I
AE
S
I
nte
rna
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
.
3
,
J
u
n
e
2
0
2
5
,
p
p
.
1863
~
1
8
6
9
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
3
.
p
p
1
8
6
3
-
1
8
6
9
1863
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Dete
c
tion o
f
pa
r
ti
a
lly
occluded a
re
a
in
face ima
g
e us
ing
U
-
Ne
t
mo
del
J
y
o
t
hs
na
Chera
pa
na
m
j
er
i
1
,
B
.
Na
re
nd
ra
K
um
a
r
Ra
o
2
1
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
,
JN
TU
A
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
Jaw
a
h
a
r
l
a
l
N
e
h
r
u
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
,
A
n
a
n
t
h
a
p
u
r
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
S
c
h
o
o
l
o
f
C
o
mp
u
t
i
n
g
,
M
o
h
a
n
B
a
b
u
U
n
i
v
e
r
si
t
y
(
Er
st
w
h
i
l
e
S
r
e
e
V
i
d
y
a
n
i
k
e
t
h
a
n
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
)
,
Ti
r
u
p
a
t
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
O
c
t 2
1
,
2
0
2
4
R
ev
is
ed
Feb
1
5
,
2
0
2
5
Acc
ep
ted
Ma
r
1
5
,
2
0
2
5
Oc
c
lu
d
e
d
fa
c
e
re
c
o
g
n
it
io
n
is
imp
o
rtan
t
tas
k
in
c
o
m
p
u
ter v
isio
n
.
To
c
o
m
p
lete
th
e
o
c
c
lu
d
e
d
fa
c
e
re
c
o
g
n
it
io
n
e
fficie
n
tl
y
,
fi
rst
we
n
e
e
d
t
o
i
d
e
n
ti
fy
t
h
e
o
c
c
lu
d
e
d
re
g
i
o
n
i
n
fa
c
e
.
Id
e
n
t
i
fy
in
g
th
e
o
c
c
lu
d
e
d
re
g
io
n
in
f
a
c
e
is
a
c
h
a
ll
e
n
g
i
n
g
tas
k
in
c
o
m
p
u
ter
v
is
io
n
.
On
e
c
a
se
o
f
fa
c
e
o
c
c
lu
sio
n
is
n
o
t
h
i
n
g
b
u
t
we
a
rin
g
m
a
sk
s,
su
n
g
las
se
s,
a
n
d
sc
a
rv
e
s.
An
o
th
e
r
c
a
se
o
f
fa
c
e
o
c
c
lu
sio
n
is
fa
c
e
is
h
id
in
g
th
e
o
th
e
r
o
b
jec
t
s
li
k
e
b
o
o
k
s,
th
i
n
g
s,
o
r
o
th
e
r
fa
c
e
s.
In
o
u
r
re
se
a
rc
h
,
id
e
n
ti
f
y
in
g
t
h
e
o
c
c
lu
d
e
d
a
re
a
wh
ich
is
c
o
ro
n
a
v
iru
s d
ise
a
se
o
f
2
0
1
9
(
COV
ID
-
19
)
m
a
sk
e
d
a
re
a
in
f
a
c
e
a
n
d
g
e
n
e
ra
te
se
g
m
e
n
tati
o
n
m
a
p
.
In
se
m
a
n
ti
c
se
g
m
e
n
tatio
n
,
d
e
e
p
lea
rn
in
g
-
b
a
se
d
tec
h
n
i
q
u
e
s
h
a
v
e
d
e
m
o
n
stra
ted
p
ro
m
isin
g
o
u
tco
m
e
s.
W
e
h
a
v
e
e
m
p
lo
y
e
d
o
n
e
o
f
t
h
e
d
e
e
p
lea
r
n
in
g
-
b
a
se
d
U
-
Ne
t
m
o
d
e
ls
to
g
e
n
e
ra
te
a
b
in
a
ry
se
g
m
e
n
tatio
n
m
a
p
o
n
m
a
sk
e
d
re
g
io
n
o
f
a
h
u
m
a
n
fa
c
e
.
It
a
c
h
iev
e
s
re
li
a
b
le
p
e
rf
o
rm
a
n
c
e
a
n
d
re
d
u
c
i
n
g
n
e
two
r
k
c
o
m
p
lex
it
y
.
We
trai
n
o
u
r
m
o
d
e
l
o
n
M
a
sk
e
d
F
a
c
e
-
Ce
leb
A
d
a
tas
e
t
a
n
d
a
c
c
u
ra
c
y
is
9
7
.
7
%
.
Re
su
l
ts
fro
m
e
x
p
e
rime
n
ts
d
e
m
o
n
stra
te
th
a
t
,
i
n
c
o
m
p
a
riso
n
t
o
th
e
m
o
st
a
d
v
a
n
c
e
d
se
m
a
n
t
ic
s
e
g
m
e
n
tatio
n
m
o
d
e
ls,
o
u
r
a
p
p
ro
a
c
h
a
c
h
iev
e
s
a
p
r
o
m
isin
g
c
o
m
p
ro
m
ise
b
e
twe
e
n
se
g
m
e
n
tati
o
n
a
c
c
u
ra
c
y
a
n
d
c
o
m
p
u
ti
n
g
e
fficie
n
c
y
.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
C
o
m
p
u
ter
v
is
io
n
Dee
p
lear
n
in
g
I
m
ag
e
s
eg
m
en
tatio
n
Occ
lu
s
io
n
U
-
Net
m
o
d
el
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
:
J
y
o
th
s
n
a
C
h
er
ap
an
am
jer
i
D
e
p
a
r
t
m
e
n
t
o
f
C
o
m
p
u
t
e
r
S
c
ie
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
J
N
T
U
A
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
J
a
w
a
h
a
r
la
l
N
e
h
r
u
T
e
c
h
n
o
l
o
g
i
ca
l
U
n
i
v
e
r
s
i
t
y
A
n
a
n
t
h
a
p
u
r
,
I
n
d
i
a
E
m
ail:
jy
o
th
s
n
a_
5
1
3
@
y
ah
o
o
.
co
m
1.
I
NT
RO
D
UCT
I
O
N
W
h
en
it
co
m
es
to
d
etec
tin
g
o
cc
lu
d
ed
a
r
ea
s
in
f
ac
e
im
a
g
e
s
,
d
ee
p
lear
n
in
g
-
b
ased
tec
h
n
i
q
u
es
h
av
e
s
h
o
wn
ex
ce
llen
t
r
esu
lts
.
T
h
e
C
OVI
D
-
1
9
m
ask
ed
ar
ea
in
f
a
cial
im
ag
es
is
th
e
o
cc
lu
d
ed
ar
ea
in
o
u
r
r
esear
ch
.
No
wad
ay
s
m
ask
s
ar
e
u
s
ed
f
o
r
d
if
f
er
e
n
t
p
u
r
p
o
s
es
s
u
ch
as
C
OVI
D
-
1
9
p
a
n
d
em
ic
wid
esp
r
e
ad
u
s
e
o
f
f
ac
e
m
ask
s
to
p
r
e
v
en
t
s
p
r
ea
d
o
f
d
is
ea
s
e,
e
s
ca
p
e
f
r
o
m
c
r
im
es,
s
av
e
o
u
r
h
ea
lth
f
r
o
m
p
o
llu
tio
n
.
T
h
ese
ar
e
th
e
ca
s
es
to
wea
r
f
ac
e
m
ask
s
.
I
n
s
u
ch
ca
s
es
v
er
y
d
if
f
icu
lt to
id
en
tify
f
ac
es wh
ile
wea
r
in
g
m
ask
s
.
Face
r
ec
o
g
n
itio
n
tech
n
o
lo
g
y
is
co
m
m
o
n
l
y
u
s
ed
to
id
en
tif
y
p
eo
p
le
b
ased
o
n
t
h
eir
f
ac
ial
f
e
atu
r
es,
an
d
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
it
p
er
f
o
r
m
s
ex
tr
em
ely
well.
Ma
n
y
r
ea
l
ti
m
e
ap
p
licatio
n
s
b
ased
o
n
f
ac
e
r
ec
o
g
n
itio
n
s
y
s
tem
s
u
ch
as
f
ac
e
au
th
en
ticatio
n
-
b
ased
p
ay
m
e
n
t sy
s
tem
s
,
f
ac
e
ac
ce
s
s
co
n
tr
o
l f
ailed
to
ef
f
ec
tiv
ely
r
ec
o
g
n
ize
th
e
m
ask
ed
f
ac
e
s
.
I
n
o
u
r
r
esear
ch
,
m
ain
o
b
jecti
v
e
is
to
id
en
tify
o
cc
l
u
d
ed
a
r
ea
in
f
ac
e
im
ag
es
wh
ich
is
f
i
r
s
t
s
tep
in
o
cc
lu
d
ed
f
ac
e
r
ec
o
g
n
itio
n
.
O
cc
lu
d
ed
f
ac
e
r
ec
o
g
n
itio
n
is
ch
allen
g
in
g
task
in
c
o
m
p
u
ter
v
is
io
n
.
T
h
is
r
esear
ch
to
p
ic
co
m
es
u
n
d
er
c
o
m
p
u
ter
v
is
io
n
wh
ich
is
th
e
s
u
b
ar
ea
o
f
ar
tific
ial
in
tellig
en
ce
.
C
o
m
p
u
ter
v
is
io
n
is
u
s
ed
to
in
ter
p
r
et
an
d
u
n
d
er
s
tan
d
v
is
u
al
wo
r
ld
.
T
o
tr
ai
n
o
u
r
m
o
d
el
u
s
in
g
d
ee
p
lear
n
in
g
-
b
ased
co
m
p
u
ter
v
is
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
3
,
J
u
n
e
2
0
2
5
:
1
8
6
3
-
1
8
6
9
1864
tech
n
iq
u
e
ca
lled
s
eg
m
en
tatio
n
tech
n
iq
u
e
wh
ic
h
is
U
-
Net
s
em
an
tic
s
eg
m
en
tatio
n
m
o
d
e
l.
B
y
d
iv
id
in
g
th
e
im
ag
es
in
to
s
em
an
tically
s
ig
n
if
ican
t
o
b
jects,
s
em
an
tic
s
eg
m
en
tatio
n
aim
s
to
in
ter
p
r
et
th
e
im
ag
es.
I
t
ca
n
b
e
u
s
ed
ex
ten
s
iv
ely
in
a
v
ar
iety
o
f
ap
p
licatio
n
s
s
u
ch
as
m
ed
ic
al
im
ag
e
an
aly
s
is
,
ag
r
icu
ltu
r
e,
an
d
r
em
o
te
s
en
s
in
g
im
ag
es.
T
h
e
m
ain
aim
o
f
t
h
is
r
esear
ch
is
to
d
etec
t
o
cc
lu
d
e
d
ar
ea
in
f
ac
e
im
a
g
e
u
s
in
g
U
-
Net
s
eg
m
en
tatio
n
m
o
d
el.
Seg
m
e
n
tatio
n
is
th
e
p
r
o
ce
s
s
o
f
ca
teg
o
r
izin
g
an
im
a
g
e
at
th
e
p
ix
el
lev
el
i
n
o
r
d
er
to
id
en
tify
th
e
p
r
ec
is
e
ar
ea
o
f
o
cc
lu
s
io
n
.
W
h
en
o
n
e
o
b
ject
h
id
es
a
p
o
r
tio
n
o
f
an
o
t
h
er
,
th
is
is
k
n
o
wn
as
o
cc
lu
s
io
n
in
a
n
im
a
g
e.
So
m
e
o
f
th
e
o
cc
l
u
d
ed
f
ac
e
im
ag
es a
s
s
h
o
wn
in
F
ig
u
r
e
1
.
Fig
u
r
e
1
.
Occ
lu
d
ed
f
ac
e
im
ag
es
F
ig
u
r
e
1
s
h
o
ws
th
e
f
ac
e
im
ag
es
co
n
tain
s
f
ac
e
with
o
b
jects
s
u
ch
as
m
ask
,
s
u
n
g
lass
,
b
ea
r
d
,
an
d
s
ca
r
f
.
W
h
en
o
cc
u
r
r
ed
th
ese
o
cc
l
u
d
e
d
f
ac
e
im
ag
es
i
n
f
ac
e
r
ec
o
g
n
it
io
n
,
v
er
y
d
if
f
ic
u
lt
to
id
en
tif
y
f
ac
es.
I
n
s
u
ch
ca
s
es
n
ee
d
to
r
em
o
v
e
o
cc
lu
s
io
n
a
n
d
id
en
tify
f
ac
es
in
f
ac
e
r
ec
o
g
n
iti
o
n
tech
n
o
lo
g
y
.
Fo
r
th
at
p
u
r
p
o
s
e,
f
ir
s
t
we
n
ee
d
to
id
en
tify
o
cc
lu
d
e
d
ar
ea
in
f
ac
e
s
an
d
th
en
r
em
o
v
e
o
cc
lu
s
io
n
i
n
f
ac
e
im
ag
es
u
s
in
g
d
ee
p
lear
n
in
g
-
b
ased
m
o
d
els.
I
n
o
u
r
r
esear
c
h
f
o
cu
s
o
n
o
cc
lu
d
ed
ar
ea
wh
ich
is
C
OVI
D
-
1
9
m
ask
ed
ar
ea
d
etec
tio
n
in
f
ac
e
im
ag
es
u
s
in
g
s
eg
m
en
tatio
n
m
o
d
el
ca
lled
U
-
Net
.
Sem
an
tic
s
eg
m
en
tatio
n
task
c
h
allen
g
in
g
o
wn
in
g
t
o
d
etec
t
o
b
jects
in
im
ag
es.
Sev
er
al
s
tu
d
ies
ex
tr
ac
t
th
e
m
ea
n
in
g
f
u
l
in
f
o
r
m
atio
n
f
r
o
m
th
e
im
ag
es
u
s
in
g
s
eg
m
en
tatio
n
task
.
Acc
o
r
d
in
g
to
Pan
et
a
l.
[
1
]
,
t
h
e
EG
-
T
r
a
n
s
UN
et
te
ch
n
i
q
u
e
c
a
n
s
ig
n
i
f
i
ca
n
tl
y
e
n
h
a
n
c
e
t
h
e
s
eg
m
e
n
t
ati
o
n
q
u
al
it
y
o
f
b
i
o
l
o
g
ic
al
p
i
ct
u
r
es
b
y
co
n
c
u
r
r
e
n
tl
y
i
m
p
le
m
e
n
ti
n
g
t
h
e
p
e
r
d
e
k
a
m
p
e
m
o
ti
o
n
al
m
et
h
o
d
(
P
E
M
)
,
t
h
e
f
e
at
u
r
e
f
u
s
i
o
n
m
o
d
u
le
b
as
ed
o
n
s
em
a
n
t
ic
g
u
id
e
d
att
en
ti
o
n
,
a
n
d
t
h
e
c
h
an
n
e
l
s
p
ati
al
at
te
n
ti
o
n
(
C
SA
_
1
)
m
o
d
u
l
e
in
to
U
-
N
et
.
On
tw
o
we
ll
-
k
n
o
w
n
co
l
o
n
o
s
c
o
p
y
d
a
tas
ets
(
C
VC
-
C
lin
icDB
a
n
d
Vas
ir
-
S
E
G)
b
y
at
t
ain
in
g
9
5
.
4
4
a
n
d
9
5
.
2
6
%
o
n
m
Dice
,
r
es
p
e
cti
v
el
y
.
A
b
e
tte
r
U
-
N
et
t
h
at
ac
h
i
e
v
es
h
i
g
h
e
r
p
er
f
o
r
m
a
n
ce
w
it
h
les
s
c
a
p
ac
it
y
b
y
u
s
i
n
g
a
s
p
a
r
s
e
l
y
li
n
k
e
d
b
l
o
c
k
wi
th
m
u
ltis
ca
l
e
f
il
te
r
s
[
2
]
.
A
s
i
n
k
h
o
le
d
atas
et
was
m
an
u
all
y
g
ath
e
r
e
d
f
r
o
m
v
a
r
i
o
u
s
s
o
u
r
ce
s
,
c
o
n
s
i
d
e
r
e
d
,
t
h
en
au
t
o
m
at
ic
all
y
m
a
r
k
e
d
w
it
h
t
h
e
s
u
g
g
es
te
d
au
to
e
n
c
o
d
e
r
,
wh
i
ch
s
h
o
r
te
n
ed
t
h
e
tim
e
u
s
e
d
f
o
r
a
n
n
o
tat
io
n
wh
ile
cr
ea
ti
n
g
ac
cu
r
a
te
m
as
k
s
,
i
n
o
r
d
e
r
t
o
e
v
al
u
ate
o
u
r
a
p
p
r
o
ac
h
.
T
h
e
p
e
r
f
o
r
m
a
n
c
e
a
n
d
d
e
p
e
n
d
a
b
il
it
y
o
f
t
h
e
m
o
d
el
wer
e
t
h
en
f
u
r
t
h
e
r
ass
ess
ed
u
s
i
n
g
a
b
e
n
c
h
m
a
r
k
n
u
cle
i
d
a
tase
t
9
4
%
o
f
th
e
t
im
e.
R
ese
ar
c
h
b
y
D
o
n
g
et
a
l
.
[
3
]
t
o
tack
le
th
e
s
ca
le
v
ar
ian
ce
is
s
u
e
an
d
im
p
r
o
v
e
th
e
s
eg
m
en
tatio
n
o
u
tco
m
es,
a
n
s
u
p
er
p
ix
el
s
eg
m
en
tatio
n
p
o
o
lin
g
(
SSP
)
lay
er
is
in
teg
r
ate
d
i
n
to
th
e
en
h
a
n
ce
d
lig
h
tweig
h
t
en
d
-
to
-
en
d
s
em
an
tic
s
eg
m
en
tat
io
n
(
E
L
E
S2
)
ar
ch
itectu
r
e
to
ac
co
m
p
lis
h
e
n
d
-
to
-
e
n
d
ef
f
icien
t
s
em
an
tic
s
eg
m
en
tatio
n
o
f
h
ig
h
-
r
eso
lu
tio
n
r
e
m
o
te
s
en
s
in
g
(
HR
R
S
)
im
ag
es.
E
L
E
S2
ca
n
r
etain
g
r
ea
t
c
o
m
p
u
tin
g
ef
f
i
cien
cy
wh
ile
ac
h
iev
in
g
p
r
o
m
is
in
g
s
eg
m
en
tatio
n
ac
cu
r
ac
y
.
W
ith
ju
s
t
1
2
.
6
2
M1
p
ar
am
eter
s
an
d
1
3
.
0
9
f
l
o
atin
g
-
p
o
in
t
o
p
e
r
atio
n
s
p
er
s
ec
o
n
d
(
FLOPs
)
,
io
n
E
L
E
S2
o
b
tain
s
m
I
o
U
o
f
8
0
.
1
6
an
d
7
3
.
2
0
%
o
n
th
e
I
PS
R
S
Po
ts
d
am
an
d
Vaih
in
g
en
_
1
d
a
taset,
r
esp
ec
tiv
ely
.
Acc
o
r
d
in
g
to
Gao
et
a
l.
[
4
]
,
a
u
n
iq
u
e
Swin
-
Un
et
n
etwo
r
k
to
en
h
an
ce
m
u
lti
-
s
ca
le
le
s
io
n
s
eg
m
en
tatio
n
p
r
ec
is
io
n
in
C
OVI
D
-
1
9
C
T
s
ca
n
s
.
R
esea
r
c
h
b
y
C
h
e
n
et
a
l
.
[
5
]
to
im
p
r
o
v
e
th
e
s
eg
m
en
tatio
n
q
u
ality
o
f
b
io
lo
g
ical
p
ictu
r
es,
a
n
ew
tr
a
n
s
f
o
r
m
er
-
b
ased
atten
tio
n
-
d
ir
e
cted
U
-
Net
ca
lled
T
r
an
s
AttUn
et
s
im
u
ltan
eo
u
s
ly
in
teg
r
ates
m
u
lti
-
lev
el
g
u
id
e
d
a
tten
tio
n
an
d
m
u
lti
-
s
ca
le
s
k
ip
co
n
n
ec
tio
n
s
in
to
U
-
Net
.
Z
u
o
et
a
l.
[
6
]
s
u
g
g
ests
in
o
r
d
er
to
ac
cu
r
ately
s
eg
m
e
n
t
cr
o
p
s
ee
d
lin
g
s
in
th
eir
n
atu
r
al
h
ab
itat
an
d
ac
co
m
p
lis
h
au
to
n
o
m
o
u
s
ass
ess
m
en
t
o
f
s
ee
d
lin
g
lo
ca
tio
n
an
d
p
h
e
n
o
ty
p
e.
Gite
et
a
l.
[
7
]
u
s
e
U
-
Net
ar
ch
itectu
r
e
u
s
ed
in
lu
n
g
s
eg
m
e
n
tatio
n
u
s
in
g
x
-
r
ay
.
Half
-
UNe
t,
T
r
an
s
UNe
t,
DS
-
T
r
an
s
UNe
t
d
u
al
s
win
tr
an
s
f
o
r
m
er
[
8
]
–
[
1
1
]
u
s
ed
f
o
r
m
ed
ica
l
im
ag
e
p
r
o
ce
s
s
in
g
.
R
esea
r
c
h
b
y
Sh
elh
am
e
r
et
a
l.
[
1
2
]
f
u
lly
co
n
v
o
lu
tio
n
al
n
et
wo
r
k
(
FC
N
)
o
n
s
em
an
tic
s
eg
m
en
tatio
n
an
d
s
ce
n
e
p
ar
s
in
g
,
ex
p
lo
r
in
g
PASC
AL
v
is
u
al
o
b
ject
class
e
s
(
VO
C
)
1
,
NYU
-
Dep
th
v
2
d
ataset
(
NYUD
v
2
)
,
an
d
s
ca
le
-
in
v
ar
ian
t
f
ea
tu
r
e
tr
an
s
f
o
r
m
(
SIFT
)
f
lo
w.
Desp
ite
th
e
f
ac
t
th
at
th
ese
ac
tiv
itie
s
h
av
e
tr
ad
itio
n
ally
m
ad
e
a
d
is
tin
ctio
n
b
etwe
en
ar
ea
s
a
n
d
o
b
jects.
R
esea
r
ch
b
y
L
i
et
a
l
.
[
1
3
]
to
p
ar
allelize
th
e
s
em
an
t
ic
s
eg
m
en
tatio
n
o
f
tar
g
et
d
etec
tio
n
,
a
f
ast
in
s
tan
c
e
s
eg
m
en
tatio
n
m
eth
o
d
b
ased
o
n
m
et
r
ic
lear
n
in
g
is
s
u
g
g
este
d
f
o
r
b
o
th
lo
g
en
d
f
ac
e
d
etec
tio
n
an
d
s
em
an
tic
s
eg
m
en
tatio
n
.
L
iter
ate
s
u
r
v
ey
o
n
d
if
f
e
r
en
t
im
a
g
e
s
eg
m
en
tatio
n
tech
n
i
q
u
es
[
1
4
]
–
[
1
9
]
.
Ob
ject
d
etec
tio
n
o
f
ae
r
ial
im
ag
es,
f
ac
e
d
etec
tio
n
an
d
s
eg
m
en
tatio
n
,
d
etec
tio
n
o
f
g
r
ap
e
clu
s
ter
s
u
s
in
g
m
ask
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
R
-
C
NN
)
[
2
0
]
–
[
2
3
]
.
T
h
e
Dee
p
lab
v
3
+
m
o
d
el'
s
ar
ch
itectu
r
e
with
atten
tio
n
m
ec
h
an
is
m
s
f
o
r
s
eg
m
en
tin
g
o
c
u
lar
im
ag
es
[
2
4
]
.
Dif
f
e
r
en
t
e
v
alu
atio
n
s
m
etr
ics
u
s
ed
f
o
r
s
em
an
tic
s
eg
m
e
n
tatio
n
tech
n
iq
u
es
[
2
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Dete
ctio
n
o
f p
a
r
tia
lly
o
cc
lu
d
e
d
a
r
ea
in
fa
ce
ima
g
e
u
s
in
g
U
-
N
et
mo
d
el
(
Jy
o
th
s
n
a
C
h
era
p
a
n
a
mjeri
)
1865
T
h
e
in
f
e
r
en
ce
f
r
o
m
liter
atu
r
e
r
ev
iew
is
th
at
s
em
an
tic
s
eg
m
en
tatio
n
is
u
s
ed
in
m
a
n
y
d
if
f
er
en
t
r
ea
l
-
tim
e
ap
p
licatio
n
s
s
u
c
h
as m
ed
ical
im
ag
e
s
eg
m
e
n
tatio
n
,
ag
r
ic
u
ltu
r
e,
o
b
ject
d
etec
tio
n
i
n
ae
r
ial
im
ag
es
,
an
d
f
ac
e
d
etec
tio
n
.
B
y
in
s
p
ir
in
g
ex
is
tin
g
wo
r
k
,
we
ap
p
ly
th
is
s
em
a
n
tic
tech
n
iq
u
e
to
o
cc
l
u
s
io
n
ar
ea
d
etec
tio
n
in
f
ac
e
im
ag
es
wh
ich
is
m
ask
ed
ar
ea
in
f
ac
e
im
ag
e.
T
h
e
m
ain
s
ig
n
i
f
ican
t
o
f
th
is
r
esear
ch
wo
r
k
is
to
g
u
id
e
to
r
em
o
v
e
th
e
o
cc
lu
s
io
n
in
f
ac
e
im
ag
es
an
d
r
ec
o
g
n
izes
th
e
f
ac
es
in
r
ea
l
-
tim
e.
W
ith
o
u
t
id
e
n
tify
in
g
o
cc
lu
s
io
n
in
f
ac
e
im
ag
es v
er
y
d
if
f
icu
lt to
r
e
m
o
v
e
o
cc
lu
s
io
n
.
T
h
e
m
ajo
r
ity
o
f
ex
is
tin
g
ap
p
r
o
ac
h
es
h
av
e
f
o
c
u
s
ed
o
n
e
n
h
an
c
in
g
th
e
n
etwo
r
k
ca
p
ac
ity
to
e
n
h
an
ce
th
e
m
o
d
el’
s
f
u
n
ctio
n
ality
.
T
h
is
will
r
esu
lt
in
s
ig
n
if
ican
t
d
is
a
d
v
an
tag
es
:
i
)
i
t
m
ak
es
th
er
e
m
o
r
e
lay
er
s
,
ii
)
i
t
in
cr
ea
s
es
th
e
lik
elih
o
o
d
o
f
o
v
er
f
itti
n
g
in
t
h
e
n
e
u
r
al
n
etw
o
r
k
,
a
n
d
iii
)
m
o
r
e
tr
ain
i
n
g
s
am
p
les
ar
e
n
ee
d
e
d
.
Su
m
m
ar
y
o
f
o
u
r
p
r
im
ar
y
c
o
n
tr
ib
u
tio
n
s
as f
o
llo
ws:
‒
A
U
-
Net
Mo
d
el
th
at
ef
f
ec
tiv
ely
id
en
tif
y
th
e
o
cc
lu
d
ed
ar
e
a
in
th
e
f
ac
e
im
ag
es
in
-
ter
m
s
o
f
th
e
b
in
ar
y
s
eg
m
en
tatio
n
m
ask
.
T
h
e
m
ai
n
ad
v
an
ta
g
e
o
f
t
h
is
m
o
d
el
is
to
id
en
tify
o
cc
lu
d
ed
a
r
ea
in
lim
ited
d
ata
s
o
u
r
ce
s
.
‒
Pro
v
id
in
g
a
co
m
p
r
eh
en
s
iv
e
r
e
v
iew
o
f
th
e
p
er
f
o
r
m
an
ce
o
f
s
e
g
m
en
tatio
n
m
o
d
els.
‒
T
h
e
s
u
p
er
io
r
ity
an
d
g
en
e
r
aliza
b
ilit
y
o
f
th
e
s
u
g
g
ested
U
-
Ne
t
f
o
r
au
to
m
atic
o
cc
lu
d
ed
ar
ea
s
eg
m
en
tatio
n
ar
e
d
em
o
n
s
tr
ated
b
y
e
x
ten
s
iv
e
ex
p
er
im
en
tal
f
in
d
in
g
s
o
n
d
ata
s
ets o
f
o
cc
lu
d
ed
f
ac
e
im
ag
es.
‒
C
o
n
d
u
ctin
g
a
c
o
m
p
ar
ativ
e
s
tu
d
y
o
f
th
r
ee
s
tate
-
of
-
th
e
-
ar
t
(
SOT
A)
s
eg
m
en
tatio
n
m
o
d
els,
n
am
ely
FC
N,
Dee
p
L
ab
v
3
+,
an
d
Py
r
am
id
s
c
en
e
p
ar
s
in
g
n
etwo
r
k
(
PS
PNET
)
.
2.
M
E
T
H
O
D
I
n
th
is
s
ec
tio
n
,
in
o
r
d
er
to
d
et
ec
t
th
e
o
cc
lu
d
e
d
ar
ea
in
f
ac
e
im
ag
es
-
in
th
is
ca
s
e,
th
e
m
ask
ed
ar
ea
in
f
ac
es
-
we
d
escr
ib
e
o
u
r
n
o
v
el
n
etwo
r
k
ar
c
h
itectu
r
e.
M
as
k
co
v
er
ed
b
y
n
o
s
e,
m
o
u
th
,
c
h
in
ar
ea
s
.
Dete
ctin
g
o
cc
lu
d
ed
ar
e
a
in
f
ac
e
im
ag
e
is
ch
allen
g
in
g
task
in
c
o
m
p
u
te
r
v
is
io
n
.
T
h
is
is
v
er
y
im
p
o
r
tan
t ta
s
k
s
f
o
r
o
cc
lu
s
io
n
r
em
o
v
al.
T
h
e
g
o
al
o
f
o
u
r
r
ese
ar
ch
is
to
g
en
er
ate
a
b
in
ar
y
s
eg
m
en
tatio
n
m
ap
o
f
th
e
m
ask
ed
o
b
ject
in
th
e
f
ac
e
im
ag
e
in
p
u
t.
2
.1
.
Understa
nd
ing
t
he
U
-
Net
ba
s
ic
a
r
chit
ec
t
ur
e
U
-
Net
is
a
d
ee
p
lear
n
in
g
-
b
a
s
ed
co
m
p
u
te
r
v
is
io
n
f
r
am
ew
o
r
k
.
U
-
Net
ar
ch
itectu
r
e
is
lo
o
k
lik
e
a
U
-
s
h
ap
ed
ar
ch
itectu
r
e
.
T
h
e
m
ain
p
u
r
p
o
s
e
o
f
U
-
Net
ar
c
h
itectu
r
e
is
u
s
ed
to
s
eg
m
en
ts
im
ag
e
s
ac
cu
r
ately
.
T
h
er
e
ar
e
co
n
tr
ac
tin
g
an
d
e
x
ten
d
in
g
p
ath
s
in
th
e
U
-
Net
d
esig
n
.
T
h
e
co
n
tr
ac
tin
g
p
ath
co
n
s
is
ts
o
f
en
co
d
er
la
y
er
s
.
T
h
is
en
co
d
er
lay
er
s
r
ec
eiv
es
c
o
n
tex
tu
al
in
f
o
r
m
atio
n
a
n
d
d
ec
r
ea
s
e
th
e
s
p
atial
r
eso
lu
tio
n
o
f
th
e
i
n
p
u
t.
T
h
e
ex
p
an
d
i
n
g
p
ath
in
cl
u
d
es
d
ec
o
d
er
lay
er
s
th
at
d
ec
o
d
e
d
th
e
alr
ea
d
y
en
co
d
ed
d
ata
an
d
u
s
e
th
e
in
f
o
r
m
atio
n
f
r
o
m
th
e
co
n
tr
ac
tin
g
p
ath
v
ia
s
k
ip
co
n
n
ec
tio
n
s
to
g
e
n
er
ate
th
e
b
in
ar
y
s
eg
m
en
tatio
n
m
ap
.
Fin
d
in
g
th
e
ap
p
r
o
p
r
iate
f
ea
tu
r
es
in
th
e
in
p
u
t
im
ag
e
is
th
e
p
r
im
ar
y
g
o
al
o
f
th
e
co
n
tr
ac
tin
g
p
ath
.
T
h
e
co
n
tr
ac
tin
g
p
ath
is
s
am
e
as
th
e
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
.
T
h
e
o
p
er
atio
n
o
f
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
is
to
c
o
n
v
o
lu
tio
n
al
o
p
er
atio
n
f
o
llo
wed
b
y
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
.
E
v
er
y
b
lo
ck
c
o
m
p
r
is
es
o
f
two
co
n
s
ec
u
tiv
e
3
×
3
co
n
v
o
lu
tio
n
al
lay
er
s
,
s
u
cc
ee
d
ed
b
y
an
ac
tiv
a
tio
n
f
u
n
ctio
n
o
f
R
eL
U
.
Fo
llo
win
g
th
e
c
o
n
v
o
lu
tio
n
o
p
er
atio
n
,
m
ax
p
o
o
lin
g
2
×
2
o
p
er
atio
n
s
with
s
tr
id
e
2
is
u
s
ed
.
T
h
e
n
ex
t
co
n
v
o
lu
tio
n
al
lay
er
d
o
u
b
les
th
e
n
u
m
b
er
o
f
f
ilter
s
em
p
lo
y
ed
af
ter
ea
ch
m
ax
p
o
o
lin
g
s
tep
.
A
lay
er
th
at
b
eg
in
s
with
6
4
f
ea
t
u
r
e
ch
an
n
els,
f
o
r
in
s
tan
ce
,
will
h
av
e
1
2
8
ch
a
n
n
els
f
o
llo
win
g
t
h
e
s
u
b
s
eq
u
en
t
p
o
o
l
in
g
an
d
c
o
n
v
o
lu
tio
n
p
r
o
ce
s
s
es.
On
th
e
o
th
er
h
an
d
,
ex
p
an
s
iv
e
p
ath
wh
ich
ta
k
es
th
e
ex
tr
ac
ted
in
p
u
t
f
ea
tu
r
es
a
n
d
g
e
n
er
ate
s
eg
m
e
n
tatio
n
m
a
s
k
.
E
ac
h
b
lo
ck
is
f
o
llo
wed
b
y
u
p
s
am
p
lin
g
lay
er
.
T
h
e
d
ec
o
d
e
r
lay
er
s
in
th
e
ex
p
an
s
iv
e
p
ath
u
p
s
am
p
lin
g
th
e
f
ea
tu
r
e
m
ap
s
.
T
h
e
d
ec
o
d
er
la
y
er
s
ca
n
d
etec
t
th
e
f
ea
tu
r
es
m
o
r
e
p
r
ec
is
ely
b
y
u
s
in
g
th
e
s
k
ip
co
n
n
ec
tio
n
s
f
r
o
m
th
e
co
n
tr
ac
tin
g
p
ath
to
th
e
ex
p
an
s
iv
e
p
at
h
,
w
h
ich
h
elp
s
to
p
r
eser
v
e
lo
s
t sp
atial
in
f
o
r
m
atio
n
.
Fig
u
r
e
2
illu
s
tr
ates
h
o
w
th
e
U
-
Net
n
etwo
r
k
co
n
v
er
ts
a
g
r
ay
s
ca
le
in
p
u
t
im
ag
e
o
f
s
ize
5
7
2
×5
7
2
×1
in
to
a
b
in
ar
y
s
eg
m
en
ted
o
u
tp
u
t
m
ap
o
f
s
ize
3
8
8
×3
8
8
×2
.
Si
n
ce
n
o
p
a
d
d
in
g
h
as
b
ee
n
u
s
ed
,
th
e
o
u
tp
u
t
s
ize
is
s
m
aller
th
an
th
e
in
p
u
t
s
ize,
w
h
ich
ca
n
b
e
o
b
s
er
v
e
d
.
W
e
h
av
e
to
k
ee
p
th
e
i
n
p
u
t
s
ize
th
e
s
am
e
if
we
ca
n
u
s
e
th
e
p
ad
d
in
g
.
T
h
e
in
p
u
t
im
ag
e
r
a
p
id
ly
d
ec
r
ea
s
es
wid
th
an
d
h
eig
h
t
as
th
e
n
u
m
b
er
o
f
ch
an
n
e
ls
g
r
o
ws
alo
n
g
th
e
co
n
tr
ac
tin
g
p
ath
.
I
f
m
o
r
e
ch
an
n
els
ex
is
t,
th
e
n
etwo
r
k
m
ay
g
ath
er
h
ig
h
e
r
-
lev
el
d
ata.
A
last
co
n
v
o
lu
tio
n
o
p
er
atio
n
at
t
h
e
b
o
ttlen
ec
k
p
r
o
d
u
ce
s
a
f
ea
tu
r
e
m
ap
with
3
0
×3
0
×
1
0
2
4
p
i
x
els.
Af
ter
r
em
o
v
in
g
th
e
f
ea
tu
r
e
m
a
p
f
r
o
m
th
e
b
o
ttlen
ec
k
,
th
e
ex
p
an
s
iv
e
p
ath
r
esizes
it
to
f
it
th
e
in
itial
in
p
u
t
s
ize
u
p
s
am
p
lin
g
lay
er
s
,
wh
ich
d
ec
r
ea
s
e
th
e
n
u
m
b
er
o
f
ch
an
n
els in
th
e
f
ea
tu
r
e
m
ap
wh
ile
im
p
r
o
v
i
n
g
its
s
p
atial
r
eso
lu
tio
n
,
ar
e
u
s
ed
to
ac
h
iev
e
th
is
.
T
h
e
d
ec
o
d
e
r
la
y
er
s
u
s
e
th
e
s
k
ip
co
n
n
ec
tio
n
s
f
r
o
m
th
e
co
n
tr
ac
tio
n
p
ath
to
lo
ca
te
a
n
d
r
ef
in
e
th
e
im
ag
e'
s
f
ea
tu
r
es.
I
n
t
h
e
en
d
,
ea
ch
p
ix
e
l
in
th
e
o
u
t
p
u
t
im
ag
e
co
r
r
esp
o
n
d
s
to
a
lab
el
in
th
e
in
p
u
t
im
a
g
e
th
at
is
co
n
n
ec
te
d
to
a
p
ar
ticu
lar
o
b
ject
o
r
class
.
E
ac
h
p
ix
el
in
t
h
is
o
u
tp
u
t
m
a
p
s
h
o
ws
eith
er
th
e
b
ac
k
g
r
o
u
n
d
o
r
th
e
f
o
r
eg
r
o
u
n
d
b
ec
au
s
e
it is
a
b
in
ar
y
s
eg
m
e
n
t
atio
n
m
ap
.
B
asic step
s
in
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
:
‒
T
h
e
in
p
u
t
im
ag
e
is
d
eliv
er
e
d
to
th
e
c
o
n
tr
ac
tin
g
p
ath
,
wh
ich
s
ee
k
s
to
ca
p
tu
r
e
r
ele
v
an
t
d
et
ails
ab
o
u
t
th
e
in
p
u
t im
ag
e
w
h
ile
r
ed
u
cin
g
th
e
s
p
atial
d
im
en
s
io
n
s
o
f
th
e
im
ag
e.
‒
Af
ter
ex
tr
ac
tin
g
c
o
m
p
lex
ch
ar
ac
ter
is
tics
an
d
p
atter
n
s
,
th
e
f
e
atu
r
e
m
ap
p
r
o
ce
e
d
s
to
th
e
ex
p
an
d
in
g
p
ath
.
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.
1
4
,
No
.
3
,
J
u
n
e
2
0
2
5
:
1
8
6
3
-
1
8
6
9
1866
‒
T
h
e
ex
p
an
d
in
g
p
ath
u
p
s
am
p
les
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
a
n
d
co
m
b
i
n
es
lear
n
ed
f
ea
tu
r
e
s
u
s
in
g
b
o
th
co
n
v
o
l
u
tio
n
an
d
u
p
-
co
n
v
o
lu
ti
o
n
p
r
o
ce
s
s
es to
p
r
o
d
u
ce
a
s
eg
m
en
tatio
n
m
ap
.
‒
Sk
ip
co
n
n
ec
tio
n
s
ar
e
u
s
ed
t
o
d
o
u
b
le
t
h
e
f
ea
tu
r
e
ch
an
n
els
an
d
co
n
ca
ten
ate
t
h
e
r
elev
a
n
t
f
ea
t
u
r
e
m
ap
f
r
o
m
th
e
co
n
tr
ac
tin
g
p
ath
.
‒
E
ac
h
p
ix
el
in
th
e
f
ea
t
u
r
e
m
a
p
s
h
o
u
ld
b
e
class
if
ied
s
ep
ar
ately
in
a
s
eg
m
e
n
tatio
n
m
ap
th
a
t
is
p
r
o
d
u
ce
d
af
ter
u
p
s
am
p
lin
g
it in
th
e
e
x
p
a
n
d
in
g
p
ath
.
Fig
u
r
e
2
.
Pro
p
o
s
ed
s
y
s
t
em
ar
c
h
itectu
r
e
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
ec
tio
n
,
we
p
r
esen
t
t
h
e
ex
p
er
im
en
tal
s
etu
p
an
d
o
u
tco
m
es
o
f
o
u
r
m
eth
o
d
.
W
e
b
eg
in
b
y
d
escr
ib
in
g
t
h
e
d
ataset
u
s
ed
f
o
r
tr
ain
i
n
g
an
d
ev
alu
atio
n
p
u
r
p
o
s
es.
T
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
is
th
en
d
escr
ib
ed
,
to
g
eth
er
with
th
e
p
ar
ticu
lar
s
an
d
co
n
f
ig
u
r
atio
n
s
th
at
wer
e
u
s
ed
.
W
e
d
ev
elo
p
ap
p
r
o
p
r
iate
ev
alu
atio
n
m
etr
ics
an
d
ap
p
ly
th
em
t
o
o
u
r
n
etwo
r
k
to
ev
alu
ate
its
p
er
f
o
r
m
a
n
ce
.
C
ar
r
ied
o
u
t
a
f
ew
test
s
to
ev
alu
ate
o
u
r
s
u
g
g
ested
m
o
d
el
ag
ain
s
t SOT
A
tech
n
iq
u
es.
3
.
1
.
Da
t
a
s
et
T
h
e
Ma
s
k
ed
Face
-
C
eleb
A
d
ata
s
et
is
av
ailab
le
to
th
e
p
u
b
lic.
T
h
is
d
ataset
is
co
n
s
tr
u
cted
f
r
o
m
C
eleb
A
d
ataset
u
s
in
g
th
e
Ma
s
k
T
h
eFac
e
to
o
l.
T
h
is
d
ataset
is
u
s
ed
in
th
e
f
ield
o
f
co
m
p
u
ter
v
is
io
n
,
esp
ec
ially
f
o
r
f
ac
e
an
aly
s
is
task
s
.
W
e
ass
ess
th
e
m
o
d
els
m
en
tio
n
e
d
p
r
ev
io
u
s
ly
u
s
in
g
th
e
m
ask
ed
-
f
ac
e
d
at
asets
.
T
h
is
d
ataset
co
n
tain
s
2
1
,
8
4
4
m
ask
e
d
f
ac
e
im
ag
es
an
d
c
o
r
r
esp
o
n
d
in
g
t
ar
g
et
o
r
g
r
o
u
n
d
tr
u
th
im
ag
es
o
n
o
cc
lu
d
ed
ar
ea
.
I
m
ag
es
o
f
s
ize
ar
e
2
5
6
×
2
5
6
p
ix
els.
T
h
ese
im
ag
es
ar
e
r
a
n
d
o
m
ly
d
iv
id
e
d
in
to
1
7
,
4
7
6
(
8
0
%)
im
ag
es
f
o
r
tr
ain
in
g
,
2
,
1
8
4
(
1
0
%)
im
ag
es
f
o
r
v
alid
atio
n
,
a
n
d
2
,
1
8
4
(
1
0
%)
im
ag
es
f
o
r
test
in
g
f
r
o
m
th
e
Ma
s
k
ed
Face
-
C
eleb
A
d
ataset.
3
.
2
.
E
v
a
lua
t
i
o
n
m
et
ric
I
n
o
u
r
e
x
p
er
im
e
n
t,
we
u
s
e
ac
cu
r
ac
y
as
th
e
m
ain
p
ar
a
m
eter
to
d
eter
m
in
e
h
o
w
clo
s
ely
th
e
p
r
ed
icted
m
ask
an
d
g
r
o
u
n
d
tr
u
th
m
atch
.
T
h
is
m
etr
ic
is
as
s
o
ciate
d
with
f
o
u
r
v
alu
es
i.e
.
,
tr
u
e
-
p
o
s
itiv
e
(
T
P),
tr
u
e
-
n
e
g
ativ
e
(
T
N)
,
f
alse
-
p
o
s
itiv
e
(
FP
)
,
an
d
f
alse
-
n
eg
ativ
e
(
FN)
.
T
h
e
f
o
r
m
u
la
f
o
r
ac
c
u
r
ac
y
is
s
h
o
wn
in
(
1
)
.
=
+
+
+
+
(
1
)
T
h
e
n
u
m
b
er
o
f
o
cc
lu
d
e
d
ar
ea
p
ix
els,
o
r
m
ask
ed
ar
ea
p
ix
el
s
th
at
h
a
v
e
b
e
en
c
o
r
r
ec
tly
id
e
n
tifie
d
,
is
r
ep
r
esen
ted
b
y
T
P.
T
h
e
n
u
m
b
er
o
f
b
ac
k
g
r
o
u
n
d
p
ix
els
th
at
a
r
e
in
co
r
r
ec
tly
class
if
ied
as
b
a
ck
g
r
o
u
n
d
is
k
n
o
wn
as
FP
.
T
h
e
n
u
m
b
er
o
f
o
cc
lu
d
e
d
ar
ea
p
ix
els
th
at
ar
e
in
co
r
r
ec
t
ly
class
if
ied
as
b
ac
k
g
r
o
u
n
d
is
k
n
o
wn
as
FP
.
T
h
e
n
u
m
b
er
o
f
b
ac
k
g
r
o
u
n
d
p
i
x
els th
at
ar
e
co
r
r
ec
tly
class
if
ied
as b
ac
k
g
r
o
u
n
d
is
k
n
o
wn
as T
N.
3
.
3
.
E
x
perim
ent
a
l
s
et
t
ing
s
T
h
e
n
etwo
r
k
is
tr
ain
ed
u
s
in
g
th
e
in
p
u
t
im
ag
es
an
d
th
e
s
eg
m
en
tatio
n
m
ap
s
th
at
co
r
r
esp
o
n
d
to
th
em
.
All
th
e
ex
p
er
im
en
ts
ar
e
im
p
le
m
en
ted
with
T
en
s
o
r
f
l
o
w
wh
ic
h
is
Py
th
o
n
f
r
am
ew
o
r
k
.
A
n
8
-
co
r
e
3
2
G+
1
V1
0
0
GPU
PC
an
d
a
3
.
8
GHz
C
PU
ar
e
u
s
ed
in
th
e
ex
p
er
im
en
t
to
t
r
ain
t
h
e
n
etwo
r
k
.
Ad
am
is
em
p
lo
y
ed
as
a
n
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
Dete
ctio
n
o
f p
a
r
tia
lly
o
cc
lu
d
e
d
a
r
ea
in
fa
ce
ima
g
e
u
s
in
g
U
-
N
et
mo
d
el
(
Jy
o
th
s
n
a
C
h
era
p
a
n
a
mjeri
)
1867
o
p
tim
izer
to
ad
ju
s
t
th
e
o
cc
lu
d
ed
f
ac
e
im
ag
e
s
eg
m
en
tatio
n
n
etwo
r
k
'
s
p
ar
am
eter
s
.
T
h
is
wo
r
k
em
p
lo
y
e
d
th
e
f
o
llo
win
g
s
u
p
e
r
p
a
r
am
eter
s
:
b
atch
s
ize
o
f
1
6
,
im
a
g
e
r
eso
lu
t
io
n
o
f
2
5
6
,
e
p
o
ch
s
o
f
1
0
0
,
an
d
s
tar
tin
g
lear
n
i
n
g
r
ate
η
o
f
0
.
0
5
.
I
n
th
is
m
an
u
s
cr
ip
t,
1
0
0
was
u
tili
ze
d
f
o
r
e
p
o
ch
s
,
wh
ich
ar
e
th
e
n
u
m
b
e
r
o
f
e
x
p
er
im
en
tal
tr
ain
in
g
r
o
u
n
d
s
.
T
h
e
im
a
g
es
u
s
ed
f
o
r
ex
p
er
im
e
n
tal
tr
ain
in
g
an
d
t
esti
n
g
wer
e
co
n
v
er
ted
in
t
o
2
5
6
×
2
5
6
r
eso
lu
tio
n
.
B
atch
_
Size
is
th
e
n
u
m
b
er
o
f
s
am
p
les
s
elec
ted
f
o
r
o
n
e
tr
ain
in
g
.
3
.
4
.
E
x
periment
a
l
r
esu
lt
s
T
o
ev
al
u
ate
th
e
ef
f
icac
y
o
f
th
e
p
r
o
p
o
s
ed
U
-
Net
we
f
ir
s
t
co
n
d
u
ct
th
e
ex
p
e
r
im
en
t
s
o
n
th
e
Ma
s
k
ed
Face
-
C
eleb
A
d
ataset
f
o
r
th
e
task
o
f
m
ask
ed
ar
ea
s
eg
m
en
tatio
n
.
T
h
e
r
esu
lts
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
as
s
h
o
wn
in
F
ig
u
r
e
3
.
T
h
e
im
ag
e
as
s
h
o
wn
in
F
ig
u
r
e
3
(
a)
i
s
th
e
m
ask
ed
f
ac
e
im
ag
e
wh
ich
is
in
p
u
t
to
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
im
a
g
e
a
s
s
h
o
wn
in
F
ig
u
r
e
3
(
b
)
is
th
e
m
ask
wh
ich
is
t
h
e
g
r
o
u
n
d
tr
u
t
h
lab
el.
T
h
is
is
wh
at
o
u
r
m
o
d
el
m
u
s
t
p
r
ed
icted
f
o
r
th
e
g
iv
e
n
m
ask
ed
f
ac
e
im
a
g
e.
T
h
e
im
a
g
e
as
s
h
o
wn
in
F
ig
u
r
e
3
(
c)
p
r
ed
icted
o
cc
lu
d
ed
r
e
g
io
n
.
T
h
e
wh
ite
r
eg
io
n
d
en
o
tes
th
e
m
ask
ed
ar
ea
wh
ich
is
o
cc
lu
d
e
ar
ea
an
d
th
e
b
lack
r
eg
i
o
n
d
en
o
tes
th
e
n
o
o
cc
lu
d
ed
a
r
ea
.
No
tice
th
at
if
th
e
m
ask
is
en
tire
ly
b
lack
th
is
m
ea
n
s
th
er
e
ar
e
n
o
o
cc
l
u
d
ed
ar
ea
d
ep
o
s
its
in
th
e
g
iv
e
n
m
ask
ed
f
ac
e
im
ag
e.
(
a)
(
b
)
(
c)
Fig
u
r
e
3
.
E
x
p
er
im
e
n
tal
r
esu
lt f
o
r
U
-
Net
m
o
d
el
o
n
Ma
s
k
ed
Face
-
C
eleb
A
d
ataset
of
(
a)
t
h
e
o
r
ig
in
al
o
cc
lu
d
es
f
ac
e
im
ag
e
,
(
b
)
g
r
o
u
n
d
tr
u
th
,
a
n
d
(
c
)
p
r
ed
icted
o
cc
lu
d
e
d
ar
ea
3
.
5
.
Co
m
pa
ra
t
iv
e
s
t
ud
y
W
e
h
av
e
co
n
tr
asted
th
e
o
b
tain
ed
ex
p
er
im
e
n
tal
r
esu
lts
with
th
o
s
e
o
f
o
th
er
s
em
an
tic
s
eg
m
en
tatio
n
ap
p
r
o
ac
h
es
in
o
r
d
er
to
ass
ess
th
e
ef
f
ec
tiv
en
ess
o
f
o
u
r
s
u
g
g
ested
m
eth
o
d
f
u
r
th
er
.
T
h
e
T
ab
le
1
s
h
o
ws
th
e
co
m
p
ar
ativ
e
an
aly
s
is
o
f
v
ar
io
u
s
ex
is
tin
g
s
em
an
tic
s
eg
m
en
tatio
n
m
o
d
els.
T
h
is
co
m
p
ar
ativ
e
an
aly
s
is
s
ay
s
th
at
s
u
p
er
io
r
ity
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
d
etec
tio
n
o
f
o
cc
lu
d
ed
r
eg
io
n
i
n
h
u
m
an
f
ac
e
im
a
g
es.
W
e
n
o
ticed
th
at
o
u
r
m
o
d
el
p
er
f
o
r
m
s
9
7
.
7
%
a
cc
u
r
ac
y
with
o
th
e
r
th
r
ee
m
et
h
o
d
s
as
s
h
o
wn
in
T
ab
le
1
.
Fro
m
th
e
co
m
p
ar
ativ
e
s
tu
d
y
as
s
h
o
wn
in
F
ig
u
r
e
4
,
it
is
o
b
s
er
v
ed
th
at
p
r
o
p
o
s
ed
s
y
s
tem
p
r
o
v
id
es
ap
p
r
ec
iab
le
ac
c
u
r
ac
y
o
f
9
7
.
7
%.
I
t
o
u
tp
er
f
o
r
m
s
co
m
p
ar
ed
t
o
ex
is
tin
g
im
ag
e
s
eg
m
e
n
tatio
n
alg
o
r
i
th
m
s
.
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
v
ar
i
o
u
s
im
ag
e
s
eg
m
e
n
tatio
n
tech
n
iq
u
es
S
.
n
o
M
e
t
h
o
d
A
c
c
u
r
a
c
y
(
%)
1
F
C
N
[
1
2
]
9
0
.
3
D
e
e
p
La
b
v
3
+
[
2
4
]
9
3
.
4
3
P
S
P
N
e
t
[
2
5
]
9
4
.
2
4
Th
e
p
r
o
p
o
s
e
d
mo
d
e
l
9
7
.
7
Fig
u
r
e
4
.
Gr
a
p
h
ical
r
ep
r
esen
tatio
n
o
f
c
o
m
p
ar
ativ
e
s
tu
d
y
b
et
wee
n
p
r
o
p
o
s
ed
s
y
s
tem
an
d
e
x
i
s
tin
g
s
y
s
tem
ap
p
r
o
ac
h
es with
r
ef
er
en
ce
to
a
cc
u
r
ac
y
86
88
90
92
94
96
98
100
F
u
ll
y
C
onvol
uti
on
a
l
Ne
two
r
k
(
F
C
N
)
De
e
p
L
a
b
V3+
P
y
r
a
mi
d
S
c
e
n
e
P
a
r
s
ing
Ne
two
r
k
(
P
S
P
N
e
t
)
T
he
P
r
opo
s
e
d
M
ode
l
A
c
c
u
r
a
c
y
M
e
t
h
o
d
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.
1
4
,
No
.
3
,
J
u
n
e
2
0
2
5
:
1
8
6
3
-
1
8
6
9
1868
3
.
6
.
Abla
t
io
n
s
t
ud
y
R
o
le
o
f
u
s
in
g
s
k
ip
c
o
n
n
ec
tio
n
s
:
we
in
v
esti
g
ate
th
e
ef
f
ec
tiv
e
n
ess
o
f
u
s
in
g
s
k
ip
co
n
n
ec
tio
n
s
.
Ass
u
m
e
th
at
a
U
-
Net
wi
th
o
u
t
s
k
ip
co
n
n
ec
tio
n
s
wo
u
ld
p
r
o
d
u
ce
a
b
lu
r
r
y
m
ess
.
I
f
y
o
u
wo
u
ld
b
e
in
t
er
ested
,
it
co
u
ld
b
e
in
ter
esti
n
g
to
tr
y
d
is
ab
lin
g
th
e
s
k
ip
co
n
n
ec
tio
n
an
d
s
ee
wh
at
h
ap
p
en
s
.
B
y
g
r
ad
u
ally
ad
d
in
g
m
o
r
e
f
ea
tu
r
es
to
a
co
n
v
o
l
u
tio
n
s
-
b
ased
v
is
io
n
m
o
d
el,
th
ey
tr
an
s
f
o
r
m
it
f
r
o
m
an
u
n
s
tab
le
an
d
in
ac
cu
r
ate
m
o
d
el
in
to
a
SOTA
o
n
e
.
B
y
av
o
id
in
g
th
e
b
o
ttlen
ec
k
a
n
d
g
iv
in
g
th
e
m
o
d
el
ac
ce
s
s
to
th
e
en
co
d
e
r
'
s
in
ter
m
ed
iate
ac
tiv
atio
n
s
-
wh
ich
in
clu
d
e
t
h
ese
f
in
e
-
g
r
ain
e
d
,
h
i
g
h
-
r
eso
lu
tio
n
d
etails
-
s
k
ip
co
n
n
ec
tio
n
s
h
elp
s
o
lv
e
th
is
is
s
u
e.
T
h
at
is
wh
at
is
m
ea
n
t b
y
r
ec
o
v
e
r
f
in
e
g
r
ain
ed
d
etails in
th
e
p
r
ed
ictio
n
.
4.
CO
NCLU
SI
O
N
AND
F
U
T
U
RE
WO
RK
T
h
is
p
ap
er
p
r
o
p
o
s
e
d
a
n
o
v
el
m
eth
o
d
f
o
r
b
i
n
ar
y
s
eg
m
en
tati
o
n
m
a
p
o
n
o
cc
lu
d
e
d
a
r
ea
d
etec
tio
n
in
f
ac
e
im
ag
es
ca
lled
U
-
Net
m
o
d
el.
T
h
is
p
r
o
p
o
s
ed
m
o
d
el
is
im
p
r
o
v
ed
v
er
s
io
n
o
f
f
u
lly
co
n
v
o
lu
tio
n
al
n
etwo
r
k
.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
s
h
o
w
th
at
o
u
r
m
o
d
el
ef
f
icien
tly
s
eg
m
e
n
tin
g
o
cc
lu
d
e
d
ar
ea
in
h
u
m
an
f
ac
e
im
ag
e,
in
o
u
r
ca
s
e
o
cc
lu
d
ed
a
r
ea
is
m
ask
e
d
ar
ea
.
Desp
ite
th
e
p
r
o
m
is
in
g
r
esu
lts
o
n
d
ataset.
E
x
p
er
i
m
en
ts
o
n
p
u
b
licly
av
ailab
le
d
ataset
wh
ich
is
Ma
s
k
ed
Face
-
C
eleb
A
d
ataset
s
h
o
ws 9
7
.
7
% seg
m
en
tatio
n
ac
c
u
r
ac
y
.
I
t p
r
o
d
u
ce
s
h
i
g
h
p
er
p
etu
al
q
u
ality
r
esu
lts
co
m
p
ar
ed
to
o
t
h
er
SOTA
im
ag
e
s
eg
m
en
tatio
n
m
eth
o
d
s
.
T
h
e
m
ai
n
ad
v
an
tag
e
o
f
o
u
r
m
o
d
el
h
as
f
ewe
r
p
ar
am
eter
s
,
lim
ited
d
ataset
an
d
f
aster
i
n
s
p
ee
d
.
T
h
is
wo
r
k
is
u
s
ef
u
l
f
o
r
m
ask
ed
f
ac
e
r
ec
o
g
n
itio
n
in
r
ea
l
-
tim
e.
T
h
er
e
ar
e
p
r
o
m
is
in
g
ar
ea
s
f
o
r
f
u
r
th
er
s
tu
d
y
an
d
ad
v
a
n
ce
m
en
ts
in
f
ac
e
m
ask
id
en
tific
atio
n
,
s
u
c
h
as
im
p
r
o
v
in
g
o
u
r
m
o
d
el'
s
p
er
f
o
r
m
a
n
c
e
in
d
if
f
icu
lt
s
itu
atio
n
s
lik
e
h
ig
h
o
cc
lu
s
io
n
o
r
co
m
p
lex
m
ask
p
atter
n
s
.
E
n
h
a
n
cin
g
th
e
d
ataset
f
o
r
d
if
f
e
r
en
t
s
izes
o
f
m
ask
s
an
d
s
h
ap
es.
W
e
also
th
in
k
th
at
m
er
g
in
g
with
f
ac
ial
r
ec
o
g
n
itio
n
r
esear
ch
h
as
a
lo
t
o
f
p
o
ten
tial.
W
e
ca
n
i
n
cr
ea
s
e
th
e
p
r
ec
is
io
n
an
d
d
ep
en
d
a
b
ilit
y
o
f
f
ac
ial
r
ec
o
g
n
i
tio
n
s
y
s
tem
s
b
y
elim
in
atin
g
m
ask
s
as a
p
r
ep
r
o
ce
s
s
in
g
s
tep
.
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
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
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
J
y
o
th
s
n
a
C
h
er
ap
an
am
jer
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
B
.
Nar
en
d
r
a
Ku
m
ar
R
ao
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
o
p
e
n
ly
av
ailab
le
at
h
ttp
s
://d
r
iv
e.
g
o
o
g
le.
co
m
/d
r
iv
e
/f
o
ld
er
s
/1
E
J
b
x
f
g
T
VHDBNv
f
e7
KzE
SwJ
o
W
c8
e4
J
2
HJ?u
s
p
=s
h
ar
e_
lin
k
.
RE
F
E
R
E
NC
E
S
[
1
]
S
.
P
a
n
,
X
.
L
i
u
,
N
.
X
i
e
,
a
n
d
Y
.
C
h
o
n
g
,
“
EG
-
Tr
a
n
sU
N
e
t
:
a
t
r
a
n
sf
o
r
m
e
r
-
b
a
se
d
U
-
N
e
t
w
i
t
h
e
n
h
a
n
c
e
d
a
n
d
g
u
i
d
e
d
m
o
d
e
l
s
f
o
r
b
i
o
me
d
i
c
a
l
i
ma
g
e
se
g
me
n
t
a
t
i
o
n
,
”
B
M
C
B
i
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
4
,
n
o
.
1
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
8
6
/
s1
2
8
5
9
-
0
2
3
-
0
5
1
9
6
-
1.
[
2
]
R
.
A
l
s
h
a
w
i
,
M
.
T.
H
o
q
u
e
,
a
n
d
M
.
C
.
F
l
a
n
a
g
i
n
,
“
A
d
e
p
t
h
-
w
i
se
s
e
p
a
r
a
b
l
e
U
-
N
e
t
a
r
c
h
i
t
e
c
t
u
r
e
w
i
t
h
m
u
l
t
i
s
c
a
l
e
f
i
l
t
e
r
s
t
o
d
e
t
e
c
t
si
n
k
h
o
l
e
s
,
”
R
e
m
o
t
e
S
e
n
s
i
n
g
,
v
o
l
.
1
5
,
n
o
.
5
,
F
e
b
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
5
0
5
1
3
8
4
.
[
3
]
H
.
D
o
n
g
,
B
.
Y
u
,
W
.
W
u
,
a
n
d
C
.
H
e
,
“
En
h
a
n
c
e
d
l
i
g
h
t
w
e
i
g
h
t
e
n
d
-
to
-
e
n
d
se
man
t
i
c
se
g
me
n
t
a
t
i
o
n
f
o
r
h
i
g
h
-
r
e
s
o
l
u
t
i
o
n
r
e
mo
t
e
se
n
si
n
g
i
ma
g
e
s,
”
I
E
EE
A
c
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
7
0
9
4
7
–
7
0
9
5
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
8
2
3
7
0
.
[
4
]
Z.
-
J.
G
a
o
,
Y
.
H
e
,
a
n
d
Y
.
Li
,
“
A
n
o
v
e
l
l
i
g
h
t
w
e
i
g
h
t
S
w
i
n
-
U
n
e
t
n
e
t
w
o
r
k
f
o
r
s
e
ma
n
t
i
c
s
e
g
m
e
n
t
a
t
i
o
n
o
f
C
O
V
I
D
-
1
9
l
e
si
o
n
i
n
C
T
i
ma
g
e
s
,
”
I
E
EE
A
c
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
9
5
0
–
9
6
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
2
3
2
7
2
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
Dete
ctio
n
o
f p
a
r
tia
lly
o
cc
lu
d
e
d
a
r
ea
in
fa
ce
ima
g
e
u
s
in
g
U
-
N
et
mo
d
el
(
Jy
o
th
s
n
a
C
h
era
p
a
n
a
mjeri
)
1869
[
5
]
B
.
C
h
e
n
,
Y
.
L
i
u
,
Z.
Zh
a
n
g
,
G
.
L
u
,
a
n
d
A
.
W
.
K
.
K
o
n
g
,
“
Tr
a
n
sA
t
t
U
n
e
t
:
M
u
l
t
i
-
Le
v
e
l
A
t
t
e
n
t
i
o
n
-
G
u
i
d
e
d
U
-
N
e
t
w
i
t
h
t
r
a
n
sf
o
r
mer
f
o
r
med
i
c
a
l
i
m
a
g
e
s
e
g
me
n
t
a
t
i
o
n
,
”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Em
e
r
g
i
n
g
T
o
p
i
c
s
i
n
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
,
v
o
l
.
8
,
n
o
.
1
,
p
p
.
5
5
–
6
8
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TET
C
I
.
2
0
2
3
.
3
3
0
9
6
2
6
.
[
6
]
X
.
Zu
o
,
H
.
Li
n
,
D
.
W
a
n
g
,
a
n
d
Z
.
C
u
i
,
“
A
me
t
h
o
d
o
f
c
r
o
p
see
d
l
i
n
g
p
l
a
n
t
s
e
g
me
n
t
a
t
i
o
n
o
n
e
d
g
e
i
n
f
o
r
ma
t
i
o
n
f
u
s
i
o
n
mo
d
e
l
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
9
5
2
8
1
–
9
5
2
9
3
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
8
7
8
2
5
.
[
7
]
S
.
G
i
t
e
,
A
.
M
i
s
h
r
a
,
a
n
d
K
.
K
o
t
e
c
h
a
,
“
En
h
a
n
c
e
d
l
u
n
g
i
ma
g
e
se
g
me
n
t
a
t
i
o
n
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
N
e
u
ra
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
5
,
n
o
.
3
1
,
p
p
.
2
2
8
3
9
–
2
2
8
5
3
,
N
o
v
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
2
1
-
0
2
1
-
0
6
7
1
9
-
8.
[
8
]
H
.
L
u
,
Y
.
S
h
e
,
J.
T
i
e
,
a
n
d
S
.
X
u
,
“
H
a
l
f
-
U
N
e
t
:
a
s
i
m
p
l
i
f
i
e
d
U
-
N
e
t
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
me
d
i
c
a
l
i
m
a
g
e
se
g
me
n
t
a
t
i
o
n
,
”
Fr
o
n
t
i
e
rs
i
n
N
e
u
ro
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
6
,
Ju
n
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
8
9
/
f
n
i
n
f
.
2
0
2
2
.
9
1
1
6
7
9
.
[
9
]
O
.
R
o
n
n
e
b
e
r
g
e
r
,
P
.
F
i
s
c
h
e
r
,
a
n
d
T.
B
r
o
x
,
“
U
-
N
e
t
:
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
f
o
r
b
i
o
me
d
i
c
a
l
i
ma
g
e
se
g
m
e
n
t
a
t
i
o
n
,
”
Me
d
i
c
a
l
I
m
a
g
e
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
p
u
t
e
r
-
Assi
s
t
e
d
I
n
t
e
rv
e
n
t
i
o
n
–
MI
C
C
AI
2
0
1
5
,
p
p
.
2
3
4
–
2
4
1
,
2
0
1
5
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
2
4
5
7
4
-
4
_
2
8
.
[
1
0
]
J.
C
h
e
n
e
t
a
l
.
,
“
Tr
a
n
sU
N
e
t
:
t
r
a
n
sf
o
r
mers
ma
k
e
st
r
o
n
g
e
n
c
o
d
e
r
s
f
o
r
m
e
d
i
c
a
l
i
ma
g
e
s
e
g
me
n
t
a
t
i
o
n
,
”
a
rXi
v
-
C
o
m
p
u
t
e
r S
c
i
e
n
c
e
,
2
0
2
1
.
[
1
1
]
A
.
L
i
n
,
B
.
C
h
e
n
,
J
.
X
u
,
Z
.
Z
h
a
n
g
,
G
.
L
u
,
a
n
d
D
.
Z
h
a
n
g
,
“
D
S
-
T
r
a
n
s
U
N
e
t
:
d
u
a
l
s
w
i
n
t
r
a
n
s
f
o
r
m
e
r
U
-
N
e
t
f
o
r
m
e
d
i
c
a
l
i
m
a
g
e
s
e
g
m
e
n
t
a
t
i
o
n
,”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
n
s
t
r
u
m
e
n
t
a
t
i
o
n
a
n
d
M
e
a
s
u
r
e
m
e
n
t
,
v
o
l
.
7
1
,
p
p
.
1
–
1
5
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
I
M
.
2
0
2
2
.
3
1
7
8
9
9
1
.
[
1
2
]
E.
S
h
e
l
h
a
mer
,
J
.
Lo
n
g
,
a
n
d
T
.
D
a
r
r
e
l
l
,
“
F
u
l
l
y
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
f
o
r
s
e
ma
n
t
i
c
se
g
me
n
t
a
t
i
o
n
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
Pa
t
t
e
rn
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
3
9
,
n
o
.
4
,
p
p
.
6
4
0
–
6
5
1
,
A
p
r
.
2
0
1
7
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
1
6
.
2
5
7
2
6
8
3
.
[
1
3
]
H
.
Li
,
J
.
Li
u
,
a
n
d
D
.
W
a
n
g
,
“
A
f
a
st
i
n
st
a
n
c
e
s
e
g
me
n
t
a
t
i
o
n
t
e
c
h
n
i
q
u
e
f
o
r
l
o
g
e
n
d
f
a
c
e
s
b
a
se
d
o
n
m
e
t
r
i
c
l
e
a
r
n
i
n
g
,
”
Fo
res
t
s
,
v
o
l
.
1
4
,
n
o
.
4
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
f
1
4
0
4
0
7
9
5
.
[
1
4
]
S
.
S
a
h
u
,
H
.
S
a
r
m
a
,
a
n
d
D
.
J
.
B
o
r
a
,
“
I
m
a
g
e
s
e
g
m
e
n
t
a
t
i
o
n
a
n
d
i
t
s
d
i
f
f
e
r
e
n
t
t
e
c
h
n
i
q
u
e
s
:
a
n
i
n
-
d
e
p
t
h
a
n
a
l
y
s
i
s
,
”
i
n
2
0
1
8
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
R
e
s
e
a
r
c
h
i
n
I
n
t
e
l
l
i
g
e
n
t
a
n
d
C
o
m
p
u
t
i
n
g
i
n
E
n
g
i
n
e
e
r
i
n
g
(
R
I
C
E
)
,
A
u
g
.
2
0
1
8
,
p
p
.
1
–
7
,
d
o
i
:
1
0
.
1
1
0
9
/
R
I
C
E
.
2
0
1
8
.
8
5
0
9
0
3
8
.
[
1
5
]
S
.
M
i
n
a
e
e
,
Y
.
Y
.
B
o
y
k
o
v
,
F
.
P
o
r
i
k
l
i
,
A
.
J.
P
l
a
z
a
,
N
.
K
e
h
t
a
r
n
a
v
a
z
,
a
n
d
D
.
T
e
r
z
o
p
o
u
l
o
s
,
“
I
mag
e
se
g
m
e
n
t
a
t
i
o
n
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
:
a
s
u
r
v
e
y
,”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Pa
t
t
e
r
n
An
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
4
,
n
o
.
7
,
p
p
.
3
5
2
3
-
3
5
4
2
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
2
1
.
3
0
5
9
9
6
8
.
[
1
6
]
H
.
R
a
m
a
d
a
n
,
C
.
La
c
h
q
a
r
,
a
n
d
H
.
Ta
i
r
i
,
“
A
su
r
v
e
y
o
f
r
e
c
e
n
t
i
n
t
e
r
a
c
t
i
v
e
i
mag
e
se
g
me
n
t
a
t
i
o
n
m
e
t
h
o
d
s
,
”
C
o
m
p
u
t
a
t
i
o
n
a
l
Vi
s
u
a
l
Me
d
i
a
,
v
o
l
.
6
,
n
o
.
4
,
p
p
.
3
5
5
–
3
8
4
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
0
9
5
-
0
2
0
-
0177
-
5.
[
1
7
]
U
.
S
e
h
a
r
a
n
d
M
.
L.
N
a
s
e
e
m,
“
H
o
w
d
e
e
p
l
e
a
r
n
i
n
g
i
s
e
m
p
o
w
e
r
i
n
g
sem
a
n
t
i
c
s
e
g
me
n
t
a
t
i
o
n
,
”
M
u
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
1
,
n
o
.
2
1
,
p
p
.
3
0
5
1
9
–
3
0
5
4
4
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
022
-
1
2
8
2
1
-
3.
[
1
8
]
I
.
A
h
med
,
M
.
A
h
ma
d
,
F
.
A
.
K
h
a
n
,
a
n
d
M
.
A
si
f
,
“
C
o
m
p
a
r
i
so
n
o
f
d
e
e
p
-
l
e
a
r
n
i
n
g
-
b
a
se
d
se
g
m
e
n
t
a
t
i
o
n
mo
d
e
l
s:
u
s
i
n
g
t
o
p
v
i
e
w
p
e
r
s
o
n
i
ma
g
e
s
,
”
I
E
EE
A
c
c
e
ss
,
v
o
l
.
8
,
p
p
.
1
3
6
3
6
1
–
1
3
6
3
7
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
0
.
3
0
1
1
4
0
6
.
[
1
9
]
H
.
Zh
a
n
g
,
H
.
S
u
n
,
W
.
A
o
,
a
n
d
G
.
D
i
mi
r
o
v
sk
i
,
“
A
su
r
v
e
y
o
n
i
n
st
a
n
c
e
se
g
me
n
t
a
t
i
o
n
:
r
e
c
e
n
t
a
d
v
a
n
c
e
s
a
n
d
c
h
a
l
l
e
n
g
e
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
n
o
v
a
t
i
v
e
C
o
m
p
u
t
i
n
g
,
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
7
,
n
o
.
3
,
p
p
.
1
0
4
1
–
1
0
5
3
,
2
0
2
1
,
d
o
i
:
1
0
.
2
4
5
0
7
/
i
j
i
c
i
c
.
1
7
.
0
3
.
1
0
4
1
.
[
2
0
]
K
.
H
e
,
G
.
G
k
i
o
x
a
r
i
,
P
.
D
o
l
l
a
r
,
a
n
d
R
.
G
i
r
sh
i
c
k
,
“
M
a
s
k
R
-
C
N
N
,
”
i
n
2
0
1
7
I
E
EE
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
o
m
p
u
t
e
r
Vi
s
i
o
n
(
I
C
C
V)
,
O
c
t
.
2
0
1
7
,
p
p
.
2
9
8
0
–
2
9
8
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
1
7
.
3
2
2
.
[
2
1
]
M
u
s
y
a
r
o
f
a
h
,
V
.
S
c
h
mi
d
t
,
a
n
d
M
.
K
a
d
a
,
“
O
b
j
e
c
t
d
e
t
e
c
t
i
o
n
o
f
a
e
r
i
a
l
i
ma
g
e
u
si
n
g
m
a
sk
-
r
e
g
i
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
(
m
a
sk
R
-
C
N
N
)
,
”
I
O
P
C
o
n
f
e
r
e
n
c
e
S
e
ri
e
s:
E
a
rt
h
a
n
d
E
n
v
i
ro
n
m
e
n
t
a
l
S
c
i
e
n
c
e
,
J
u
n
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
5
5
-
1
3
1
5
/
5
0
0
/
1
/
0
1
2
0
9
0
.
[
2
2
]
K
.
L
i
n
e
t
a
l
.
,
“
F
a
c
e
d
e
t
e
c
t
i
o
n
a
n
d
s
e
g
me
n
t
a
t
i
o
n
b
a
se
d
o
n
i
mp
r
o
v
e
d
mas
k
R
-
C
N
N
,
”
D
i
s
c
re
t
e
D
y
n
a
m
i
c
s
i
n
N
a
t
u
re
a
n
d
S
o
c
i
e
t
y
,
v
o
l
.
2
0
2
0
,
n
o
.
1
,
p
p
.
1
–
1
1
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
0
/
9
2
4
2
9
1
7
.
[
2
3
]
L.
S
h
e
n
e
t
a
l
.
,
“
F
u
si
n
g
a
t
t
e
n
t
i
o
n
mec
h
a
n
i
sm
w
i
t
h
M
a
s
k
R
-
C
N
N
f
o
r
i
n
s
t
a
n
c
e
s
e
g
me
n
t
a
t
i
o
n
o
f
g
r
a
p
e
c
l
u
s
t
e
r
i
n
t
h
e
f
i
e
l
d
,
”
Fr
o
n
t
i
e
r
s
i
n
Pl
a
n
t
S
c
i
e
n
c
e
,
v
o
l
.
1
3
,
J
u
l
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
8
9
/
f
p
l
s.
2
0
2
2
.
9
3
4
4
5
0
.
[
2
4
]
C.
-
Y
.
H
s
u
,
R
.
H
u
,
Y
.
X
i
a
n
g
,
X
.
L
o
n
g
,
a
n
d
Z.
L
i
,
“
I
mp
r
o
v
i
n
g
t
h
e
D
e
e
p
l
a
b
v
3
+
m
o
d
e
l
w
i
t
h
a
t
t
e
n
t
i
o
n
m
e
c
h
a
n
i
s
ms
a
p
p
l
i
e
d
t
o
e
y
e
d
e
t
e
c
t
i
o
n
a
n
d
s
e
g
m
e
n
t
a
t
i
o
n
,
”
M
a
t
h
e
m
a
t
i
c
s
,
v
o
l
.
1
0
,
n
o
.
1
5
,
J
u
l
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
m
a
t
h
1
0
1
5
2
5
9
7
.
[
2
5
]
C
.
Z
h
a
n
g
,
J
.
Zh
a
o
,
a
n
d
Y
.
F
e
n
g
,
“
R
e
se
a
r
c
h
o
n
sem
a
n
t
i
c
s
e
g
me
n
t
a
t
i
o
n
b
a
se
d
o
n
i
m
p
r
o
v
e
d
P
S
P
N
e
t
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
t
e
l
l
i
g
e
n
t
P
e
rc
e
p
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r V
i
si
o
n
(
C
I
P
C
V)
,
M
a
y
2
0
2
3
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
I
P
C
V
5
8
8
8
3
.
2
0
2
3
.
0
0
0
1
2
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
J
y
o
th
sna
Cher
a
p
a
n
a
m
je
r
i
is
p
u
rsu
in
g
h
e
r
P
h
.
D
.
JN
TUA,
An
a
n
tap
u
r
,
m
a
ste
r
d
e
g
re
e
in
2
0
0
9
in
S
V
Un
iv
e
rsit
y
a
n
d
b
a
c
h
e
lo
r
d
e
g
re
e
in
2
0
0
5
i
n
J
NTUH
,
Hy
d
e
ra
b
a
d
,
An
d
h
ra
P
ra
d
e
sh
,
In
d
ia.
S
h
e
is
c
u
rre
n
tl
y
d
o
i
n
g
P
h
.
D.
in
JN
TUA,
An
a
n
ta
p
u
r.
He
r
a
re
a
o
f
in
tere
sts
i
s
a
rti
ficia
l
in
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
c
o
m
p
u
ter
v
isio
n
,
d
e
e
p
lea
rn
in
g
,
d
a
ta
sc
ien
c
e
,
a
n
d
I
o
T
.
S
h
e
h
a
s
1
5
p
l
u
s
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
.
G
ATE
q
u
a
li
fied
i
n
2
0
0
7
.
APRCE
T
q
u
a
li
fied
i
n
2
0
1
9
.
S
h
e
re
g
istere
d
f
o
r
G
o
o
g
l
e
S
c
h
o
lar
a
n
d
Re
se
a
rc
h
G
a
te
f
o
r
late
st
d
e
v
e
lo
p
m
e
n
ts
in
tec
h
n
o
l
o
g
y
.
S
h
e
h
a
s
5
p
u
b
l
ica
ti
o
n
s
in
t
h
e
fiel
d
o
f
a
rti
ficia
l
in
tell
ig
e
n
c
e
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
jy
o
t
h
sn
a
m
tec
h
@g
m
a
il
.
c
o
m
or
jy
o
th
s
n
a
_
c
se
@5
1
3
@y
a
h
o
o
.
c
o
m
.
B
.
Na
r
e
n
d
r
a
K
u
m
a
r
Ra
o
wa
s
o
b
tain
e
d
Ba
c
h
e
lo
r
De
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
E
n
g
i
n
e
e
rin
g
fr
o
m
Un
i
v
e
rsit
y
o
f
M
a
d
ra
s,
M
.
Tec
h
.
a
n
d
P
h
.
D.
in
c
o
m
p
u
ter
sc
ien
c
e
fro
m
JN
TU,
Hy
d
e
ra
b
a
d
.
He
h
a
s
m
o
re
th
a
n
2
2
y
e
a
rs
o
f
e
x
p
e
rien
c
e
in
a
r
e
a
o
f
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
e
n
g
in
e
e
rin
g
wh
i
c
h
i
n
c
lu
d
e
s
fo
u
r
y
e
a
rs
o
f
in
d
u
str
ial
e
x
p
e
rien
c
e
a
n
d
six
tee
n
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
.
Re
se
a
rc
h
in
tere
sts
i
n
c
lu
d
e
s
o
ftwa
re
e
n
g
i
n
e
e
rin
g
,
d
e
e
p
lea
rn
in
g
,
a
n
d
e
m
b
e
d
d
e
d
sy
ste
m
s
.
Cu
rre
n
tl
y
h
e
is
wo
rk
i
n
g
a
s
P
ro
fe
ss
o
r
a
n
d
He
a
d
,
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
i
n
e
e
rin
g
a
t
M
o
h
a
n
Ba
b
u
Un
i
v
e
rsity
(Er
stwh
il
e
S
re
e
Vid
y
a
n
ik
e
th
a
n
En
g
in
e
e
ri
n
g
C
o
ll
e
g
e
)
.
He
h
a
s
3
5
p
u
b
li
c
a
ti
o
n
s
in
h
is
c
r
e
d
it
t
il
l
d
a
te
i
n
re
p
u
te
d
jo
u
rn
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s
.
H
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
n
a
re
n
d
ra
k
u
m
a
rra
o
b
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.