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.
14
,
No
.
6
,
Dec
em
b
er
20
25
,
p
p
.
4
6
5
2
~
4
6
6
0
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
6
.
p
p
4
6
5
2
-
4
6
6
0
4652
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Deep learni
ng
appro
a
ches for Bra
i
lle det
ection a
nd
cla
ss
ificatio
n:
co
mpa
ra
tive a
na
ly
sis
Su
re
k
ha
J
a
nra
o
,
T
a
v
i
o
n F
er
na
nd
es,
O
j
a
s
G
o
la
t
ka
r,
Swa
ra
j
Dus
a
ne
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
K
.
J
.
S
o
ma
i
y
a
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
M
u
m
b
a
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
Ma
y
1
8
,
2
0
2
4
R
ev
is
ed
Sep
8
,
2
0
2
5
Acc
ep
ted
Oct
1
8
,
2
0
2
5
Th
is
stu
d
y
p
r
o
p
o
se
s
a
h
y
b
rid
a
p
p
ro
a
c
h
to
Bra
il
le
tran
sla
ti
o
n
lev
e
ra
g
in
g
t
h
e
stre
n
g
th
s
o
f
b
o
th
YO
LO
fo
r
o
b
je
c
t
d
e
tec
ti
o
n
a
n
d
m
u
lt
it
u
d
e
o
f
c
la
ss
ifi
c
a
ti
o
n
m
o
d
e
ls
su
c
h
a
s
Re
sN
e
t,
a
n
d
Re
sN
e
t
fo
r
a
c
c
u
ra
te
Bra
il
le
c
h
a
ra
c
ter
c
las
sifica
ti
o
n
fro
m
ima
g
e
s.
Up
o
n
c
o
m
p
a
r
in
g
n
u
m
e
ro
u
s
m
o
d
e
ls
o
n
v
a
rio
u
s
p
e
rfo
rm
a
n
c
e
m
e
tri
c
s,
Re
sN
e
t
a
n
d
De
n
se
Ne
t
o
u
tp
e
rf
o
rm
e
d
o
th
e
r
m
o
d
e
ls,
e
x
h
ib
it
in
g
h
ig
h
a
c
c
u
ra
c
y
(0
.
9
4
8
7
a
n
d
0
.
9
6
4
7
re
sp
e
c
ti
v
e
ly
)
a
n
d
F
1
-
sc
o
re
s
(0
.
9
4
8
1
a
n
d
0
.
9
6
6
6
)
d
u
e
to
th
e
ir
d
e
e
p
,
d
e
n
se
ly
c
o
n
n
e
c
ted
a
rc
h
it
e
c
t
u
re
s
a
d
e
p
t
a
t
c
a
p
tu
rin
g
i
n
tri
c
a
te
Bra
il
le
p
a
tt
e
rn
s.
CNN
s
with
p
o
o
l
in
g
sh
o
we
d
b
a
lan
c
e
d
re
su
lt
s,
wh
il
e
M
o
b
il
e
Ne
tV2
'
s
li
g
h
twe
ig
h
t
d
e
sig
n
li
m
it
e
d
c
o
m
p
lex
c
las
sifica
ti
o
n
.
Re
sN
e
Xt'
s
m
u
lt
i
-
p
a
th
lea
rn
i
n
g
a
c
h
iev
e
d
re
sp
e
c
tab
le
p
e
rfo
rm
a
n
c
e
b
u
t
lag
g
e
d
b
e
h
in
d
R
e
sN
e
t
a
n
d
De
n
se
Ne
t.
In
t
h
e
fu
t
u
re
th
e
re
su
lt
s
fro
m
o
u
r
st
u
d
y
c
o
u
l
d
b
e
fu
r
th
e
r
e
x
p
lo
re
d
o
n
c
o
n
trac
ted
Bra
il
le
re
c
o
g
n
it
io
n
,
b
e
a
d
a
p
ted
t
o
v
a
rio
u
s
Bra
il
le
c
o
d
e
s,
a
n
d
o
p
ti
m
ize
d
fo
r
m
o
b
il
e
d
e
v
ice
s
,
fo
r
re
a
l
-
ti
m
e
Bra
il
le d
e
tec
ti
o
n
a
n
d
tran
sla
t
io
n
o
n
sm
a
rtp
h
o
n
e
s.
K
ey
w
o
r
d
s
:
B
r
aille
clas
s
if
icatio
n
B
r
aille
d
etec
tio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Den
s
eNe
t
Mo
b
ileNetV2
R
esNet
R
esNeX
t
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
:
Su
r
ek
h
a
J
an
r
a
o
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
K.
J
.
So
m
aiy
a
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
Sio
n
,
Mu
m
b
ai
4
0
0
0
2
2
,
I
n
d
ia
E
m
ail:
s
u
r
ek
h
a.
jan
r
a
o
@
s
o
m
aiy
a.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
Vis
u
al
im
p
air
m
en
t
af
f
ec
ts
a
s
ig
n
if
ican
t
p
o
r
tio
n
o
f
th
e
g
lo
b
al
p
o
p
u
latio
n
,
with
I
n
d
ia
al
o
n
e
h
a
v
in
g
ap
p
r
o
x
im
ately
7
0
m
illi
o
n
v
is
u
ally
im
p
air
ed
in
d
iv
id
u
als,
in
cl
u
d
in
g
4
.
9
5
m
illi
o
n
b
lin
d
p
e
o
p
le
an
d
0
.
2
4
m
illi
o
n
ch
ild
r
en
.
Fo
r
t
h
ese
p
eo
p
le,
B
r
aille
r
em
ain
s
th
e
p
r
im
ar
y
m
ed
iu
m
f
o
r
liter
ac
y
an
d
co
m
m
u
n
icatio
n
.
A
s
tu
d
y
co
n
d
u
cte
d
in
Delh
i
NC
R
h
as
s
h
o
wn
th
at
v
is
u
ally
im
p
air
ed
in
d
iv
id
u
als
f
ac
e
s
ig
n
if
ican
t
m
ed
i
ca
tio
n
m
an
a
g
em
en
t
ch
allen
g
es
d
u
e
to
lack
o
f
e
x
te
n
s
iv
e
u
s
e
o
f
B
r
aille
in
d
aily
li
f
e,
wh
ich
ca
n
lead
to
s
ev
er
e
h
ea
l
th
co
n
s
eq
u
en
ce
s
[
1
]
.
Ho
wev
e
r
,
th
e
in
te
g
r
atio
n
o
f
ad
v
a
n
ce
d
tech
n
o
lo
g
ies
h
as
t
h
e
p
o
ten
tial
to
s
ig
n
if
ica
n
tly
en
h
an
ce
in
clu
s
io
n
a
n
d
co
n
n
ec
tiv
ity
in
o
u
r
in
cr
ea
s
in
g
l
y
d
ig
ital
wo
r
ld
.
R
ec
en
t
ad
v
a
n
c
em
en
ts
in
clu
d
e
b
r
aille
d
is
p
lay
s
with
s
er
v
o
m
o
to
r
s
u
tili
zin
g
Go
o
g
le'
s
T
ess
er
ac
t
O
C
R
en
g
in
e
[
2
]
,
I
o
T
d
ev
ices
d
e
s
ig
n
ed
f
o
r
r
ec
o
g
n
izin
g
s
p
ec
if
i
c
B
r
aille
ch
ar
ac
ter
s
[
3
]
–
[
8
]
,
a
n
d
t
h
e
B
r
ailleNe
t
m
o
d
el,
wh
ich
in
c
o
r
p
o
r
ates
f
o
r
eg
r
o
u
n
d
atten
tio
n
a
n
d
s
em
an
tic
le
ar
n
in
g
[
9
]
.
B
u
ild
in
g
o
n
th
ese
d
ev
elo
p
m
en
ts
,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
o
f
f
er
s
ig
n
if
ican
t p
o
te
n
tial f
o
r
B
r
aille
d
etec
tio
n
an
d
class
if
icatio
n
,
en
ab
lin
g
b
r
o
ad
er
r
ec
o
g
n
itio
n
o
f
d
iv
er
s
e
B
r
aille
ch
ar
ac
ter
s
an
d
en
h
an
ci
n
g
ac
ce
s
s
ib
ilit
y
f
o
r
v
is
u
ally
im
p
air
ed
in
d
iv
i
d
u
als
[
1
0
]
–
[
1
5
]
.
Nasir
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
an
An
d
r
o
id
ap
p
lica
tio
n
th
at
u
tili
ze
s
a
C
NN
-
b
ased
v
is
u
al
r
ec
o
g
n
itio
n
m
o
d
el
with
th
e
T
en
s
o
r
Flo
w
o
b
ject
API
an
d
s
in
g
le
s
h
o
t
d
ete
cto
r
(
SS
D)
u
s
in
g
a
p
r
e
-
tr
ain
ed
Mo
b
ileNetV2
.
T
h
i
s
ap
p
licatio
n
ass
is
ts
v
is
u
ally
i
m
p
air
ed
u
s
er
s
in
th
eir
d
aily
ac
tiv
ities
with
o
u
t
th
e
n
ee
d
f
o
r
an
i
n
ter
n
et
co
n
n
ec
tio
n
,
h
ig
h
li
g
h
tin
g
t
h
e
p
r
ac
tical
p
o
ten
tial o
f
C
NNs in
r
ea
l
-
life
task
s
.
W
h
i
l
e
s
o
m
e
s
t
u
d
i
e
s
h
a
v
e
co
m
p
a
r
ed
th
e
p
er
f
o
r
m
a
n
c
e
o
f
v
a
r
io
u
s
C
N
N
m
o
d
e
l
s
f
o
r
B
r
a
i
l
l
e
r
e
c
o
g
n
i
t
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
io
n
[
1
7
]
,
[
1
8
]
,
a
co
m
p
r
eh
e
n
s
iv
e
a
n
a
ly
s
i
s
o
f
ad
v
an
c
ed
d
e
e
p
l
e
ar
n
in
g
a
r
c
h
i
t
e
c
tu
r
e
s
s
u
c
h
a
s
R
e
s
N
e
t
,
D
e
n
s
e
N
e
t
,
a
n
d
R
e
s
Ne
X
t
i
s
s
t
i
l
l
l
a
ck
i
n
g
[
1
9
]
–
[
2
3
]
.
T
h
i
s
g
a
p
in
r
e
s
e
a
r
ch
l
im
i
t
s
o
u
r
u
n
d
er
s
t
a
n
d
i
n
g
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es fo
r
B
r
a
ille d
etec
tio
n
a
n
d
cla
s
s
ifica
tio
n
:
co
m
p
a
r
a
tive
a
n
a
lysi
s
(
S
u
r
ek
h
a
Ja
n
r
a
o
)
4653
t
h
e
f
u
l
l
p
o
t
e
n
t
i
al
o
f
th
e
s
e
ar
c
h
i
t
e
c
tu
r
e
s
f
o
r
B
r
a
i
l
l
e
r
e
co
g
n
i
t
io
n
t
a
s
k
s
.
A
d
d
i
ti
o
n
a
l
l
y
,
t
h
e
co
m
p
u
t
a
t
io
n
a
l
e
f
f
i
c
i
en
cy
c
r
u
c
i
a
l
f
o
r
r
ea
l
-
t
im
e
B
r
a
i
l
l
e
r
e
co
g
n
i
t
i
o
n
o
n
m
o
b
i
l
e
d
ev
i
c
e
s
i
s
o
f
t
e
n
o
v
e
r
lo
o
k
ed
.
T
h
i
s
o
v
e
r
s
ig
h
t
l
i
m
i
t
s
t
h
e
p
r
ac
t
i
c
a
l
ap
p
l
ic
a
b
i
l
i
t
y
o
f
m
an
y
p
r
o
p
o
s
ed
s
o
lu
t
i
o
n
s
,
e
s
p
e
c
i
a
l
ly
f
o
r
e
v
e
r
y
d
ay
u
s
e
b
y
v
i
s
u
a
l
l
y
i
m
p
a
i
r
e
d
in
d
iv
i
d
u
a
l
s
.
Fu
r
th
e
r
m
o
r
e,
m
o
s
t
r
e
s
e
a
r
ch
f
o
cu
s
e
s
o
n
s
p
e
c
if
i
c
l
an
g
u
a
g
e
s
,
w
i
t
h
a
n
o
t
a
b
le
l
a
ck
o
f
s
o
l
u
t
i
o
n
s
f
o
r
d
iv
er
s
e
l
a
n
g
u
ag
e
r
e
q
u
i
r
em
e
n
t
s
[
2
4
]
,
[
2
5
]
.
T
o
a
d
d
r
e
s
s
t
h
e
s
e
ch
a
l
l
en
g
e
s
,
we
p
r
o
p
o
s
e
a
h
y
b
r
i
d
a
p
p
r
o
a
ch
t
h
a
t
l
e
v
e
r
ag
e
s
th
e
s
t
r
e
n
g
t
h
s
o
f
b
o
t
h
th
e
Y
O
L
O
m
o
d
e
l
f
o
r
r
a
p
id
B
r
a
i
l
le
ch
a
r
a
c
t
e
r
d
e
t
ec
t
i
o
n
an
d
s
p
e
c
i
a
l
i
z
ed
C
N
N
s
f
o
r
d
e
t
a
i
l
ed
c
la
s
s
i
f
i
c
a
t
i
o
n
.
T
h
i
s
s
e
p
a
r
a
t
io
n
o
f
d
e
t
e
c
t
io
n
an
d
c
l
a
s
s
i
f
i
ca
t
i
o
n
p
r
o
c
e
s
s
e
s
a
i
m
s
t
o
b
a
l
an
c
e
s
p
e
ed
a
n
d
a
c
cu
r
a
cy
,
p
r
o
v
i
d
in
g
a
s
c
a
l
ab
l
e
,
m
o
d
u
l
a
r
s
o
lu
t
i
o
n
a
d
ap
t
a
b
le
to
v
ar
i
o
u
s
u
s
e
c
a
s
e
s
a
n
d
p
o
t
en
t
i
a
l
l
y
i
m
p
r
o
v
i
n
g
o
v
er
a
l
l
e
f
f
i
c
i
en
cy
i
n
B
r
a
i
l
l
e
r
e
co
g
n
i
t
io
n
.
I
n
th
is
p
a
p
er
,
we
p
r
esen
t
a
t
h
o
r
o
u
g
h
co
m
p
a
r
is
o
n
o
f
v
ar
io
u
s
C
NN
m
o
d
els'
p
er
f
o
r
m
an
ce
in
B
r
aille
ch
ar
ac
ter
class
if
icatio
n
.
T
h
e
m
o
d
els
co
m
p
a
r
ed
in
clu
d
e
ar
ch
it
ec
tu
r
es
lik
e
R
esNet,
Den
s
eNe
t,
Mo
b
ileNetV2
,
a
n
d
R
esNeX
t.
W
ith
th
is
s
tu
d
y
,
we
aim
to
c
o
n
tr
ib
u
te
to
th
e
d
ev
elo
p
m
en
t
o
f
m
o
r
e
ef
f
icien
t,
v
er
s
a
tile,
an
d
p
r
ac
tically
ap
p
licab
le
B
r
aille
d
etec
tio
n
a
n
d
class
if
icatio
n
s
y
s
tem
s
,
u
ltima
tely
en
h
an
ci
n
g
th
e
q
u
ality
o
f
life
f
o
r
v
i
s
u
ally
im
p
air
ed
in
d
i
v
id
u
als wo
r
ld
wi
d
e.
2.
M
E
T
H
O
D
2
.
1
.
B
ra
ille c
ha
ra
ct
er
det
ec
t
io
n us
ing
YO
L
O
v
8
T
o
ac
h
iev
e
a
tr
an
s
lated
B
r
aille,
th
e
f
ir
s
t
task
wo
u
l
d
b
e
to
d
etec
t
th
e
B
r
aille
ch
ar
ac
ter
s
p
r
esen
t
in
th
e
in
p
u
t im
ag
e.
T
h
is
is
ac
h
iev
ed
b
y
tak
in
g
ad
v
a
n
tag
e
o
f
th
e
Y
OL
Ov
8
o
b
ject
d
etec
tio
n
m
o
d
e
l tr
ain
ed
o
n
a
B
r
aille
d
ataset
[
2
6
]
.
T
h
is
d
ataset
co
n
s
is
ted
o
f
2
0
0
0
an
n
o
tated
im
ag
es
o
f
b
r
aille
ch
ar
ac
ter
s
in
v
ar
io
u
s
s
ce
n
es.
T
h
e
d
etec
ted
c
h
ar
ac
ter
s
wer
e
c
r
o
p
p
e
d
an
d
g
iv
e
n
as
in
p
u
t
to
th
e
class
if
icatio
n
m
o
d
el
wh
ich
th
en
tr
an
s
lated
th
em
.
T
h
e
u
s
e
o
f
two
s
ep
ar
ate
C
NN
m
o
d
els
f
o
r
th
e
task
s
o
f
d
etec
tio
n
an
d
class
if
icatio
n
en
s
u
r
es
ac
cu
r
ac
y
b
y
lev
er
ag
in
g
th
e
s
tr
en
g
th
s
o
f
ea
ch
m
o
d
el.
T
h
e
g
en
er
al
p
r
o
ce
s
s
o
f
d
etec
tin
g
B
r
aille
ch
a
r
ac
ter
s
u
s
in
g
YOL
Ov
8
is
d
ep
icted
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
Gen
e
r
al
wo
r
k
i
n
g
o
f
B
r
aille
ch
ar
ac
ter
d
etec
tio
n
m
o
d
el
I
n
itially
,
R
GB
im
ag
e
th
at
c
o
n
tain
s
B
r
aille
ch
ar
ac
ter
s
is
co
n
v
er
ted
f
r
o
m
th
e
R
GB
co
lo
r
s
p
ac
e
to
th
e
YC
b
C
r
co
lo
r
s
p
ac
e,
a
w
id
ely
u
s
ed
co
lo
r
en
co
d
i
n
g
s
y
s
tem
in
d
ig
ital
im
ag
e
an
d
v
id
eo
p
r
o
c
ess
in
g
.
T
h
is
allo
w
s
f
o
r
b
etter
h
an
d
lin
g
o
f
l
u
m
in
an
ce
an
d
c
h
r
o
m
i
n
an
ce
wh
ic
h
en
h
an
ce
s
th
e
co
n
tr
ast
b
etwe
en
B
r
aille
d
o
ts
an
d
th
eir
b
ac
k
g
r
o
u
n
d
allo
win
g
f
o
r
o
p
ti
m
al
d
etec
tio
n
an
d
class
if
icatio
n
.
Af
ter
co
n
v
e
r
s
io
n
p
r
e
p
r
o
ce
s
s
in
g
is
ap
p
lied
o
n
th
e
im
ag
e
to
r
em
o
v
e
b
ac
k
g
r
o
u
n
d
clu
tter
an
d
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
b
y
elim
in
atin
g
n
o
is
e
an
d
s
u
b
s
eq
u
en
tly
f
alse
p
o
s
itiv
es.
YOL
Ov
8
m
o
d
el
d
i
v
id
es
th
e
in
p
u
t
im
ag
e
i
n
to
a
g
r
id
an
d
m
ak
es
p
r
e
d
ictio
n
s
with
in
ea
ch
g
r
id
ce
ll
r
eg
ar
d
in
g
th
e
lo
ca
tio
n
o
f
b
o
u
n
d
in
g
b
o
x
es
an
d
class
p
r
o
b
ab
ilit
ies
f
o
r
t
h
e
o
b
jects
it
d
etec
ts
.
T
h
e
m
o
d
el
ca
lcu
lates
co
n
f
id
en
ce
s
co
r
es
to
d
e
ter
m
i
n
e
th
e
lik
elih
o
o
d
th
at
a
p
ar
ti
cu
lar
b
o
u
n
d
in
g
b
o
x
co
n
tain
s
a
B
r
aille
ch
ar
ac
ter
.
T
o
in
cr
ea
s
e
d
etec
tio
n
ac
cu
r
a
cy
in
th
e
co
n
tex
t
o
f
B
r
aille,
th
e
p
r
e
-
p
r
o
ce
s
s
ed
im
a
g
e
s
er
v
es
as
th
e
in
p
u
t
t
o
YOL
Ov
8
,
wh
ich
d
etec
ts
in
d
iv
id
u
al
r
aised
d
o
ts
th
at
f
o
r
m
B
r
aille
ch
a
r
ac
ter
s
.
T
h
ese
a
r
r
an
g
e
m
en
ts
o
f
d
o
ts
with
in
a
2
×
3
g
r
id
ar
e
i
d
en
tifie
d
b
ased
o
n
th
eir
s
ize,
s
h
ap
e
,
an
d
s
p
atial
ar
r
an
g
em
e
n
t.
2
.
2
.
B
ra
ille c
ha
ra
ct
er
cla
s
s
if
ica
t
io
n
Af
ter
d
etec
tio
n
,
th
e
YOL
Ov
8
m
o
d
el
r
et
u
r
n
s
b
o
u
n
d
in
g
b
o
x
co
o
r
d
in
ates
f
o
r
ea
ch
o
f
t
h
e
B
r
aille
ch
ar
ac
ter
s
wh
ich
ar
e
p
r
o
ce
s
s
ed
f
u
r
th
e
r
to
e
x
tr
ac
t
th
e
ch
a
r
ac
ter
s
.
T
h
is
is
p
er
f
o
r
m
e
d
with
th
e
aid
o
f
th
e
Op
en
C
V,
wh
ich
tak
es
th
e
b
o
u
n
d
in
g
b
o
x
co
o
r
d
in
ates
r
etu
r
n
e
d
b
y
YOL
Ov
8
an
d
c
r
o
p
s
t
h
e
r
ele
v
an
t
r
eg
io
n
f
r
o
m
th
e
o
r
ig
in
a
l
im
ag
e,
is
o
latin
g
ev
er
y
B
r
aille
ch
ar
ac
ter
.
T
h
ese
cr
o
p
p
ed
c
h
ar
ac
ter
s
ca
n
n
o
w
b
e
s
en
t a
s
an
in
p
u
t to
s
ev
er
al
d
ee
p
lear
n
in
g
-
b
ased
m
o
d
els
to
class
if
y
th
em
ac
cu
r
ately
.
I
n
th
i
s
s
tu
d
y
,
we
aim
to
ev
alu
ate
an
d
c
o
m
p
a
r
e
th
e
p
er
f
o
r
m
an
ce
o
f
v
ar
i
o
u
s
d
ee
p
lear
n
in
g
m
o
d
el
ar
ch
itectu
r
e
s
in
clas
s
if
y
in
g
B
r
aille
ch
ar
ac
ter
s
.
Fo
r
B
r
aille
ch
ar
ac
ter
class
if
icatio
n
,
th
e
d
ataset
[
2
7
]
u
s
ed
co
n
s
titu
ted
n
u
m
er
o
u
s
im
ag
es
o
f
ea
ch
B
r
aille
ch
ar
ac
ter
in
a
2
8
×
28
B
W
s
ca
le.
T
h
e
d
ata
also
wen
t
t
h
r
o
u
g
h
s
ev
e
r
al
ty
p
es
o
f
d
ata
au
g
m
en
tatio
n
in
clu
d
in
g
wid
th
h
e
ig
h
t
s
h
if
t,
r
o
tatio
n
,
etc.
First,
we
co
n
s
id
er
ed
C
N
Ns
with
p
o
o
lin
g
,
with
th
r
ee
co
n
v
o
l
u
tio
n
al
lay
er
s
with
3
2
,
6
4
,
an
d
1
2
8
f
ilter
s
r
esp
ec
tiv
ely
.
T
h
is
C
NN
b
ased
B
r
aille
class
if
ier
ap
p
lies
a
s
im
p
le
ar
c
h
itectu
r
e
t
h
at
is
f
o
llo
wed
b
y
m
a
x
p
o
o
lin
g
to
r
ed
u
ce
s
p
atial
d
im
en
s
io
n
s
.
T
h
e
co
n
v
o
lu
tio
n
al
la
y
er
s
ef
f
icie
n
tly
ca
p
tu
r
e
th
e
f
in
e
s
p
atial
d
et
ail
an
d
h
ier
a
r
ch
ic
al
f
ea
tu
r
es
in
th
e
in
p
u
t
im
ag
e
wit
h
a
k
e
r
n
el
s
ize
o
f
3
×
3
.
T
h
e
m
o
d
el
also
in
clu
d
es
f
u
lly
co
n
n
ec
t
ed
lay
er
s
to
p
er
f
o
r
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
6
5
2
-
4
6
6
0
4654
class
if
icatio
n
with
h
id
d
en
lay
er
s
ize
o
f
5
1
2
.
T
h
e
d
r
o
p
o
u
t
l
ay
er
is
th
en
in
co
r
p
o
r
ated
in
t
h
e
f
u
lly
co
n
n
ec
ted
s
ec
tio
n
an
d
aim
s
to
r
e
d
u
ce
th
e
o
v
er
f
itti
n
g
an
d
e
n
h
an
ce
th
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
.
T
h
is
is
f
o
llo
wed
b
y
r
esid
u
al
n
eu
r
al
n
etwo
r
k
s
(
R
esNet
)
th
at
f
ea
tu
r
es
r
esid
u
al
b
lo
ck
s
wh
ich
in
teg
r
ates
s
k
ip
co
n
n
ec
tio
n
s
to
f
ac
ilit
ate
g
r
ad
ien
t
f
lo
w
an
d
r
ed
u
ce
th
e
p
r
o
b
lem
o
f
v
an
is
h
in
g
g
r
ad
ie
n
t.
I
n
itially
,
th
e
n
etwo
r
k
b
eg
in
s
with
a
co
n
v
o
lu
tio
n
an
d
m
ax
p
o
o
lin
g
lay
er
,
it is
th
en
f
o
llo
wed
b
y
p
r
e
-
r
esid
u
al
b
lo
ck
s
t
h
at
p
r
o
g
r
ess
iv
ely
d
ec
r
ea
s
e
s
p
atial
d
im
en
s
io
n
s
an
d
d
ee
p
e
n
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
E
ac
h
r
esid
u
al
b
lo
c
k
co
n
tain
s
two
co
n
v
o
l
u
tio
n
al
lay
er
s
s
u
p
p
o
r
ted
b
y
b
atc
h
n
o
r
m
aliza
tio
n
(
B
atch
No
r
m
)
an
d
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
ac
t
iv
atio
n
to
s
tab
ilize
tr
ain
in
g
wh
ile
also
en
h
an
cin
g
f
ea
tu
r
e
lear
n
in
g
.
An
o
th
er
m
o
d
el
th
at
we
ex
am
in
ed
was,
d
en
s
ely
co
n
n
ec
ted
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
s
(
Den
s
eNe
t)
.
T
h
e
class
if
ier
tak
es
th
e
ad
v
an
tag
e
o
f
a
d
en
s
e
c
o
n
n
ec
t
iv
ity
p
atter
n
wh
e
r
e
ea
ch
lay
er
is
co
n
n
ec
ted
to
ev
e
r
y
p
r
ev
io
u
s
lay
er
en
s
u
r
in
g
ef
f
icien
t
lear
n
in
g
as
well
as
p
r
o
m
o
tes
f
ea
tu
r
e
r
eu
s
e.
T
h
e
ar
ch
itectu
r
e
s
tar
ts
with
a
n
in
itial
co
n
v
o
lu
tio
n
al
lay
er
f
o
llo
wed
b
y
d
en
s
e
b
lo
c
k
s
wh
er
e
th
e
in
p
u
ts
ar
e
co
n
ca
ten
ated
with
th
e
o
u
tp
u
ts
o
f
th
e
p
r
ev
io
u
s
lay
e
r
s
.
T
h
e
n
u
m
b
er
o
f
f
ea
tu
r
e
m
ap
s
an
d
d
o
wn
s
am
p
le
s
p
atial
d
im
en
s
io
n
s
ar
e
r
e
d
u
ce
d
b
y
in
t
r
o
d
u
cin
g
tr
an
s
itio
n
lay
e
r
s
b
et
wee
n
th
e
d
en
s
e
b
l
o
ck
s
.
W
e
h
av
e
also
tr
ied
o
t
h
er
m
o
d
els
s
u
ch
as
Mo
b
ileNetV2
an
d
R
esNeX
t.
Mo
b
ileNetV2
is
d
esig
n
ed
f
o
r
ef
f
icien
t
in
f
er
e
n
ce
o
n
m
o
b
ile
an
d
em
b
ed
d
e
d
d
e
v
ices
with
lim
ited
co
m
p
u
tatio
n
al
r
es
o
u
r
ce
s
.
I
t
ac
h
iev
es
ef
f
icien
cy
th
r
o
u
g
h
d
ep
t
h
-
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
s
wh
ich
s
ig
n
if
ican
tly
r
ed
u
ce
s
th
e
c
o
m
p
u
tatio
n
al
co
s
t
wh
ile
m
ain
tain
in
g
r
e
p
r
esen
tatio
n
al
c
ap
ac
ity
.
Mo
b
ileNetV2
’
s
lig
h
t
weig
h
t
ar
ch
itectu
r
e
m
ak
in
g
it
s
u
itab
le
f
o
r
r
ea
l
-
tim
e
in
f
er
en
ce
o
n
d
ev
ices
h
a
v
in
g
c
o
n
s
tr
ain
ed
r
eso
u
r
ce
s
,
esp
ec
iall
y
in
s
ce
n
ar
i
o
s
wh
er
e
c
o
m
p
u
tatio
n
al
e
f
f
icien
cy
is
o
f
u
tm
o
s
t im
p
o
r
ta
n
ce
.
L
astl
y
,
th
e
R
esNeXt
m
o
d
el,
wh
ich
is
an
ad
v
a
n
ce
d
v
ar
ian
t
o
f
R
esNet
m
o
d
el
th
at
aim
s
t
o
en
h
an
ce
f
ea
tu
r
e
d
iv
er
s
ity
b
y
e
m
p
lo
y
in
g
th
e
co
n
ce
p
t
o
f
ca
r
d
i
n
ality
b
y
u
s
in
g
g
r
o
u
p
ed
co
n
v
o
lu
tio
n
s
.
T
h
e
m
o
d
el
b
eg
in
s
,
with
an
in
itial
co
n
v
o
l
u
tio
n
an
d
B
atch
No
r
m
lay
er
,
i
t
is
th
en
s
u
cc
ee
d
ed
b
y
f
o
u
r
R
esNet
b
lo
ck
s
th
at
d
o
wn
s
am
p
le
s
p
atial
d
im
en
s
io
n
s
an
d
in
cr
ea
s
es
th
e
f
ea
tu
r
e
d
ep
th
.
L
astl
y
,
th
e
m
o
d
el
co
n
cl
u
d
es
with
th
e
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
an
d
a
f
u
lly
co
n
n
ec
ted
lay
er
f
o
r
class
if
icatio
n
u
tili
zin
g
ca
r
d
in
ality
to
en
h
a
n
ce
r
ep
r
esen
tatio
n
al
p
o
wer
an
d
ef
f
icien
c
y
with
o
u
t
s
ig
n
if
ican
t in
cr
ea
s
e
in
p
ar
am
et
er
co
u
n
t.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
F
o
r
m
o
d
e
l
ev
a
lu
a
t
i
o
n
,
s
t
a
n
d
a
r
d
m
e
t
r
ic
s
l
i
k
e
a
cc
u
r
a
c
y
,
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l,
a
n
d
F
1
-
s
c
o
r
e
w
er
e
c
a
l
c
u
l
a
t
ed
f
o
r
e
a
c
h
m
o
d
e
l
v
a
r
i
a
n
t
.
E
a
ch
m
o
d
e
l
wa
s
t
r
a
i
n
ed
f
o
r
1
0
e
p
o
c
h
s
.
C
o
n
f
u
s
i
o
n
m
a
tr
i
c
e
s
f
r
o
m
d
i
f
f
e
r
en
t
m
o
d
e
l
v
a
r
ia
n
t
s
w
e
r
e
a
l
s
o
c
o
m
p
a
r
ed
t
o
i
d
e
n
t
i
f
y
c
o
m
m
o
n
m
i
s
c
l
a
s
s
i
f
i
c
a
t
i
o
n
s
an
d
ar
e
a
s
o
f
d
is
a
g
r
e
em
e
n
t
a
m
o
n
g
t
h
e
m
o
d
e
l
s
.
3
.
1
.
Co
nv
o
lutio
na
l neura
l net
wo
rk
s
wit
h po
o
lin
g
Fro
m
Fig
u
r
e
2
we
ca
n
in
f
e
r
th
at
t
h
e
lo
s
s
cu
r
v
e
d
em
o
n
s
tr
ates
a
s
tead
y
d
ec
lin
e
an
d
ev
en
tu
al
co
n
v
er
g
en
ce
.
T
h
e
ac
cu
r
ac
y
c
u
r
v
e
s
h
o
ws
a
r
a
p
id
in
itial
in
cr
e
ase,
in
d
icatin
g
s
tr
o
n
g
lear
n
i
n
g
p
r
o
g
r
ess
.
Fro
m
th
e
co
n
f
u
s
io
n
m
atr
ix
as
s
h
o
wn
in
Fig
u
r
e
3
,
we
ca
n
i
d
en
tify
a
f
e
w
in
s
tan
ce
s
o
f
m
is
class
if
icatio
n
,
h
o
wev
er
,
th
ese
ar
e
n
eg
lig
ib
le
an
d
d
o
n
o
t sig
n
if
ica
n
tly
im
p
ac
t th
e
r
ea
l
-
wo
r
ld
a
p
p
licatio
n
.
Fig
u
r
e
2
.
Gr
a
p
h
s
co
n
tain
i
n
g
lo
s
s
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
an
d
F1
s
co
r
e
o
f
C
NN
with
p
o
o
lin
g
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
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es fo
r
B
r
a
ille d
etec
tio
n
a
n
d
cla
s
s
ifica
tio
n
:
co
m
p
a
r
a
tive
a
n
a
lysi
s
(
S
u
r
ek
h
a
Ja
n
r
a
o
)
4655
Fig
u
r
e
3
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
C
NN
with
p
o
o
lin
g
m
o
d
el
3
.
2
.
ResNet
T
h
e
p
er
f
o
r
m
an
ce
r
esu
lts
o
f
R
esNet
ar
e
ev
alu
ated
an
d
s
h
o
wn
in
th
e
Fig
u
r
e
s
4
an
d
5
.
I
t
is
ev
i
d
en
t
f
r
o
m
th
e
f
ig
u
r
es
th
at
th
e
r
e
is
a
s
p
ik
e
in
v
alid
atio
n
lo
s
s
an
d
d
r
o
p
i
n
ac
cu
r
ac
y
ar
o
u
n
d
e
p
o
ch
5
,
s
u
g
g
esti
n
g
in
s
tab
ilit
y
b
u
t,
d
esp
ite
th
is
,
th
e
m
o
d
el
r
ec
o
v
er
s
in
d
u
e
tim
e
b
o
asti
n
g
th
e
ab
ilit
y
to
o
v
er
co
m
e
b
r
ief
s
etb
ac
k
s
.
T
h
e
f
in
al
ep
o
ch
s
s
h
o
w
a
s
lig
h
t
g
ap
b
etwe
en
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
,
b
u
t
it
is
n
o
t
s
ig
n
if
ican
t
en
o
u
g
h
to
s
u
g
g
est
o
v
er
f
itti
n
g
.
Fig
u
r
e
6
d
e
m
o
n
s
tr
ates
th
e
ac
tu
al
d
etec
tio
n
a
n
d
c
lass
if
icatio
n
p
er
f
o
r
m
e
d
b
y
th
e
h
y
b
r
id
Yo
lo
v
8
an
d
R
esNet
m
o
d
el
o
n
im
ag
es h
av
i
n
g
B
r
aille
tex
t.
Fig
u
r
e
4
.
Gr
a
p
h
s
co
n
tain
i
n
g
lo
s
s
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
an
d
F1
s
co
r
e
o
f
R
esNet
m
o
d
el
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.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
6
5
2
-
4
6
6
0
4656
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
R
esNet
Fig
u
r
e
6
.
W
o
r
k
i
n
g
o
f
B
r
aille
lan
g
u
ag
e
class
if
icatio
n
m
o
d
el
f
o
r
im
ag
es u
s
in
g
R
esNet
3.
3
.
DenseNet
T
h
e
p
er
f
o
r
m
a
n
ce
r
esu
lt
o
f
De
n
s
eNe
t m
o
d
el
is
ev
alu
ated
an
d
s
h
o
wn
in
th
e
Fig
u
r
e
7
.
W
e
ca
n
in
f
er
th
at
th
e
tr
ain
in
g
p
r
o
ce
s
s
s
h
o
ws
s
ig
n
if
ican
t
in
s
tab
ilit
y
,
with
lar
g
e
f
lu
ctu
atio
n
s
in
v
alid
atio
n
l
o
s
s
an
d
ac
cu
r
ac
y
b
u
t,
th
is
is
ev
en
tu
ally
co
n
v
er
g
ed
to
a
h
ig
h
-
p
e
r
f
o
r
m
an
ce
s
tate.
T
h
er
e
is
a
n
o
ticea
b
le
g
ap
b
et
wee
n
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
in
t
h
e
later
ep
o
ch
s
,
s
u
g
g
esti
n
g
o
v
er
f
itti
n
g
b
u
t,
it is
n
o
t sev
e
r
ely
im
p
ac
t
in
g
g
en
e
r
aliza
tio
n
.
Fig
u
r
e
7
.
Gr
a
p
h
s
co
n
tain
i
n
g
lo
s
s
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
an
d
F1
s
co
r
e
o
f
Den
s
eNe
t 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
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es fo
r
B
r
a
ille d
etec
tio
n
a
n
d
cla
s
s
ifica
tio
n
:
co
m
p
a
r
a
tive
a
n
a
lysi
s
(
S
u
r
ek
h
a
Ja
n
r
a
o
)
4657
3.
4
.
M
o
bil
eNe
t
V2
T
h
e
Mo
b
ileNetV2
m
o
d
el
d
em
o
n
s
tr
ated
r
ea
s
o
n
a
b
le
lear
n
in
g
p
r
o
g
r
ess
,
with
d
ec
r
ea
s
in
g
lo
s
s
d
esp
ite
lo
w
ac
cu
r
ac
y
,
as
s
h
o
wn
in
Fig
u
r
e
8
.
Ho
wev
er
,
a
s
ig
n
if
ican
t
n
u
m
b
er
o
f
m
is
class
if
icatio
n
s
ar
e
ev
id
en
t
i
n
th
e
co
n
f
u
s
io
n
m
atr
ix
. T
h
is
cr
itically
im
p
ac
ts
th
e
m
o
d
el'
s
r
ea
l
-
wo
r
ld
ap
p
licab
ilit
y
.
Fig
u
r
e
8
.
Gr
a
p
h
s
co
n
tain
i
n
g
lo
s
s
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
&
F1
s
co
r
e
o
f
Mo
b
ileNetV2
m
o
d
el
3.
5
.
ResNeXt
T
h
e
p
e
r
f
o
r
m
an
ce
r
esu
lt
o
f
De
n
s
eNe
t
m
o
d
el
is
e
v
alu
ated
an
d
s
h
o
wn
in
th
e
Fig
u
r
e
9
.
As
we
ca
n
s
ee
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
s
h
o
ws
a
s
m
o
o
th
lear
n
in
g
c
u
r
v
e
with
b
o
t
h
tr
ain
in
g
an
d
v
alid
atio
n
l
o
s
s
d
ec
r
ea
s
in
g
s
tead
ily
.
Als
o
,
u
n
lik
e
th
e
Den
s
eNe
t
m
o
d
el,
th
is
m
o
d
el
s
h
o
ws
s
tab
le
lear
n
in
g
with
o
u
t
d
r
am
a
tic
f
lu
ctu
atio
n
s
i
n
p
er
f
o
r
m
an
ce
.
Ov
er
all,
it
s
h
o
w
s
m
o
r
e
c
o
n
s
is
ten
t
p
er
f
o
r
m
an
c
e
ac
r
o
s
s
ep
o
c
h
s
co
m
p
ar
ed
to
Den
s
eNe
t,
b
u
t
with
lo
wer
o
v
er
all
ac
c
u
r
ac
y
.
Fig
u
r
e
9
.
Gr
a
p
h
s
co
n
tain
i
n
g
lo
s
s
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
F1
s
co
r
e
o
f
R
esNeXt
m
o
d
el
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.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
6
5
2
-
4
6
6
0
4658
3.
6
.
Dis
cus
s
io
n
W
e
h
av
e
c
o
m
p
ar
e
d
th
e
p
e
r
f
o
r
m
an
ce
o
f
all
th
e
m
o
d
els
u
n
d
er
co
n
s
id
er
atio
n
i
n
T
ab
le
1
.
B
ased
o
n
th
a
t
we
h
av
e
in
f
er
r
e
d
th
at
R
esNet
a
n
d
Den
s
eNe
t
d
em
o
n
s
tr
ated
s
u
p
er
io
r
p
e
r
f
o
r
m
an
ce
,
with
h
ig
h
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
es,
in
d
icatin
g
ef
f
ec
tiv
e
lear
n
i
n
g
an
d
g
en
er
a
lizatio
n
ca
p
ab
ilit
ies.
On
e
o
f
th
e
m
ajo
r
lim
itatio
n
s
th
at
wer
e
f
ac
ed
d
u
r
in
g
o
u
r
r
esear
ch
was
th
e
lack
o
f
h
ig
h
-
q
u
ality
d
ataset
lead
in
g
to
s
u
b
-
o
p
tim
al
d
etec
tio
n
f
o
r
r
ea
l
wo
r
ld
im
ag
es.
I
n
a
p
r
ac
t
ical
ap
p
licatio
n
th
is
lim
itatio
n
will
n
ee
d
to
b
e
ad
d
r
ess
ed
b
y
cr
ea
tin
g
a
n
ew
ex
ten
s
iv
e
d
ataset
th
at
wo
u
ld
wo
r
k
b
etter
in
r
ea
l
w
o
r
ld
c
o
n
d
itio
n
s
.
Desp
ite
th
is
C
NN
wi
th
p
o
o
lin
g
s
h
o
wed
b
alan
ce
d
p
e
r
f
o
r
m
an
ce
a
n
d
w
ill
b
en
ef
it
f
r
o
m
f
u
r
th
er
t
r
ain
i
n
g
.
Mo
b
ileNetV2
ex
h
ib
ited
s
ig
n
if
ican
tly
lo
wer
ac
cu
r
ac
y
an
d
p
r
e
cisi
o
n
,
s
u
g
g
e
s
tin
g
lim
itatio
n
s
in
its
lig
h
twei
g
h
t
ar
ch
itectu
r
e
f
o
r
co
m
p
le
x
i
m
ag
e
class
if
icatio
n
task
s
.
I
n
co
m
p
ar
is
o
n
to
Ad
r
u
e
t
a
l.
[
1
7
]
,
wh
ich
ex
p
lo
r
ed
m
o
d
els
lik
e
R
e
s
Net5
0
,
Xce
p
tio
n
,
a
n
d
Sq
u
ee
ze
Net,
o
u
r
wo
r
k
f
o
c
u
s
es
o
n
C
NN
with
p
o
o
lin
g
,
R
esNet,
Den
s
eNe
t,
an
d
Mo
b
ileNetV2
.
W
h
ile
th
e
r
ef
er
en
ce
p
ap
er
o
b
s
er
v
e
d
a
h
ig
h
ac
c
u
r
ac
y
o
f
9
4
.
5
5
%
u
s
in
g
a
cu
s
to
m
C
NN,
o
u
r
R
esNet
an
d
Den
s
eNe
t
m
o
d
els
s
h
o
wed
b
etter
ac
cu
r
ac
y
an
d
g
en
e
r
aliza
tio
n
,
with
f
ewe
r
in
s
tan
ce
s
o
f
in
s
tab
ilit
y
ac
r
o
s
s
ep
o
ch
s
th
er
eb
y
s
h
o
win
g
h
i
g
h
p
o
ten
tial
f
o
r
B
r
aille
ch
ar
ac
ter
r
ec
o
g
n
itio
n
in
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
.
R
esNeXt
ac
h
iev
ed
r
esp
ec
tab
le
p
er
f
o
r
m
a
n
c
e,
b
u
t
s
lig
h
tly
lo
wer
co
m
p
ar
ed
t
o
R
es
Net
an
d
D
en
s
eNe
t,
in
d
icatin
g
p
o
ten
tial
f
o
r
im
p
r
o
v
em
en
t
with
ad
d
itio
n
al
tr
ain
in
g
o
r
ar
ch
itectu
r
al
m
o
d
if
icatio
n
s
.
T
ab
le
1.
C
o
m
p
a
r
is
o
n
o
f
m
o
d
el
s
u
s
ed
f
o
r
B
r
aille
c
lass
if
icatio
n
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
C
N
N
w
i
t
h
p
o
o
l
i
n
g
0
.
8
7
8
2
0
.
8
8
0
3
0
.
8
7
7
3
0
.
8
7
2
R
e
sN
e
t
0
.
9
4
8
7
0
.
9
5
5
4
0
.
9
4
5
4
0
.
9
4
8
D
e
n
seN
e
t
0
.
9
6
4
7
0
.
9
7
1
5
0
.
9
6
6
3
0
.
9
6
6
M
o
b
i
l
e
N
e
t
V
2
0
.
2
1
4
7
0
.
2
9
1
1
0
.
2
1
1
1
0
.
2
1
2
R
e
sN
e
X
t
0
.
8
6
8
6
0
.
9
0
5
4
0
.
8
6
6
7
0
.
8
6
3
4.
CO
NCLU
SI
O
N
Ou
r
s
tu
d
y
th
o
r
o
u
g
h
l
y
ex
am
in
e
d
an
d
ev
alu
ated
th
e
p
e
r
f
o
r
m
an
ce
o
f
s
ev
er
al
d
ee
p
lear
n
in
g
ar
c
h
itectu
r
es
in
clu
d
in
g
C
NNs
with
p
o
o
lin
g
,
R
esNet,
Den
s
e
Net,
Mo
b
ileN
etV2
,
an
d
R
esNeXt.
W
e
lev
er
ag
ed
th
e
YOL
Ov
8
o
b
ject
d
etec
tio
n
m
o
d
el
f
o
r
e
f
f
icien
t
B
r
aille
ch
ar
ac
ter
d
etec
tio
n
f
r
o
m
im
a
g
es
wh
ich
is
th
en
f
o
llo
wed
b
y
th
e
ap
p
licatio
n
o
f
d
if
f
er
en
t
d
ee
p
l
ea
r
n
in
g
m
o
d
el
f
o
r
th
e
class
if
icatio
n
o
f
ch
ar
ac
te
r
s
.
T
h
e
co
m
p
ar
ativ
e
an
aly
s
is
o
f
th
e
d
ee
p
lear
n
in
g
m
o
d
els
h
a
s
b
r
o
u
g
h
t
u
s
to
th
e
co
n
cl
u
s
io
n
th
at
th
e
R
esNet
an
d
De
n
s
eNe
t
m
o
d
els
h
av
e
o
u
tp
er
f
o
r
m
ed
o
t
h
er
m
o
d
els
b
y
d
em
o
n
s
tr
atio
n
h
ig
h
er
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
an
d
F1
-
s
co
r
es.
T
h
ese
ar
ch
itectu
r
es
in
v
o
l
v
e
d
e
n
s
ely
c
o
n
n
ec
ted
la
y
er
s
wh
ich
e
n
ab
les
ef
f
ec
tiv
e
lear
n
in
g
an
d
g
e
n
er
ali
za
tio
n
ca
p
ab
ilit
ies.
Su
ch
f
ea
tu
r
es
ar
e
ess
en
tial
f
o
r
ca
p
tu
r
in
g
t
h
e
in
tr
icac
ies
an
d
s
p
atial
ar
r
an
g
em
en
t
o
f
B
r
aille
ch
ar
ac
ter
s
.
On
th
e
o
th
er
h
a
n
d
,
th
e
C
NN
with
p
o
o
lin
g
m
o
d
el
h
as
a
m
e
d
io
cr
e
p
er
f
o
r
m
an
ce
in
d
ictin
g
its
p
o
ten
tial
f
o
r
B
r
aille
class
if
icatio
n
task
h
o
wev
er
,
t
h
e
Mo
b
ileNetV2
’
s
lig
h
t
-
wei
g
h
t
ar
ch
itectu
r
e
an
d
d
esig
n
f
o
r
ef
f
icien
cy
d
id
n
o
t
p
er
f
o
r
m
u
p
to
th
e
m
ar
k
.
R
esNeXt
ar
ch
itectu
r
e
p
er
f
o
r
m
ed
w
ell
b
u
t
s
lig
h
tly
less
th
an
R
esNet
an
d
Den
s
eNe
t,
it
h
o
wev
er
s
till
h
o
ld
s
p
r
o
m
is
e
f
o
r
f
u
tu
r
e
a
r
ch
itectu
r
al
o
p
tim
izat
io
n
s
o
r
en
s
em
b
le
tec
h
n
iq
u
es
th
at
co
u
ld
p
o
te
n
tially
en
h
an
ce
its
p
e
r
f
o
r
m
an
ce
.
Ou
r
cu
r
r
en
t
r
esear
ch
is
s
o
lely
f
o
cu
s
ed
o
n
tr
a
n
s
latio
n
o
f
in
d
i
v
id
u
a
l
B
r
aille
ce
ll
in
to
a
n
E
n
g
lis
h
letter
,
letter
b
y
letter
.
Fu
tu
r
e
wo
r
k
co
u
ld
ex
p
lo
r
e
th
e
r
ec
o
g
n
itio
n
an
d
tr
an
s
latio
n
o
f
co
n
tr
ac
ted
B
r
aille,
wh
er
e
en
tire
wo
r
d
s
o
r
co
m
m
o
n
letter
co
m
b
in
atio
n
s
ar
e
r
ep
r
e
s
en
ted
b
y
s
in
g
le
B
r
aille
ce
lls
.
E
ac
h
co
u
n
t
r
y
h
as
its
o
wn
B
r
aille
co
d
e
,
o
u
r
r
esear
ch
co
n
s
id
er
s
Un
if
ied
E
n
g
lis
h
B
r
aille
(
E
B
U)
c
o
d
e
,
f
u
tu
r
e
wo
r
k
c
o
u
ld
in
v
o
lv
e
wo
r
k
in
g
o
n
o
th
er
B
r
aille
co
d
es.
Fu
r
th
er
m
o
r
e
,
th
e
b
est
-
p
e
r
f
o
r
m
in
g
m
o
d
els
(
R
esNet
an
d
Den
s
eNe
t)
co
u
ld
b
e
o
p
tim
ized
f
o
r
m
o
b
ile
d
ev
ices,
en
ab
lin
g
r
ea
l
-
tim
e
B
r
aille
d
etec
tio
n
an
d
tr
a
n
s
latio
n
o
n
s
m
ar
tp
h
o
n
es.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
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
Su
r
ek
h
a
J
an
r
a
o
✓
✓
✓
✓
✓
✓
✓
✓
T
av
io
n
Fer
n
a
n
d
es
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ojas Go
latk
ar
✓
✓
✓
✓
✓
✓
✓
✓
✓
Swar
aj
Du
s
an
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
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es fo
r
B
r
a
ille d
etec
tio
n
a
n
d
cla
s
s
ifica
tio
n
:
co
m
p
a
r
a
tive
a
n
a
lysi
s
(
S
u
r
ek
h
a
Ja
n
r
a
o
)
4659
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
a
u
t
h
o
r
s
c
o
n
f
i
r
m
t
h
a
t
t
h
e
d
at
a
s
u
p
p
o
r
t
i
n
g
t
h
e
f
i
n
d
i
n
g
s
o
f
t
h
i
s
s
t
u
d
y
a
r
e
a
v
a
i
la
b
l
e
wi
t
h
i
n
t
h
e
a
r
t
i
c
le
.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
A
l
m
u
k
a
i
n
z
i
,
A
.
A
l
m
u
h
a
r
e
b
,
F
.
A
l
d
w
i
san
,
a
n
d
W
.
A
l
q
u
a
y
d
h
i
b
,
“
M
e
d
i
c
a
t
i
o
n
u
s
e
p
a
t
t
e
r
n
s
i
n
t
h
e
v
i
s
u
a
l
l
y
i
m
p
a
i
r
e
d
i
n
S
a
u
d
i
A
r
a
b
i
a
a
n
d
t
h
e
i
m
p
o
r
t
a
n
c
e
o
f
a
p
p
l
y
i
n
g
B
r
a
i
l
l
e
l
a
b
e
l
i
n
g
,
”
S
a
u
d
i
P
h
a
rm
a
c
e
u
t
i
c
a
l
J
o
u
rn
a
l
,
v
o
l
.
2
8
,
n
o
.
3
,
p
p
.
2
7
4
–
2
8
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
s
p
s.
2
0
2
0
.
0
1
.
0
0
6
.
[
2
]
W
.
P
o
r
n
s
u
k
v
i
t
o
o
n
,
A
.
N
i
m
k
o
o
m
p
a
i
,
a
n
d
P
.
S
i
r
i
k
o
n
g
t
h
a
m,
“
N
u
mer
i
c
a
l
i
m
a
g
e
s
c
l
a
ssi
f
i
c
a
t
i
o
n
f
o
r
B
r
a
i
l
l
e
w
i
t
h
s
e
r
v
o
m
o
t
o
r
s
u
s
i
n
g
O
C
R
t
e
c
h
n
i
q
u
e
,
”
i
n
2
0
2
2
7
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
B
u
si
n
e
ss
a
n
d
I
n
d
u
st
r
i
a
l
Re
se
a
rc
h
(
I
C
BI
R)
,
2
0
2
2
,
p
p
.
6
1
–
66
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
B
I
R
5
4
5
8
9
.
2
0
2
2
.
9
7
8
6
4
2
8
.
[
3
]
M
.
A
.
H
u
ss
a
i
n
,
R
.
I
.
R
i
f
a
t
,
S
.
B
.
I
q
b
a
l
,
S
.
B
i
sw
a
s,
M
.
G
.
R
a
b
i
u
l
A
l
a
m,
a
n
d
M
.
T.
R
e
z
a
,
“
D
e
e
p
l
e
a
r
n
i
n
g
b
a
se
d
b
a
n
g
l
a
v
o
i
c
e
t
o
B
r
a
i
l
l
e
c
h
a
r
a
c
t
e
r
c
o
n
v
e
r
si
o
n
s
y
st
e
m,
”
i
n
2
0
2
2
I
EEE
1
3
t
h
A
n
n
u
a
l
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
El
e
c
t
ro
n
i
c
s
a
n
d
M
o
b
i
l
e
C
o
m
m
u
n
i
c
a
t
i
o
n
C
o
n
f
e
re
n
c
e
(
I
EM
C
O
N
)
,
2
0
2
2
,
p
p
.
2
6
2
–
2
6
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
EM
C
O
N
5
6
8
9
3
.
2
0
2
2
.
9
9
4
6
6
1
9
.
[
4
]
F
.
F
a
i
sa
l
,
M
.
H
a
s
a
n
,
S
.
S
a
b
r
i
n
,
M
.
Z.
H
a
san
,
a
n
d
A
.
H
.
S
i
d
d
i
q
u
e
,
“
V
o
i
c
e
a
c
t
i
v
a
t
e
d
p
o
r
t
a
b
l
e
B
r
a
i
l
l
e
w
i
t
h
a
u
d
i
o
f
e
e
d
b
a
c
k
,
”
i
n
2
0
2
1
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ro
b
o
t
i
c
s,
E
l
e
c
t
ri
c
a
l
a
n
d
S
i
g
n
a
l
Pr
o
c
e
s
si
n
g
T
e
c
h
n
i
q
u
e
s
(
I
C
RE
S
T
)
,
2
0
2
1
,
p
p
.
4
1
8
–
4
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
R
EST
5
1
5
5
5
.
2
0
2
1
.
9
3
3
1
0
0
4
.
[
5
]
S
.
R
a
ma
c
h
a
n
d
r
a
n
,
D
.
G
u
r
u
r
a
j
,
K
.
N
.
P
a
l
l
a
v
i
,
a
n
d
N
.
R
a
j
a
n
,
“
T
e
x
t
t
o
B
r
a
i
l
l
e
c
o
n
v
e
r
t
i
n
g
c
o
mm
u
n
i
c
a
t
i
o
n
d
e
v
i
c
e
f
o
r
t
h
e
v
i
s
u
a
l
a
n
d
h
e
a
r
i
n
g
i
m
p
a
i
r
e
d
p
e
r
s
o
n
s
,
”
i
n
2
0
2
1
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
I
n
f
o
rm
a
t
i
c
s
(
I
C
C
C
I
)
,
2
0
2
1
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
C
I
5
0
8
2
6
.
2
0
2
1
.
9
4
0
2
5
9
0
.
[
6
]
S
.
S
r
i
j
a
,
P
.
K
a
w
y
a
,
T.
A
.
R
e
d
d
y
,
a
n
d
M
.
D
h
a
n
a
l
a
k
s
h
m
i
,
“
R
a
s
p
b
e
r
r
y
P
i
b
a
sed
w
e
a
r
a
b
l
e
r
e
a
d
e
r
f
o
r
v
i
s
u
a
l
l
y
i
m
p
a
i
r
e
d
p
e
o
p
l
e
w
i
t
h
h
a
p
t
i
c
f
e
e
d
b
a
c
k
,
”
i
n
2
0
2
0
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
E
l
e
c
t
r
o
n
i
c
s
a
n
d
S
u
s
t
a
i
n
a
b
l
e
C
o
m
m
u
n
i
c
a
t
i
o
n
S
y
s
t
e
m
s
(
I
C
ES
C
)
,
2
0
2
0
,
p
p
.
2
5
7
–
260
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
ESC4
8
9
1
5
.
2
0
2
0
.
9
1
5
6
0
0
5
.
[
7
]
A
.
K
a
r
me
l
,
A
.
S
h
a
r
ma
,
M
.
P
a
n
d
y
a
,
a
n
d
D
.
G
a
r
g
,
“
I
o
T
b
a
se
d
a
ssi
st
i
v
e
d
e
v
i
c
e
f
o
r
d
e
a
f
,
d
u
mb
a
n
d
b
l
i
n
d
p
e
o
p
l
e
,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
6
5
,
p
p
.
2
5
9
–
2
6
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.
2
0
2
0
.
0
1
.
0
8
0
.
[
8
]
A
.
D
e
y
,
S
.
B
i
sw
a
s,
a
n
d
D
.
-
N
.
Le
,
“
R
e
c
o
g
n
i
t
i
o
n
o
f
W
h
-
q
u
e
st
i
o
n
si
g
n
g
e
s
t
u
r
e
s
i
n
v
i
d
e
o
s
t
r
e
a
ms
u
si
n
g
a
n
a
t
t
e
n
t
i
o
n
d
r
i
v
e
n
C
3
D
-
B
i
LST
M
n
e
t
w
o
r
k
,
”
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
.
2
9
2
0
–
2
9
3
1
,
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
.
2
7
6
.
[
9
]
G
.
W
a
n
g
,
Y
.
G
u
,
W
.
W
a
n
g
,
Z
.
Y
u
,
a
n
d
S
.
S
o
n
g
,
“
B
r
a
i
l
l
e
d
e
t
e
c
t
i
o
n
m
o
d
e
l
b
a
se
d
o
n
f
o
r
e
g
r
o
u
n
d
a
t
t
e
n
t
i
o
n
a
n
d
sem
a
n
t
i
c
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
3
I
EE
E
5
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
ren
c
e
o
n
P
o
w
e
r,
I
n
t
e
l
l
i
g
e
n
t
C
o
m
p
u
t
i
n
g
a
n
d
S
y
s
t
e
m
s
(
I
C
PI
C
S
)
,
2
0
2
3
,
p
p
.
8
7
–
92
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
P
I
C
S
5
8
3
7
6
.
2
0
2
3
.
1
0
2
3
5
7
1
0
.
[
1
0
]
P
.
P
o
t
d
a
r
,
D
.
H
a
r
d
ma
n
,
E.
A
l
m
a
n
z
o
r
,
a
n
d
F
.
I
i
d
a
,
“
H
i
g
h
-
s
p
e
e
d
t
a
c
t
i
l
e
B
r
a
i
l
l
e
r
e
a
d
i
n
g
v
i
a
b
i
o
m
i
me
t
i
c
s
l
i
d
i
n
g
i
n
t
e
r
a
c
t
i
o
n
s
,
”
I
EE
E
Ro
b
o
t
i
c
s
a
n
d
Au
t
o
m
a
t
i
o
n
L
e
t
t
e
rs
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
2
6
1
4
–
2
6
2
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
L
R
A
.
2
0
2
4
.
3
3
5
6
9
7
8
.
[
1
1
]
C
.
C
h
e
l
l
a
sw
a
my
,
T.
S
.
G
e
e
t
h
a
,
K
.
H
a
r
i
h
a
r
a
n
,
A
.
D
.
R
a
j
,
K
.
A
r
c
h
a
n
a
,
a
n
d
S
.
B
a
b
i
t
h
a
r
a
n
i
,
“
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
B
r
a
i
l
l
e
t
e
c
h
n
o
l
o
g
y
f
o
r
v
i
s
u
a
l
a
n
d
h
e
a
r
i
n
g
i
m
p
a
i
r
e
d
p
e
o
p
l
e
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
S
m
a
r
t
S
y
st
e
m
s
f
o
r
a
p
p
l
i
c
a
t
i
o
n
s
i
n
El
e
c
t
r
i
c
a
l
S
c
i
e
n
c
e
s
(
I
C
S
S
ES
)
,
2
0
2
3
,
p
p
.
1
–
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
S
S
ES5
8
2
9
9
.
2
0
2
3
.
1
0
1
9
9
9
3
5
.
[
1
2
]
T.
K
a
u
s
a
r
,
S
.
M
a
n
z
o
o
r
,
A
.
K
a
u
sar
,
Y
.
L
u
,
M
.
W
a
s
i
f
,
a
n
d
M
.
A
.
A
sh
r
a
f
,
“
D
e
e
p
l
e
a
r
n
i
n
g
s
t
r
a
t
e
g
y
f
o
r
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
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
9
,
p
p
.
1
6
9
3
5
7
–
1
6
9
3
7
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
1
3
8
2
4
0
.
[
1
3
]
R
.
P
r
a
t
h
i
p
a
,
M
.
A
r
u
n
,
M
.
P
r
e
mK
u
m
a
r
,
S
.
S
a
t
h
i
y
a
p
r
i
y
a
,
a
n
d
P
.
S
r
i
d
e
v
i
,
“
C
N
N
b
a
se
d
p
e
r
so
n
a
l
a
ss
i
st
i
v
e
s
y
st
e
m
f
o
r
d
e
a
f
-
b
l
i
n
d
,
”
i
n
2
0
2
2
4
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
i
rc
u
i
t
s,
C
o
n
t
r
o
l
,
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
C
o
m
p
u
t
i
n
g
(
I
4
C
)
,
2
0
2
2
,
p
p
.
2
2
1
–
2
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
4
C
5
7
1
4
1
.
2
0
2
2
.
1
0
0
5
7
8
8
2
.
[
1
4
]
S
.
R
a
mac
h
a
n
d
r
a
n
,
N
.
R
a
j
a
n
,
K
.
N
.
P
a
l
l
a
v
i
,
J.
S
u
b
a
s
h
r
e
e
,
S
.
S
u
c
h
i
t
h
r
a
,
a
n
d
B
.
S
o
n
a
l
,
“
C
o
mm
u
n
i
c
a
t
i
o
n
d
e
v
i
c
e
f
o
r
t
h
e
v
i
su
a
l
a
n
d
h
e
a
r
i
n
g
i
mp
a
i
r
e
d
p
e
r
s
o
n
s
t
o
c
o
n
v
e
r
t
b
r
a
i
l
l
e
c
h
a
r
a
c
t
e
r
s
t
o
E
n
g
l
i
s
h
t
e
x
t
,
”
i
n
2
0
2
1
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Em
e
rg
i
n
g
S
m
a
r
t
C
o
m
p
u
t
i
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s (E
S
C
I
)
,
2
0
2
1
,
p
p
.
5
8
7
–
5
9
2
,
d
o
i
:
1
0
.
1
1
0
9
/
ESCI
5
0
5
5
9
.
2
0
2
1
.
9
3
9
6
8
5
9
.
[
1
5
]
C
.
X
u
,
“
A
p
p
l
y
i
n
g
M
LP
a
n
d
C
N
N
o
n
h
a
n
d
w
r
i
t
i
n
g
i
ma
g
e
s
f
o
r
i
ma
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
t
a
sk
,
”
i
n
2
0
2
2
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ad
v
a
n
c
e
d
E
l
e
c
t
r
o
n
i
c
Ma
t
e
r
i
a
l
s,
C
o
m
p
u
t
e
rs
a
n
d
S
o
f
t
w
a
r
e
E
n
g
i
n
e
e
ri
n
g
(
A
EM
C
S
E)
,
2
0
2
2
,
p
p
.
8
3
0
–
835
,
d
o
i
:
1
0
.
1
1
0
9
/
A
E
M
C
S
E
5
5
5
7
2
.
2
0
2
2
.
0
0
1
6
7
.
[
1
6
]
H
.
M
.
N
a
si
r
,
N
.
M
.
A
.
B
r
a
h
i
n
,
M
.
M
.
M
.
A
m
i
n
u
d
d
i
n
,
M
.
S
.
M
i
s
p
a
n
,
a
n
d
M
.
F
.
Zu
l
k
i
f
l
i
,
“
A
n
d
r
o
i
d
b
a
se
d
a
p
p
l
i
c
a
t
i
o
n
f
o
r
v
i
s
u
a
l
l
y
i
mp
a
i
r
e
d
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
I
AE
S
I
n
t
e
rn
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
(
I
J
-
AI
)
,
v
o
l
.
1
0
,
n
o
.
4
,
p
p
.
8
7
9
–
8
8
8
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
0
.
i
4
.
p
p
8
7
9
-
8
8
8
.
[
1
7
]
S
.
L
.
A
d
r
u
,
S
.
Jo
h
n
so
n
,
a
n
d
M
.
H
e
m
a
l
a
t
h
a
,
“
C
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
C
N
N
mo
d
e
l
s
f
o
r
b
r
a
i
l
l
e
c
h
a
r
a
c
t
e
r
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
M
o
b
i
l
e
,
I
n
t
e
l
l
i
g
e
n
t
,
a
n
d
U
b
i
q
u
i
t
o
u
s
C
o
m
p
u
t
i
n
g
C
o
n
f
e
re
n
c
e
(
MIU
C
C
)
,
2
0
2
3
,
p
p
.
1
–
7
,
d
o
i
:
1
0
.
1
1
0
9
/
M
I
U
C
C
5
8
8
3
2
.
2
0
2
3
.
1
0
2
7
8
3
2
1
.
[
1
8
]
T.
O
k
a
m
o
t
o
,
T.
S
h
i
mo
n
o
,
Y
.
Ts
u
b
o
i
,
M
.
I
z
u
m
i
,
a
n
d
Y
.
T
a
k
a
n
o
,
“
B
r
a
i
l
l
e
b
l
o
c
k
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
n
d
g
u
i
d
e
f
o
r
v
i
s
u
a
l
l
y
i
mp
a
i
r
e
d
p
e
o
p
l
e
,
”
i
n
2
0
2
0
I
E
EE
2
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
S
y
m
p
o
si
u
m
o
n
I
n
d
u
s
t
ri
a
l
E
l
e
c
t
r
o
n
i
c
s
(
I
S
I
E)
,
2
0
2
0
,
p
p
.
4
8
7
–
492
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
I
E4
5
0
6
3
.
2
0
2
0
.
9
1
5
2
5
7
6
.
[
1
9
]
S
.
K
u
l
k
a
r
n
i
a
n
d
S
.
H
a
r
n
o
o
r
k
a
r
,
“
C
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
Re
s
e
a
r
c
h
J
o
u
r
n
a
l
o
f
En
g
i
n
e
e
ri
n
g
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
7
,
n
o
.
6
,
p
p
.
1
4
5
9
–
1
4
6
4
,
2
0
2
0
.
[
2
0
]
W
.
R
a
w
a
t
a
n
d
Z.
W
a
n
g
,
“
D
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
i
m
a
g
e
c
l
a
s
si
f
i
c
a
t
i
o
n
:
a
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
,
”
N
e
u
r
a
l
C
o
m
p
u
t
a
t
i
o
n
,
v
o
l
.
2
9
,
n
o
.
9
,
p
p
.
2
3
5
2
–
2
4
4
9
,
2
0
1
7
,
d
o
i
:
1
0
.
1
1
6
2
/
N
EC
O
_
a
_
0
0
9
9
0
.
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.
14
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
6
5
2
-
4
6
6
0
4660
[
2
1
]
A
.
B
.
D
e
sai
,
D
.
R
.
G
a
n
g
o
d
k
a
r
,
B
.
P
a
n
t
,
a
n
d
K
.
P
a
n
t
,
“
C
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
si
s
u
si
n
g
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
m
o
d
e
l
s VG
G
1
6
,
R
e
s
n
e
t
5
0
a
n
d
X
c
e
p
t
i
o
n
t
o
p
r
e
d
i
c
t
p
n
e
u
mo
n
i
a
,
”
i
n
2
0
2
2
2
n
d
I
n
t
e
rn
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
C
o
m
p
u
t
a
t
i
o
n
a
l
T
e
c
h
n
o
l
o
g
i
e
s
(
C
I
S
C
T
)
,
2
0
2
2
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
I
S
C
T5
5
3
1
0
.
2
0
2
2
.
1
0
0
4
6
5
0
7
.
[
2
2
]
S
.
M
.
B
a
d
a
w
y
,
A
.
E.
-
N
.
A
.
M
o
h
a
me
d
,
A
.
A
.
H
e
f
n
a
w
y
,
H
.
E.
Zi
d
a
n
,
M
.
T.
G
a
d
A
l
l
a
h
,
a
n
d
G
.
M
.
El
-
B
a
n
b
y
,
“
C
l
a
ssi
f
i
c
a
t
i
o
n
o
f
b
r
e
a
st
u
l
t
r
a
s
o
u
n
d
i
ma
g
e
s
b
a
se
d
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
-
a
c
o
m
p
a
r
a
t
i
v
e
st
u
d
y
,
”
i
n
2
0
2
1
I
n
t
e
r
n
a
t
i
o
n
a
l
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
s
C
o
n
f
e
re
n
c
e
(
I
T
C
-
E
g
y
p
t
)
,
2
0
2
1
,
p
p
.
1
–
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
TC
-
Eg
y
p
t
5
2
9
3
6
.
2
0
2
1
.
9
5
1
3
9
7
2
.
[
2
3
]
P
.
E
l
a
n
g
o
v
a
n
a
n
d
M
.
K
.
N
a
t
h
,
“
A
n
o
v
e
l
s
h
a
l
l
o
w
C
o
n
v
N
e
t
-
1
8
f
o
r
ma
l
a
r
i
a
p
a
r
a
s
i
t
e
d
e
t
e
c
t
i
o
n
i
n
Th
i
n
B
l
o
o
d
s
mea
r
i
ma
g
e
s,”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
2
,
n
o
.
5
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s4
2
9
7
9
-
0
2
1
-
0
0
7
6
3
-
w.
[
2
4
]
V
.
J
h
a
a
n
d
K
.
P
a
r
v
a
t
h
i
,
“
B
r
a
i
l
l
e
t
r
a
n
sl
i
t
e
r
a
t
i
o
n
o
f
H
i
n
d
i
h
a
n
d
w
r
i
t
t
e
n
t
e
x
t
s
u
si
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
f
o
r
c
h
a
r
a
c
t
e
r
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
f
S
c
i
e
n
t
i
f
i
c
a
n
d
T
e
c
h
n
o
l
o
g
y
Re
se
a
r
c
h
,
v
o
l
.
8
,
n
o
.
1
0
,
p
p
.
1
1
8
8
–
1
1
9
3
,
2
0
1
9
.
[
2
5
]
T.
H
u
a
n
g
e
t
a
l
.
,
“
Tr
a
n
s
l
a
t
i
n
g
B
r
a
i
l
l
e
i
n
t
o
C
h
i
n
e
se
b
a
se
d
o
n
i
m
p
r
o
v
e
d
C
B
H
G
mo
d
e
l
,
”
D
i
sp
l
a
y
s
,
v
o
l
.
7
8
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
i
sp
l
a
.
2
0
2
3
.
1
0
2
4
4
5
.
[
2
6
]
C
.
L
i
,
“
B
r
a
i
l
l
e
d
a
t
a
s
e
t
f
o
r
sce
n
e
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
K
a
g
g
l
e
.
2
0
2
1
.
A
c
c
e
ss
e
d
:
F
e
b
.
5
,
2
0
2
4
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
se
t
s/
c
h
a
n
g
j
i
a
n
l
i
/
b
r
a
i
l
l
e
-
d
a
t
a
s
e
t
-
f
o
r
-
sc
e
n
e
-
t
e
x
t
-
r
e
c
o
g
n
i
t
i
o
n
[
2
7
]
U
.
K
u
m
a
r
a
v
e
l
a
n
,
“
B
r
a
i
l
l
e
c
h
a
r
a
c
t
e
r
d
a
t
a
se
t
,
”
K
a
g
g
l
e
.
2
0
1
9
.
A
c
c
e
ss
e
d
:
F
e
b
.
1
6
,
2
0
2
2
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
se
t
s/
s
h
a
n
k
s
0
4
6
5
/
b
r
a
i
l
l
e
-
c
h
a
r
a
c
t
e
r
-
d
a
t
a
set
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Dr
.
S
u
r
e
k
h
a
J
a
n
r
a
o
e
m
p
lo
y
e
d
a
t
K.
J.
S
o
m
a
iy
a
In
stit
u
te
o
f
Te
c
h
n
o
l
o
g
y
a
s
a
n
As
sista
n
t
P
ro
fe
ss
o
r
.
S
h
e
h
a
s
b
e
e
n
a
tea
c
h
e
r
fo
r
six
tee
n
y
e
a
rs.
S
h
e
g
ra
d
u
a
ted
fro
m
M
u
m
b
a
i
Un
iv
e
rsity
with
a
b
a
c
h
e
lo
r'
s
a
n
d
m
a
ste
r'
s
d
e
g
re
e
in
c
o
m
p
u
ter
e
n
g
i
n
e
e
rin
g
.
Un
d
e
r
t
h
e
d
irec
ti
o
n
o
f
Dr.
De
v
e
n
S
h
a
h
,
sh
e
c
o
m
p
let
e
d
h
e
r
d
o
c
t
o
ra
te
in
m
a
c
h
in
e
lea
rn
in
g
a
n
d
t
h
e
in
tern
e
t
o
f
th
in
g
s.
S
h
e
h
a
s a
lmo
st
2
0
p
a
p
e
rs
p
u
b
li
sh
e
d
i
n
i
n
tern
a
ti
o
n
a
l
jo
u
rn
a
l
s.
S
h
e
is
in
tere
ste
d
in
d
a
t
a
m
in
in
g
,
m
a
c
h
in
e
lea
rn
in
g
,
a
rti
fi
c
ial
in
telli
g
e
n
c
e
,
a
n
d
t
h
e
in
ter
n
e
t
o
f
t
h
in
g
s
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
su
re
k
h
a
.
jan
ra
o
@s
o
m
a
iy
a
.
e
d
u
.
Ta
v
i
o
n
Fer
n
a
n
d
e
s
is
a
d
e
d
ica
ted
stu
d
e
n
t
p
u
rs
u
in
g
a
b
a
c
h
e
l
o
r'
s
d
e
g
re
e
in
Co
m
p
u
ter
E
n
g
in
e
e
rin
g
with
Ho
n
o
u
rs
in
Arti
ficia
l
In
telli
g
e
n
c
e
a
n
d
M
a
c
h
in
e
Lea
rn
i
n
g
a
t
K.
J.
S
o
m
a
iy
a
I
n
stit
u
te
o
f
Tec
h
n
o
l
o
g
y
.
He
is
a
c
ti
v
e
ly
e
n
g
a
g
e
d
in
re
se
a
rc
h
fo
c
u
sin
g
o
n
Bra
il
l
e
c
las
sifica
ti
o
n
u
sin
g
d
e
e
p
lea
rn
i
n
g
tec
h
n
iq
u
e
s.
He
is
v
e
r
y
p
a
ss
io
n
a
te
a
b
o
u
t
lev
e
ra
g
in
g
tec
h
n
o
l
o
g
y
f
o
r
so
c
ial
im
p
a
c
t
a
n
d
a
ims
to
m
a
k
e
a
d
v
a
n
c
e
m
e
n
ts
in
a
c
c
e
s
sib
il
it
y
fo
r
v
isu
a
ll
y
imp
a
ired
in
d
iv
id
u
a
ls.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
tav
io
n
.
f@so
m
a
iy
a
.
e
d
u
.
O
ja
s G
o
la
t
k
a
r
is
a
d
il
i
g
e
n
t
u
n
d
e
rg
ra
d
u
a
te
stu
d
e
n
t
p
u
rsu
i
n
g
a
Ba
c
h
e
lo
r'
s d
e
g
re
e
in
C
o
m
p
u
ter
En
g
in
e
e
ri
n
g
wi
th
Ho
n
o
u
rs
i
n
Art
ifi
c
ial
In
telli
g
e
n
c
e
a
n
d
M
a
c
h
in
e
Lea
rn
in
g
a
t
K.
J.
S
o
m
a
iy
a
In
sti
tu
te
o
f
Tec
h
n
o
lo
g
y
.
He
p
a
rti
c
i
p
a
ted
in
t
h
e
Bra
il
l
e
c
las
sifica
ti
o
n
a
n
d
a
c
n
e
d
e
tec
ti
o
n
&
c
las
sifica
ti
o
n
p
ro
jec
ts
,
wh
ich
we
re
b
o
th
p
iv
o
tal
m
o
m
e
n
ts
in
h
is
j
o
u
r
n
e
y
a
s
a
n
AI
a
n
d
M
L
e
x
p
e
rt.
Wi
th
a
k
e
e
n
in
tere
st
in
e
x
p
l
o
ri
n
g
i
n
n
o
v
a
ti
v
e
AI
b
a
se
d
so
lu
ti
o
n
s
t
o
c
o
m
m
o
n
p
ro
b
lem
s,
h
e
d
e
m
o
n
stra
tes
a
c
o
m
m
e
n
d
a
b
le
c
o
m
m
it
m
e
n
t
to
a
d
v
a
n
c
in
g
th
e
field
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
o
jas
.
g
@s
o
m
a
i
y
a
.
e
d
u
.
S
wa
r
a
j
Dus
a
n
e
is
a
c
o
m
m
it
ted
stu
d
e
n
t
e
n
r
o
ll
e
d
in
a
Ba
c
h
e
lo
r'
s
p
ro
g
ra
m
i
n
Co
m
p
u
ter
E
n
g
in
e
e
rin
g
with
Ho
n
o
u
rs
in
Arti
ficia
l
In
telli
g
e
n
c
e
a
n
d
M
a
c
h
in
e
Lea
rn
i
n
g
a
t
K.
J.
S
o
m
a
iy
a
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
.
His
re
se
a
rc
h
in
tere
sts
c
e
n
ter
a
ro
u
n
d
u
ti
li
z
i
n
g
d
e
e
p
lea
rn
in
g
m
e
th
o
d
s to
c
las
sify
Bra
il
le.
He
is d
e
e
p
ly
p
a
ss
io
n
a
te ab
o
u
t
h
a
rn
e
ss
in
g
tec
h
n
o
l
o
g
y
to
c
re
a
te
p
o
siti
v
e
so
c
ial
c
h
a
n
g
e
,
p
a
rti
c
u
larly
i
n
e
n
h
a
n
c
in
g
a
c
c
e
ss
ib
il
it
y
fo
r
th
e
v
isu
a
ll
y
im
p
a
ired
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
sw
a
ra
j.
d
u
sa
n
e
@s
o
m
a
iy
a
.
e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.