M
u
d
d
a
p
u
r
,
a
n
d
P
.
M
.
D
h
u
l
a
v
v
a
g
o
l
,
“
En
h
a
n
c
e
d
t
e
x
t
d
e
t
e
c
t
i
o
n
i
n
n
a
t
u
r
a
l
sc
e
n
e
s
u
s
i
n
g
a
d
v
a
n
c
e
d
mac
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s,
”
E
n
g
i
n
e
e
ri
n
g
,
T
e
c
h
n
o
l
o
g
y
&
Ap
p
l
i
e
d
S
c
i
e
n
c
e
Re
se
a
rc
h
,
v
o
l
.
1
5
,
n
o
.
2
,
p
p
.
2
2
1
1
4
–
2
2
1
1
8
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
4
8
0
8
4
/
e
t
a
sr
.
1
0
0
2
9
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Da
y
a
n
a
n
d
a
K
o
d
a
l
a
J
a
y
a
r
a
m
is
wo
rk
in
g
a
s
a
ss
istan
t
p
ro
fe
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
,
Vid
y
a
v
a
rd
h
a
k
a
Co
ll
e
g
e
o
f
En
g
i
n
e
e
rin
g
(A
u
to
n
o
m
o
u
s
In
stit
u
te)
a
ffil
iate
d
to
Visv
e
sv
a
ra
y
a
Tec
h
n
o
lo
g
ica
l
Un
i
v
e
rsity
In
d
ia.
He
h
a
s
re
c
e
iv
e
d
h
is
M
.
Tec
h
.
a
n
d
B
.
E.
fro
m
Visv
e
sv
a
ra
y
a
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsit
y
,
Be
lag
a
v
i,
Ka
rn
a
tak
a
,
In
d
ia.
He
is
p
u
rsu
in
g
h
is
P
h
.
D.
fro
m
Visv
e
sv
a
ra
y
a
Tec
h
n
o
l
o
g
ica
l
Un
i
v
e
rsity
,
Be
lag
a
v
i
,
Ka
rn
a
tak
a
,
In
d
ia.
His
tea
c
h
in
g
a
n
d
re
se
a
rc
h
in
tere
sts
a
re
in
th
e
field
o
f
d
a
ta
m
in
in
g
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
ima
g
e
p
ro
c
e
ss
in
g
.
He
h
a
s
a
ro
u
n
d
to
tal
tea
c
h
in
g
e
x
p
e
rien
c
e
o
f
1
1
y
e
a
rs.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
a
y
a
n
a
n
d
a
.
k
e
m
@g
m
a
il
.
c
o
m
.
Dr
.
Pu
tte
g
o
wda
De
v
e
g
o
wd
a
is
wo
rk
in
g
a
s
p
ro
fe
ss
o
r
a
n
d
h
e
a
d
in
th
e
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
,
ATM
E
Co
l
leg
e
o
f
En
g
i
n
e
e
rin
g
a
ffil
iate
d
to
Visv
e
sv
a
ra
y
a
Tec
h
n
o
lo
g
ica
l
Un
iv
e
rsity
,
In
d
ia.
He
h
a
s
re
c
e
iv
e
d
h
is
P
h
.
D.
fr
o
m
M
y
su
ru
Un
iv
e
rsity
,
M
y
su
r
u
,
Ka
rn
a
tak
a
,
In
d
ia.
His
tea
c
h
in
g
a
n
d
re
se
a
rc
h
in
tere
sts
a
re
in
t
h
e
field
o
f
d
a
ta
m
in
in
g
,
v
i
d
e
o
p
r
o
c
e
ss
in
g
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
ima
g
e
p
ro
c
e
ss
in
g
.
He
h
a
s
a
ro
u
n
d
t
o
ta
l
tea
c
h
in
g
e
x
p
e
rien
c
e
o
f
2
0
y
e
a
rs.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
p
u
t
t
e
g
o
wd
a
.
7
7
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.