I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
,
p
p
.
1
503
~
1
5
1
6
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
3
8
.
i
3
.
pp
1
5
0
3
-
1
5
1
6
1503
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
Remo
v
e gla
ss
es diff
usio
n mo
del
an i
nno
v
a
tive con
diti
o
ned of eye
g
la
ss
es remo
v
a
l wit
h ima
g
e di
ff
usio
n mo
del
Yuliza
1
,
Ra
chm
a
t
M
uwa
rdi
1
,
G
a
la
t
ia
E
rica
Yehez
k
iel
2
,
M
irna
Yu
nita
2
,
L
enni
3
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s
M
e
r
c
u
B
u
a
n
a
,
Ja
k
a
r
t
a
,
I
n
d
o
n
e
s
i
a
2
S
c
h
o
o
l
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
B
e
i
j
i
n
g
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
,
B
e
i
j
i
n
g
,
C
h
i
n
a
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s
M
u
h
a
mm
a
d
i
y
a
h
Ta
n
g
e
r
a
n
g
,
Ta
n
g
e
r
a
n
g
,
I
n
d
o
n
e
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ju
n
5
,
2
0
2
4
R
ev
is
ed
No
v
18
,
2
0
2
4
Acc
ep
ted
No
v
24
,
2
0
2
4
Th
e
p
re
se
n
c
e
o
f
e
y
e
g
las
se
s
in
fa
c
ial
ima
g
e
s
p
o
se
s
c
h
a
ll
e
n
g
e
s
fo
r
ima
g
e
p
ro
c
e
ss
in
g
,
p
a
rti
c
u
larly
i
n
fa
c
ial
re
c
o
g
n
i
ti
o
n
.
T
h
is
p
a
p
e
r
in
tro
d
u
c
e
s
th
e
re
m
o
v
e
g
las
se
s
d
iffu
sio
n
m
o
d
e
l
(
RG
DM),
a
c
o
n
d
it
i
o
n
e
d
d
e
n
o
isin
g
d
iffu
sio
n
p
ro
b
a
b
il
isti
c
m
o
d
e
l
(DD
P
M
)
d
e
s
ig
n
e
d
fo
r
p
re
c
ise
g
las
se
s
re
m
o
v
a
l.
RG
DM
e
m
p
lo
y
s
c
o
n
d
i
ti
o
n
a
l
m
o
d
e
li
n
g
to
f
o
c
u
s
o
n
t
h
e
g
las
se
s
re
g
io
n
w
h
il
e
se
a
m
les
sly
re
sto
rin
g
fa
c
ial
fe
a
tu
r
e
s.
An
e
y
e
s
p
o
siti
o
n
a
c
c
u
ra
c
y
m
e
c
h
a
n
ism
,
lev
e
ra
g
in
g
fa
c
ial
lan
d
m
a
rk
s,
e
n
s
u
re
s
a
c
c
u
ra
te
e
y
e
re
sto
ra
ti
o
n
p
o
s
t
-
re
m
o
v
a
l.
Co
m
p
re
h
e
n
si
v
e
e
v
a
lu
a
ti
o
n
s
o
n
t
h
e
Ce
leb
A
d
a
tas
e
t
d
e
m
o
n
stra
te
RG
DM’
s
su
p
e
rio
r
p
e
rfo
rm
a
n
c
e
,
a
c
h
iev
in
g
t
h
e
lo
we
st
F
ré
c
h
e
t
i
n
c
e
p
ti
o
n
d
ista
n
c
e
(F
ID)
o
f
2
7
.
0
9
a
n
d
lea
rn
e
d
p
e
rc
e
p
tu
a
l
ima
g
e
p
a
tch
sim
il
a
rit
y
(LP
I
P
S
)
o
f
0
.
2
9
9
,
o
u
t
p
e
rfo
rm
in
g
sta
te
-
of
-
t
h
e
-
a
rt
m
e
th
o
d
s
su
c
h
a
s
3
D
s
y
n
t
h
e
ti
c
,
c
y
c
le
-
c
o
n
siste
n
t
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
r
k
(
Cy
c
leG
AN
)
,
a
n
d
e
y
e
g
las
se
s
re
m
o
v
a
l
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
(
ERG
AN
)
.
Th
e
se
re
su
lt
s
h
ig
h
li
g
h
t
th
e
m
o
d
e
l’s
e
ffe
c
ti
v
e
n
e
ss
i
n
p
ro
d
u
c
in
g
n
a
t
u
ra
l
a
n
d
h
i
g
h
-
fi
d
e
l
it
y
fa
c
ial
re
c
o
n
stru
c
ti
o
n
s.
T
h
is
wo
r
k
a
d
v
a
n
c
e
s
g
las
se
s
re
m
o
v
a
l
tec
h
n
o
lo
g
y
a
n
d
u
n
d
e
rsc
o
re
s
th
e
sig
n
ifi
c
a
n
c
e
o
f
c
o
n
d
it
io
n
a
l
m
o
d
e
ls
i
n
ima
g
e
p
ro
c
e
ss
in
g
.
T
h
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
h
a
s
p
ra
c
ti
c
a
l
imp
li
c
a
ti
o
n
s
fo
r
fa
c
ial
re
c
o
g
n
it
io
n
a
n
d
ima
g
e
e
n
h
a
n
c
e
m
e
n
t,
p
a
v
i
n
g
th
e
wa
y
fo
r
m
o
r
e
a
c
c
u
ra
te
a
n
d
ro
b
u
st
re
a
l
-
wo
rld
a
p
p
li
c
a
ti
o
n
s.
K
ey
w
o
r
d
s
:
C
o
n
d
itio
n
al
m
o
d
elin
g
Dee
p
lear
n
in
g
Dif
f
u
s
io
n
m
o
d
el
Facial
im
ag
es
Glass
e
s
r
em
o
v
al
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
:
Yu
liza
Dep
ar
tm
en
t o
f
E
lectr
ical
E
n
g
i
n
ee
r
in
g
,
U
n
iv
er
s
itas
Me
r
cu
B
u
an
a
J
ak
ar
ta,
I
n
d
o
n
esia
E
m
ail: y
u
liza@
m
er
cu
b
u
an
a
.
a
c.
id
1.
I
NT
RO
D
UCT
I
O
N
Facial
im
ag
es
ar
e
cr
itical
v
is
u
al
d
ata
s
o
u
r
ce
s
in
v
a
r
io
u
s
a
p
p
licatio
n
s
,
in
clu
d
in
g
f
ac
ial
r
ec
o
g
n
itio
n
,
im
ag
e
an
aly
s
is
,
an
d
d
ig
ital
co
m
m
u
n
icatio
n
.
Ho
wev
er
,
t
h
e
p
r
esen
ce
o
f
ey
e
g
lass
es
in
f
ac
ial
im
ag
es
in
tr
o
d
u
ce
s
s
ig
n
if
ican
t
ch
allen
g
es,
p
a
r
ticu
lar
ly
in
ac
c
u
r
ately
r
e
p
r
es
en
tin
g
f
ac
ial
f
ea
t
u
r
es,
wh
ic
h
ar
e
cr
u
cial
f
o
r
ap
p
licatio
n
s
s
u
ch
as
s
ec
u
r
ity
s
y
s
tem
s
,
wh
er
e
p
r
ec
is
e
id
en
tific
atio
n
is
p
ar
am
o
u
n
t.
R
ef
lectio
n
s
o
n
th
e
g
lass
s
u
r
f
ac
e,
len
s
d
is
to
r
tio
n
s
,
an
d
d
iv
er
s
e
f
r
am
e
s
ty
les
co
n
tr
ib
u
t
e
to
v
is
u
al
ar
tifa
cts
th
at
o
b
s
tr
u
ct
ac
c
u
r
ate
f
ac
ial
r
ec
o
g
n
itio
n
[
1
]
,
[
2
]
.
I
n
ap
p
licatio
n
s
s
u
ch
as
s
ec
u
r
ity
s
y
s
tem
s
,
wh
er
e
p
r
ec
is
e
id
en
tific
ati
o
n
is
p
a
r
am
o
u
n
t,
t
h
e
p
r
esen
ce
o
f
g
lass
es
b
ec
o
m
es
a
s
ig
n
if
ican
t
b
o
ttlen
ec
k
[
3
]
.
Facial
r
ec
o
g
n
itio
n
tec
h
n
o
lo
g
y
[
3
]
,
[
4
]
also
k
n
o
wn
as
FR
T
s
,
ar
e
o
n
e
o
f
s
ev
er
al
b
io
m
etr
ic
to
o
ls
o
r
m
o
d
alities
d
e
v
elo
p
ed
to
d
etec
t
an
d
id
en
tif
y
p
er
s
o
n
s
wh
e
n
th
ei
r
p
h
o
to
s
ar
e
tak
en
b
y
a
ca
m
er
a
len
s
[
5
]
,
[
6
]
.
I
t
is
a
co
r
n
er
s
to
n
e
in
c
o
n
tem
p
o
r
ar
y
s
ec
u
r
it
y
an
d
id
en
tific
atio
n
s
y
s
tem
s
,
[
7
]
-
[
9
]
wh
ich
r
elies
h
ea
v
ily
o
n
th
e
ac
c
u
r
ate
r
e
p
r
esen
tatio
n
o
f
f
ac
ial
f
ea
tu
r
es.
T
h
e
p
r
esen
ce
o
f
g
lass
es
co
m
p
licates th
is
p
r
o
ce
s
s
,
lead
in
g
to
p
o
ten
tial m
is
id
en
tific
atio
n
s
an
d
r
e
d
u
ce
d
s
y
s
tem
ef
f
ica
cy
[
1
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
:
1
5
0
3
-
1
5
1
6
1504
I
m
ag
e
in
p
ai
n
tin
g
h
as
s
lo
wly
g
r
o
wn
in
im
p
o
r
tan
ce
as
a
s
tu
d
y
ar
ea
in
d
i
g
ital
im
ag
e
p
r
o
ce
s
s
in
g
d
u
e
to
th
e
ad
v
an
ce
m
en
t
o
f
s
cien
ce
a
n
d
tech
n
o
lo
g
y
,
th
e
r
a
p
id
u
p
g
r
ad
in
g
o
f
h
ar
d
war
e,
an
d
th
e
d
e
s
ir
e
f
o
r
h
ig
h
-
q
u
ality
im
ag
es
[
1
1
]
.
I
t
ca
n
r
esto
r
e
t
h
e
tex
tu
r
e
o
f
an
im
ag
e
,
ex
tr
ac
t
its
h
ig
h
-
lev
el
ab
s
tr
ac
t
p
r
o
p
er
ties
,
as
well
as
s
em
an
tic
im
ag
er
y
lik
e
h
u
m
a
n
f
ac
es
[
1
2
]
,
wh
ich
is
wh
y
im
ag
e
in
p
ai
n
tin
g
is
o
f
ten
u
s
ed
i
n
r
esear
ch
es
wh
er
e
im
ag
e
g
en
e
r
atio
n
is
n
ec
ess
ar
y
,
s
u
ch
as
th
e
r
em
o
v
al
o
f
g
la
s
s
es.
C
o
cc
ia
[
1
3
]
em
p
h
asized
in
2
0
1
7
,
p
r
o
b
lem
-
d
r
iv
en
in
n
o
v
atio
n
s
o
f
ten
lead
to
s
o
lu
tio
n
s
th
at
p
r
o
v
id
e
a
c
o
m
p
etitiv
e
ed
g
e.
T
h
e
ev
o
lu
tio
n
o
f
tech
n
o
lo
g
y
,
as
d
is
cu
s
s
ed
b
y
C
o
cc
ia
in
2
0
1
9
[
1
4
]
,
[
1
5
]
,
in
v
o
l
v
es
th
e
co
n
t
in
u
o
u
s
s
u
b
s
titu
tio
n
o
f
o
ld
er
m
eth
o
d
s
with
m
o
r
e
ef
f
i
cien
t
o
n
es.
I
n
th
is
co
n
tex
t,
o
u
r
r
esear
ch
r
ep
r
esen
ts
an
in
cr
em
en
tal
in
n
o
v
atio
n
,
b
u
il
d
in
g
u
p
o
n
ex
is
tin
g
g
en
er
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
(
GAN
)
-
b
ased
a
p
p
r
o
ac
h
es
a
n
d
ev
o
lv
in
g
t
o
war
d
s
a
m
o
r
e
ef
f
ec
tiv
e
d
if
f
u
s
io
n
m
o
d
el
f
r
am
ewo
r
k
.
W
h
ile
p
r
e
v
io
u
s
s
tu
d
ies,
s
u
ch
as
th
e
o
n
e
co
n
d
u
cted
in
2
0
2
3
b
y
Sad
i
k
-
E
-
T
awh
ee
d
et
a
l.
[
1
6
]
,
h
a
v
e
f
o
cu
s
ed
o
n
r
em
o
v
in
g
s
p
ec
if
ic
a
r
tifa
cts
lik
e
g
la
r
e
an
d
r
ef
lectio
n
s
,
o
u
r
ap
p
r
o
a
ch
tack
les
th
e
r
o
o
t
ca
u
s
e
b
y
r
em
o
v
in
g
th
e
g
lass
es
th
em
s
elv
es.
T
h
is
co
m
p
r
eh
en
s
iv
e
s
o
lu
tio
n
n
o
t
o
n
ly
en
h
an
ce
s
th
e
clar
ity
o
f
f
ac
ial
im
ag
es b
u
t a
ls
o
im
p
r
o
v
e
s
f
ac
ial
r
ec
o
g
n
itio
n
a
cc
u
r
ac
y
,
as h
ig
h
lig
h
ted
b
y
Ma
o
et
a
l
.
[
1
7
]
in
2
0
2
1
.
T
h
is
p
ap
er
in
tr
o
d
u
ce
s
th
e
r
e
m
o
v
e
g
lass
es
d
if
f
u
s
io
n
m
o
d
el
(
R
GDM
)
,
a
n
o
v
el
ap
p
r
o
ac
h
t
h
at
d
ep
ar
ts
f
r
o
m
tr
ad
itio
n
al
GAN
-
b
ased
f
r
am
ewo
r
k
s
b
y
lev
er
a
g
in
g
a
d
en
o
is
in
g
d
if
f
u
s
io
n
p
r
o
b
a
b
ilis
ti
c
m
o
d
el
(
DDPM)
,
wh
ich
r
ep
r
esen
ts
a
s
ig
n
if
ican
t
ad
v
an
ce
m
en
t
in
th
is
ar
ea
.
T
h
e
R
GDM
in
teg
r
ates
a
co
n
d
itio
n
in
g
m
o
d
u
le
th
at
f
o
cu
s
es
s
p
ec
if
ically
o
n
th
e
g
lass
es
ar
ea
,
en
ab
lin
g
p
r
ec
is
e
in
p
ain
tin
g
an
d
p
r
eser
v
in
g
th
e
in
teg
r
ity
o
f
o
th
e
r
f
ac
ial
f
ea
tu
r
es.
A
d
d
itio
n
ally
,
an
in
n
o
v
ativ
e
e
y
es
p
o
s
itio
n
ac
cu
r
ac
y
m
ec
h
a
n
is
m
is
in
co
r
p
o
r
ate
d
,
u
tili
zin
g
lan
d
m
ar
k
in
f
o
r
m
atio
n
to
en
s
u
r
e
th
e
g
en
er
ate
d
e
y
es
alig
n
p
er
f
ec
tly
with
th
eir
o
r
ig
in
al
p
o
s
itio
n
s
.
T
h
is
a
p
p
r
o
ac
h
n
o
t
o
n
ly
ad
d
r
ess
es
th
e
lim
itati
o
n
s
o
f
p
r
ev
i
o
u
s
m
eth
o
d
s
b
u
t
a
ls
o
o
f
f
er
s
a
m
o
r
e
s
ca
lab
le
an
d
ad
ap
tab
le
s
o
lu
tio
n
f
o
r
g
lass
es r
em
o
v
al
in
d
iv
er
s
e
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
[
1
8
]
.
I
n
th
is
s
tu
d
y
,
we
u
tili
ze
th
e
aid
o
f
s
eg
m
en
t
an
y
th
in
g
m
o
d
el
(
SAM)
[
1
9
]
th
at
is
s
o
lely
tr
ain
ed
o
n
r
ec
o
g
n
izin
g
g
lass
es
in
s
eg
m
en
tin
g
an
d
m
ask
in
g
t
h
e
g
lass
es.
W
e
th
en
co
m
b
in
e
th
e
u
s
e
o
f
d
if
f
u
s
io
n
m
o
d
el
an
d
a
co
n
d
itio
n
in
g
m
o
d
u
le
in
o
r
d
er
to
o
n
l
y
c
o
n
d
u
ct
th
e
in
p
ai
n
tin
g
in
th
e
m
ask
ed
g
lass
es
ar
ea
to
r
em
o
v
e
th
e
g
lass
es
f
r
o
m
th
e
o
r
ig
in
al
im
ag
es.
At
t
h
is
s
tag
e,
af
ter
th
e
in
p
ain
tin
g
ar
ea
h
as
b
ee
n
co
n
f
ir
m
e
d
b
y
th
e
co
n
d
itio
n
in
g
m
o
d
u
le,
th
e
g
la
s
s
es
ar
e
r
em
o
v
ed
th
r
o
u
g
h
im
ag
e
in
p
ain
tin
g
u
s
in
g
th
e
d
i
f
f
u
s
io
n
m
o
d
el.
T
h
e
ab
ilit
y
to
ac
c
u
r
ately
an
d
g
r
ac
e
f
u
lly
r
e
m
o
v
e
g
lass
es
f
r
o
m
f
ac
ial
im
ag
es
ad
d
r
ess
es
a
co
m
m
o
n
v
i
s
u
al
c
h
allen
g
e.
T
h
is
tech
n
o
lo
g
y
h
as
wid
esp
r
e
ad
ap
p
licatio
n
s
in
p
h
o
to
g
r
ap
h
y
,
p
er
s
o
n
al
im
a
g
e
en
h
an
ce
m
e
n
t,
an
d
e
v
en
f
ac
ial
r
ec
o
g
n
itio
n
,
wh
er
e
clea
r
,
g
l
ass
es
-
f
r
ee
im
ag
es
ar
e
o
f
ten
p
r
ef
er
r
ed
.
T
h
er
ef
o
r
e,
in
o
r
d
er
to
p
r
eser
v
e
th
e
in
d
iv
id
u
al
s
ty
le
a
n
d
f
ac
ial
f
e
atu
r
es,
we
also
co
r
r
o
b
o
r
ated
an
ad
d
itio
n
al
m
o
d
if
icatio
n
t
o
war
d
s
th
e
p
r
o
p
o
s
ed
m
eth
o
d
,
wh
ich
is
t
h
e
ey
es
p
o
s
itio
n
ac
cu
r
ac
y
[
2
0
]
.
T
h
r
o
u
g
h
t
h
e
g
i
v
en
d
ataset
attr
ib
u
tes,
th
e
attr
ib
u
tes
r
esp
o
n
s
ib
le
f
o
r
th
e
ey
es’
p
o
s
it
io
n
s
is
also
in
s
er
ted
i
n
to
th
e
m
o
d
el,
w
h
ich
p
r
o
v
i
d
es
th
e
n
e
ce
s
s
ar
y
in
f
o
r
m
atio
n
f
o
r
th
e
m
o
d
el
to
g
en
er
ate
t
h
e
s
u
p
p
o
s
ed
e
y
es in
th
e
c
o
r
r
ec
t p
o
s
itio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el,
r
em
o
v
e
g
lass
es
d
if
f
u
s
io
n
m
o
d
el
,
s
tan
d
s
at
th
e
f
o
r
e
f
r
o
n
t
o
f
ey
eg
lass
es
r
em
o
v
al
tech
n
iq
u
es.
Ou
r
p
r
im
ar
y
g
o
al
is
to
s
ig
n
if
ican
tly
en
h
an
c
e
f
ac
e
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
b
y
s
y
s
tem
atica
lly
r
em
o
v
in
g
ey
eg
lass
es f
r
o
m
f
ac
ial
im
ag
es.
T
h
e
m
ain
c
o
n
tr
ib
u
tio
n
s
o
f
o
u
r
m
o
d
el
in
clu
d
e:
-
Ou
r
p
r
o
p
o
s
ed
m
o
d
el
lev
er
ag
e
s
th
e
DDPM
a
s
its
f
o
u
n
d
atio
n
al
f
r
am
ewo
r
k
,
en
s
u
r
in
g
ef
f
e
ctiv
e
d
en
o
i
s
in
g
th
r
o
u
g
h
a
d
i
f
f
u
s
io
n
p
r
o
ce
s
s
tailo
r
ed
f
o
r
in
tr
icate
d
ata
d
is
tr
ib
u
tio
n
s
.
-
T
h
e
in
co
r
p
o
r
atio
n
o
f
a
co
n
d
itio
n
al
n
e
two
r
k
em
p
lo
y
in
g
b
in
ar
y
m
ask
s
to
allo
w
f
o
r
a
m
eticu
lo
u
s
an
d
ac
cu
r
ate
id
en
tific
atio
n
o
f
e
y
eg
lass
es
-
o
cc
lu
d
ed
r
eg
io
n
s
d
u
r
in
g
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
.
-
T
h
e
m
o
d
el
f
ea
tu
r
es
a
r
o
b
u
s
t
ey
e
p
r
ec
is
io
n
ac
cu
r
ac
y
,
le
v
er
ag
ed
f
r
o
m
th
e
C
eleb
A
d
at
aset
'
s
in
tr
in
s
ic
lan
d
m
ar
k
attr
ib
u
tes,
in
o
r
d
er
t
o
en
s
u
r
e
th
e
p
r
ec
is
e
p
o
s
itio
n
i
n
g
o
f
th
e
g
en
e
r
ated
e
y
es,
m
ai
n
tain
in
g
th
e
m
in
th
e
ex
ac
t sam
e
p
o
s
itio
n
as in
t
h
e
o
r
ig
in
al
im
a
g
e.
T
h
e
s
u
b
s
eq
u
en
t
s
ec
tio
n
s
o
f
t
h
is
p
ap
er
will
elab
o
r
ate
o
n
t
h
e
R
GDM
ap
p
r
o
ac
h
,
b
e
g
in
n
i
n
g
with
a
d
is
cu
s
s
io
n
o
f
r
elate
d
w
o
r
k
a
n
d
th
e
th
e
o
r
etica
l
f
o
u
n
d
atio
n
o
f
d
if
f
u
s
io
n
m
o
d
els
in
s
ec
t
io
n
2
an
d
3
.
T
h
e
m
eth
o
d
o
l
o
g
y
in
s
ec
tio
n
4
will
d
etail
th
e
R
GDM
f
r
am
ewo
r
k
,
in
clu
d
in
g
th
e
co
n
d
itio
n
in
g
m
o
d
u
le,
t
h
e
ey
e
s
p
o
s
itio
n
ac
cu
r
ac
y
m
ec
h
a
n
is
m
,
d
ataset
d
etails,
im
p
lem
en
tatio
n
s
p
ec
if
ics,
an
d
ev
alu
atio
n
m
etr
ics.
Sectio
n
5
will
p
r
esen
t
th
e
ex
p
er
im
en
tal
r
esu
lts
o
f
o
u
r
m
o
d
el.
T
h
e
d
is
cu
s
s
io
n
in
s
ec
tio
n
6
will
co
m
p
ar
e
R
GDM
ag
ain
s
t
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
,
d
e
m
o
n
s
tr
atin
g
its
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
g
lass
es
r
em
o
v
al.
Fin
ally
,
th
e
p
a
p
er
will
co
n
clu
d
e
o
n
th
e
im
p
licatio
n
s
o
f
th
ese
f
in
d
in
g
s
f
o
r
f
u
tu
r
e
r
esear
ch
an
d
p
r
ac
tical
ap
p
licatio
n
s
in
s
ec
tio
n
7
.
2.
RE
L
AT
E
D
WO
RK
2
.
1
.
Dif
f
us
io
n
m
o
del
Sin
ce
t
h
e
p
u
b
licatio
n
o
f
d
if
f
u
s
io
n
m
o
d
el,
th
e
u
tili
za
tio
n
o
f
s
aid
m
eth
o
d
in
im
ag
e
p
r
o
ce
s
s
in
g
,
p
ar
ticu
lar
ly
im
ag
e
g
en
e
r
atio
n
o
r
im
ag
e
i
n
p
ain
tin
g
,
h
as
b
ee
n
s
tead
ily
r
is
in
g
.
No
t
f
ew
p
u
b
licatio
n
s
h
av
e
also
b
ee
n
p
u
b
lis
h
ed
with
th
e
f
o
c
u
s
o
f
co
m
p
ar
in
g
im
ag
e
g
e
n
er
atio
n
b
ased
o
n
d
if
f
u
s
io
n
m
o
d
el
t
o
im
ag
e
g
e
n
er
atio
n
b
ased
o
n
p
r
e
v
io
u
s
k
n
o
wn
m
o
d
els,
s
u
ch
as
GAN.
Fo
cu
s
in
g
o
n
p
u
b
licatio
n
s
r
elea
s
ed
in
r
ec
en
t
y
ea
r
s
,
Dh
ar
iwal
an
d
Nich
o
l
[
2
1
]
co
m
p
ar
ed
d
if
f
u
s
io
n
m
o
d
el
to
c
u
r
r
e
n
t
s
tate
-
of
-
th
e
-
a
r
t
GAN
m
o
d
els
in
im
ag
e
s
y
n
th
esis
.
T
h
ey
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
emo
ve
g
la
s
s
es d
iffu
s
io
n
mo
d
el
a
n
in
n
o
va
tive
co
n
d
itio
n
ed
o
f e
ye
g
la
s
s
es remo
va
l wi
th
…
(
Yu
liz
a
)
1505
co
n
d
u
cte
d
th
e
ex
p
e
r
im
en
tatio
n
s
o
n
b
o
th
u
n
c
o
n
d
itio
n
al
an
d
co
n
d
itio
n
al
im
a
g
e
s
y
n
t
h
esis
,
wh
er
e
d
if
f
u
s
io
n
m
o
d
el
p
r
o
v
es
th
at
t
h
e
im
a
g
e
s
y
n
th
esis
it
g
en
er
ates
ac
h
iev
e
s
u
p
er
io
r
im
a
g
e
s
am
p
le
q
u
alit
y
co
m
p
ar
ed
to
th
e
GANs
m
o
d
el.
I
n
2
0
2
2
,
L
u
g
m
a
y
r
et
a
l
.
[
2
2
]
p
r
o
p
o
s
ed
R
ePain
t
as
an
in
p
ain
tin
g
ap
p
r
o
ac
h
b
a
s
ed
o
n
DDPM
th
at
ca
n
b
e
u
s
ed
with
m
ask
s
.
T
h
e
au
th
o
r
s
o
n
ly
alter
t
h
e
iter
atio
n
s
o
f
th
e
r
ev
er
s
e
d
if
f
u
s
io
n
an
d
d
o
es
n
o
t
co
n
d
u
ct
an
y
m
o
d
if
icatio
n
s
o
r
co
n
d
itio
n
in
g
s
o
n
th
e
o
r
ig
in
al
DDPM
m
o
d
el,
w
h
ich
p
r
o
v
es
to
g
en
er
ate
d
iv
er
s
e
a
n
d
h
ig
h
-
q
u
ality
im
ag
es
as
a
r
esu
lt
a
n
d
o
u
tp
er
f
o
r
m
s
s
tate
-
of
-
th
e
-
ar
t
GAN
an
d
au
to
r
e
g
r
ess
iv
e
m
o
d
els.
I
n
2
0
2
3
,
Xie
et
a
l.
[
2
3
]
p
r
o
p
o
s
ed
Sm
ar
t
B
r
u
s
h
f
o
r
o
b
ject
i
n
p
ain
tin
g
g
u
id
ed
b
y
tex
t
an
d
s
h
ap
e
b
ased
o
n
d
if
f
u
s
io
n
m
o
d
el.
T
h
e
au
th
o
r
s
also
p
r
o
p
o
s
ed
a
n
o
v
el
o
b
ject
-
m
ask
p
r
ed
ictio
n
with
th
e
g
o
al
o
f
b
etter
p
r
es
er
v
in
g
th
e
im
a
g
e’
s
b
ac
k
g
r
o
u
n
d
.
2
.
2
.
I
ma
g
e
i
np
a
inting
I
m
ag
e
in
p
ain
tin
g
is
th
e
p
r
o
ce
s
s
o
f
f
in
is
h
in
g
o
r
r
esto
r
in
g
a
m
is
s
in
g
p
o
r
tio
n
o
f
a
n
im
ag
e,
o
r
elim
in
atin
g
ce
r
tain
elem
en
ts
o
f
a
n
im
ag
e
th
at
wer
e
i
n
itially
in
s
er
ted
in
to
it
[
2
4
]
.
T
h
e
tr
a
d
itio
n
al
m
eth
o
d
s
o
f
im
ag
e
in
p
ain
ti
n
g
a
r
e
ca
teg
o
r
i
ze
d
in
to
th
r
ee
g
r
o
u
p
s
:
d
if
f
u
s
io
n
-
b
ased
tech
n
iq
u
es,
p
atch
-
b
a
s
ed
tech
n
iq
u
es,
an
d
co
n
v
o
l
u
tio
n
f
ilter
-
b
ased
tech
n
iq
u
es,
wh
ile
th
e
m
o
d
er
n
a
p
p
r
o
ac
h
r
elies
m
o
r
e
u
p
o
n
d
ee
p
lear
n
in
g
-
b
ased
m
eth
o
d
s
[
2
5
]
.
T
h
a
n
k
s
to
th
e
ad
v
an
ce
m
e
n
t
o
f
im
ag
e
p
r
o
c
ess
in
g
to
o
ls
an
d
th
e
f
lex
ib
ili
ty
o
f
d
ig
ital
im
ag
e
ed
itin
g
,
au
t
o
m
atic
im
ag
e
in
p
a
in
tin
g
h
as
f
o
u
n
d
s
ig
n
if
ican
t
a
p
p
licatio
n
s
in
c
o
m
p
u
ter
v
is
io
n
an
d
h
as
g
r
o
wn
t
o
b
e
a
s
ig
n
if
ican
t a
n
d
d
if
f
icu
lt a
r
ea
o
f
r
esear
c
h
in
im
ag
e
p
r
o
ce
s
s
in
g
[
2
6
]
,
[
2
7
]
.
Fo
cu
s
in
g
o
n
th
e
u
s
ag
e
o
f
im
a
g
e
in
p
ain
tin
g
f
o
r
g
lass
es
r
em
o
v
al,
th
e
f
ir
s
t
m
eth
o
d
was
p
r
o
p
o
s
ed
b
y
Par
k
et
a
l.
[
2
8
]
in
2
0
0
5
u
s
in
g
r
ec
u
r
s
iv
e
er
r
o
r
co
m
p
en
s
atio
n
.
T
h
is
m
eth
o
d
b
e
g
in
s
b
y
id
e
n
tif
y
in
g
t
h
e
ar
ea
s
th
at
th
e
g
lass
es
h
av
e
o
b
s
c
u
r
ed
b
e
f
o
r
e
cr
ea
tin
g
a
f
ac
e
im
ag
e
d
e
v
o
id
o
f
g
lass
es
with
th
e
h
elp
o
f
r
ec
u
r
s
iv
e
er
r
o
r
co
m
p
en
s
atio
n
u
s
in
g
p
r
in
cip
al
co
m
p
o
n
en
t a
n
aly
s
is
(
PC
A
)
r
ec
o
n
s
tr
u
ctio
n
.
T
h
e
r
esu
ltan
t im
a
g
e
is
d
ev
o
id
o
f
an
y
tr
ac
es
o
f
th
e
g
lass
es
'
f
r
am
e,
r
ef
lectio
n
,
o
r
s
h
ad
in
g
,
s
u
g
g
es
tin
g
th
at
th
e
m
eth
o
d
s
u
cc
ess
f
u
lly
ad
d
r
ess
es
th
e
is
s
u
e
o
f
g
lass
es
o
cc
lu
s
io
n
alth
o
u
g
h
th
e
r
e
ar
e
s
till
s
o
m
e
lim
it
atio
n
s
o
f
r
em
o
v
in
g
g
lass
es
wi
th
d
ar
k
f
r
am
es
an
d
g
r
ad
ated
len
s
co
lo
r
s
.
Fiv
e
y
ea
r
s
later
,
W
an
g
et
a
l.
[
2
9
]
p
r
o
p
o
s
ed
a
m
eth
o
d
to
r
e
m
o
v
e
g
lass
es
f
r
o
m
f
ac
e
im
ag
es
b
ased
o
n
th
e
ac
tiv
e
ap
p
ea
r
an
ce
m
o
d
el
(
AAM
)
[
3
0
]
.
T
h
e
m
o
d
el
is
b
u
ilt
b
y
m
er
g
in
g
s
h
ap
e
a
n
d
ap
p
ea
r
an
ce
v
ar
ian
ts
in
a
s
h
ap
e
-
n
o
r
m
alize
d
f
ash
io
n
.
T
h
e
r
esu
lts
o
f
th
e
ex
p
er
im
e
n
ts
r
ev
ea
l
th
at
th
e
s
u
g
g
ested
m
eth
o
d
is
an
ef
f
ec
tiv
e
s
o
lu
tio
n
f
o
r
r
ec
o
g
n
izin
g
f
ac
es
o
b
s
cu
r
ed
b
y
th
ick
-
r
im
m
ed
ey
e
g
l
ass
es,
wh
ich
was
a
n
o
tab
le
p
r
o
b
lem
with
[
2
8
]
m
eth
o
d
.
An
o
t
h
er
f
iv
e
y
ea
r
s
af
ter
,
L
ian
g
et
a
l.
[
3
1
]
p
r
o
p
o
s
ed
a
g
lass
es
d
etec
t
io
n
m
o
d
el
th
at
was
b
ased
o
n
Z
h
u
an
d
R
am
an
an
[
3
2
]
co
n
ce
p
t
o
f
tr
ee
-
p
icto
r
ial
s
tr
u
ctu
r
e.
A
d
o
u
b
le
-
lay
e
r
ed
f
ilter
m
ad
e
o
f
in
p
ain
tin
g
a
n
d
d
ee
p
lear
n
in
g
is
u
s
ed
to
r
e
m
o
v
e
th
e
g
lass
es.
Af
ter
co
n
d
u
ctin
g
a
co
m
p
r
e
h
en
s
iv
e
i
n
v
esti
g
atio
n
,
th
e
a
u
th
o
r
s
d
is
co
v
er
ed
a
d
r
awb
ac
k
o
f
th
eir
ap
p
r
o
ac
h
.
Giv
e
n
th
at
th
e
e
d
g
e
in
f
o
r
m
atio
n
o
n
g
lass
es
with
o
u
t
r
im
s
is
to
o
"we
ak
"
to
id
en
tif
y
th
em
as
g
lass
es,
th
is
af
f
ec
ts
th
e
d
etec
tio
n
r
ate
f
o
r
s
u
ch
in
d
iv
id
u
als.
I
n
r
ec
en
t
y
ea
r
s
,
a
n
o
v
el
im
ag
e
-
to
-
im
ag
e
GAN
[
3
3
]
f
r
a
m
ewo
r
k
f
o
r
ey
eg
lass
es
r
em
o
v
al,
ca
lled
B
y
eGla
s
s
e
s
GAN,
is
p
r
o
p
o
s
ed
b
y
L
ee
an
d
L
ai
in
2
0
2
0
[
3
4
]
.
T
h
e
co
m
p
o
n
e
n
ts
o
f
B
y
eGla
s
s
e
s
GAN
ar
e
an
en
co
d
er
,
a
s
eg
m
en
tatio
n
d
ec
o
d
er
,
an
d
a
f
ac
e
d
ec
o
d
er
.
T
h
e
ex
p
er
im
en
ts
d
em
o
n
s
tr
ate
th
at
ev
en
f
o
r
s
em
i
-
tr
an
s
p
ar
en
t
co
lo
r
ey
e
g
lass
es
o
r
s
p
ec
tacle
s
with
g
lar
e,
B
y
e
Glass
e
s
GAN
m
ay
p
r
o
d
u
ce
ae
s
th
etica
lly
p
leasin
g
r
esu
lts
in
th
e
f
ac
ial
im
ag
es
wi
th
th
e
g
lass
e
s
r
em
o
v
ed
.
T
h
en
,
Hu
et
a
l
.
[
3
5
]
p
r
o
p
o
s
ed
th
e
e
y
eg
lass
es
r
em
o
v
al
g
en
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
(
E
R
GAN)
,
a
u
n
if
ied
ey
eg
lass
r
em
o
v
al
m
o
d
el,
in
2
0
2
1
.
T
h
i
s
m
eth
o
d
s
u
g
g
ested
a
GAN
-
b
ased
ar
ch
itectu
r
e
to
r
em
o
v
e
v
ar
i
o
u
s
k
in
d
s
o
f
g
lass
e
s
in
th
e
wild
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
lear
n
s
to
s
wap
th
e
ey
e
r
eg
io
n
in
two
f
ac
es
w
h
en
g
i
v
en
two
im
ag
es
o
f
f
ac
e
s
wea
r
in
g
an
d
n
o
t
wea
r
in
g
ey
eg
lass
es.
T
h
at
s
am
e
y
ea
r
,
C
h
en
g
an
d
C
ao
[
3
6
]
p
r
o
p
o
s
ed
an
o
th
er
m
eth
o
d
u
s
in
g
t
h
e
s
am
e
ac
r
o
n
y
m
,
ca
lled
E
R
GAN
(
h
ig
h
p
er
f
o
r
m
GAN
f
o
r
ey
e
g
lass
es
r
em
o
v
al
)
.
T
h
is
m
eth
o
d
is
b
ased
o
n
th
e
s
u
cc
ess
f
u
l
im
p
lem
en
tatio
n
o
f
th
e
f
ac
e
attr
ib
u
te
ed
itin
g
task
in
t
h
e
GAN
.
I
t
i
n
tr
o
d
u
ce
s
th
e
m
o
r
e
s
o
p
h
is
tica
ted
GAN
in
v
er
s
io
n
m
o
d
el
I
DI
n
v
er
t
[
3
7
]
an
d
th
e
I
n
ter
f
ac
eGA
N
[
3
8
]
m
et
h
o
d
.
T
h
e
f
in
d
in
g
s
o
f
th
e
ex
p
er
im
e
n
t
in
d
icate
th
at
th
is
m
eth
o
d
h
elp
s
to
in
cr
ea
s
e
th
e
r
em
o
v
al
ac
cu
r
ac
y
o
f
g
lass
es.
Ho
wev
er
,
it
is
a
ls
o
clea
r
f
r
o
m
th
e
ex
p
er
im
en
tal
f
in
d
i
n
g
s
th
at
th
is
m
eth
o
d
r
eq
u
ir
es i
m
p
r
o
v
em
e
n
t in
ter
m
s
o
f
h
o
w
r
ea
l
-
w
o
r
ld
im
a
g
es a
r
e
p
r
o
ce
s
s
ed
,
as it f
alls
s
h
o
r
t in
s
ev
er
al
asp
ec
ts
.
A
n
o
t
h
e
r
a
p
p
r
o
a
c
h
p
r
o
p
o
s
e
d
t
h
a
t
s
a
m
e
y
e
a
r
i
s
t
h
e
m
u
l
t
i
m
o
d
a
l
a
s
y
m
m
e
t
r
i
c
d
u
al
l
e
a
r
n
i
n
g
f
r
am
e
w
o
r
k
by
L
i
n
e
t
a
l
.
[
3
9
]
,
w
h
i
c
h
u
t
i
li
z
es
u
n
s
u
p
e
r
v
i
s
e
d
le
a
r
n
i
n
g
f
o
r
e
y
e
g
la
s
s
e
s
r
e
m
o
v
a
l
.
W
h
i
l
e
i
n
n
o
v
a
t
iv
e
,
t
h
i
s
m
e
t
h
o
d
f
a
l
ls
s
h
o
r
t
i
n
t
e
r
m
s
o
f
p
r
e
c
is
i
o
n
a
n
d
q
u
a
l
i
t
y
c
o
m
p
a
r
e
d
t
o
s
u
p
e
r
v
is
e
d
l
ea
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
.
I
n
2
0
2
2
,
L
y
u
e
t
a
l
.
[
4
0
]
p
r
o
p
o
s
e
d
a
m
e
t
h
o
d
u
s
i
n
g
3
D
s
y
n
t
h
e
t
i
c
d
a
t
a
t
o
r
e
m
o
v
e
e
y
e
g
l
as
s
es
a
n
d
s
h
a
d
o
ws
,
w
h
i
c
h
p
r
o
v
id
e
d
a
f
o
u
n
d
a
t
i
o
n
f
o
r
f
u
r
t
h
e
r
r
e
s
e
a
r
c
h
.
H
o
w
e
v
e
r
,
t
h
is
a
p
p
r
o
a
c
h
is
li
m
i
t
e
d
b
y
it
s
r
e
l
ian
c
e
o
n
s
y
n
t
h
et
i
c
d
a
t
a
,
w
h
i
c
h
m
a
y
n
o
t
f
u
l
l
y
c
a
p
t
u
r
e
t
h
e
v
a
r
i
a
b
i
l
i
t
y
o
f
r
e
a
l
-
w
o
r
l
d
im
a
g
e
s
.
C
o
c
ci
a
[
4
1
]
d
i
s
c
u
s
s
e
d
i
n
2
0
1
8
t
h
e
c
l
a
s
s
i
f
i
c
at
i
o
n
o
f
i
n
n
o
v
a
t
i
o
n
s
b
a
s
e
d
o
n
t
h
e
i
r
i
n
t
e
r
a
ct
i
o
n
w
it
h
e
x
is
t
i
n
g
t
e
c
h
n
o
l
o
g
i
e
s
.
O
u
r
R
G
D
M
m
o
d
e
l
e
x
e
m
p
l
i
f
i
es
a
n
i
n
c
r
e
m
e
n
t
a
l
i
n
n
o
v
a
t
i
o
n
t
h
at
e
n
h
a
n
c
e
s
t
h
e
i
n
t
e
r
a
c
ti
o
n
b
e
t
we
e
n
e
s
t
a
b
l
is
h
e
d
m
e
t
h
o
d
s
,
s
u
c
h
a
s
G
A
N
s
a
n
d
d
i
f
f
u
s
i
o
n
m
o
d
e
l
s
,
o
f
f
e
r
i
n
g
a
m
o
r
e
p
r
e
c
i
s
e
a
n
d
a
d
a
p
t
a
b
l
e
s
o
l
u
t
i
o
n
f
o
r
g
l
a
s
s
es
r
e
m
o
v
a
l
i
n
f
ac
i
a
l
i
m
a
g
e
s
.
Fro
m
th
e
ab
o
v
e
p
ass
ag
e,
we
ca
n
co
n
clu
d
e
th
at
cu
r
r
en
t
ex
i
s
tin
g
g
lass
es
r
em
o
v
al
m
eth
o
d
s
,
m
ain
ly
th
o
s
e
th
at
ar
e
p
r
o
p
o
s
ed
in
r
ec
en
t
y
ea
r
s
,
r
ely
o
n
co
n
v
en
tio
n
a
l
GAN
-
b
ased
f
r
am
ewo
r
k
s
.
Ou
r
p
r
o
p
o
s
ed
m
o
d
el,
R
GDM
,
d
ep
ar
ts
f
r
o
m
th
e
co
n
v
en
tio
n
al
u
s
e
o
f
a
GAN
[
3
3
]
b
y
u
tili
zin
g
a
DDPM
[
4
2
]
,
[
4
3
]
in
s
tead
.
As
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
:
1
5
0
3
-
1
5
1
6
1506
d
if
f
u
s
io
n
m
o
d
els
h
a
v
e
r
ec
e
n
tly
s
h
o
wn
s
tate
-
of
-
th
e
-
ar
t
r
esu
l
ts
[
4
3
]
-
[
4
5
]
wh
ile
also
o
u
t
p
er
f
o
r
m
in
g
GANs
in
ter
m
s
o
f
ac
c
u
r
ac
y
an
d
v
a
r
iety
[
2
1
]
,
[
4
6
]
,
o
u
r
m
eth
o
d
ca
n
a
u
t
o
m
atica
lly
r
em
o
v
e
g
lass
es
f
r
o
m
f
ac
ial
im
ag
es
in
a
m
o
r
e
p
r
ec
is
e
m
an
n
er
.
3.
B
ACK
G
RO
UND:
D
E
NO
I
S
I
NG
DIFF
USI
O
N
P
RO
B
AB
I
L
I
S
T
I
C
M
O
D
E
L
I
n
th
is
p
ap
er
,
we
u
s
e
d
en
o
is
i
n
g
d
if
f
u
s
io
n
p
r
o
b
ab
ilis
tic
m
o
d
el
as
th
e
b
ase
m
o
d
el
f
o
r
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
.
DDPM
[
4
2
]
,
[
4
3
]
,
w
h
ich
is
s
h
o
wn
i
n
Fig
u
r
e
1
,
is
s
h
o
r
tly
k
n
o
wn
as
d
if
f
u
s
io
n
m
o
d
el.
I
t
is
a
m
o
d
el
in
s
p
ir
ed
b
y
n
o
n
e
q
u
ilib
r
iu
m
th
er
m
o
d
y
n
am
ics
[
4
7
]
,
wh
ich
is
a
class
o
f
laten
t v
ar
iab
le
m
o
d
e
ls
with
th
e
f
o
r
m
:
(
0
)
∶
=
∫
(
0
:
)
1
:
)
(
1
)
I
ts
b
asis
is
a
p
ar
am
eter
ized
Ma
r
k
o
v
ch
ain
tr
ain
ed
with
v
a
r
iatio
n
al
in
f
er
e
n
ce
to
g
e
n
er
ate
s
am
p
les
th
at
m
atch
th
e
d
ata
in
a
f
i
n
ite
am
o
u
n
t o
f
t
im
e.
Fig
u
r
e
1
.
Ov
e
r
v
iew
o
f
th
e
d
e
n
o
is
in
g
d
if
f
u
s
io
n
p
r
o
b
ab
ilis
tic
m
o
d
el
I
t star
ts
b
y
s
am
p
lin
g
f
r
o
m
a
d
i
s
tr
ib
u
tio
n
with
n
o
is
e
,
g
r
ad
u
all
y
p
r
o
d
u
cin
g
less
-
n
o
is
y
s
am
p
le
s
−
1
,
−
2
,
…
,
u
n
til a
f
in
al
s
am
p
le
0
is
r
ea
ch
ed
.
T
h
e
r
ev
er
s
e
p
r
o
ce
s
s
is
ca
lle
d
th
e
jo
in
t d
is
tr
ib
u
tio
n
(
0
:
)
,
wh
ich
is
a
Ma
r
k
o
v
ch
ain
with
lear
n
ed
g
a
u
s
s
ian
tr
an
s
itio
n
s
.
T
h
e
f
o
r
war
d
p
r
o
ce
s
s
,
also
k
n
o
w
n
as
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
,
is
th
e
ap
p
r
o
x
im
ate
p
o
s
ter
io
r
(
1
:
|
0
)
th
at
a
Ma
r
k
o
v
c
h
ain
is
f
ix
ed
to
,
a
d
d
in
g
g
au
s
s
ian
n
o
i
s
e
to
th
e
d
ata
g
r
ad
u
ally
b
ased
o
n
a
v
a
r
ian
ce
s
ch
ed
u
le
1
,
…
,
.
B
y
o
p
tim
izin
g
th
e
s
tan
d
ar
d
v
a
r
iatio
n
al
b
o
u
n
d
o
n
n
eg
ativ
e
lo
g
lik
elih
o
o
d
,
tr
ain
i
n
g
ca
n
t
h
en
b
e
ca
r
r
ied
o
u
t.
Fo
r
a
n
o
is
y
s
am
p
le
,
its
n
o
is
e
co
m
p
o
n
en
t
c
an
b
e
p
r
ed
icted
u
s
in
g
a
f
u
n
cti
o
n
(
,
)
.
Giv
en
a
d
ata
s
am
p
le
0
,
we
ca
n
ad
d
n
o
is
e
to
th
e
im
ag
e
b
y
u
tili
zin
g
th
e
f
o
r
war
d
d
if
f
u
s
io
n
Ma
r
k
o
v
p
r
o
ce
s
s
o
v
er
a
s
ch
ed
u
led
v
ar
ian
ce
o
f
an
d
ti
m
estep
.
T
h
e
y
ar
e
r
an
d
o
m
ly
d
r
awn
to
g
eth
er
to
cr
ea
te
a
n
o
is
ed
s
am
p
le
,
wh
ich
g
en
er
ates e
v
e
r
y
s
am
p
le
in
a
m
in
ib
atch
t
o
tr
ain
th
is
f
u
n
ctio
n
.
T
h
e
f
o
r
war
d
p
r
o
ce
s
s
m
ay
b
e
ap
p
lied
as:
(
1
:
|
0
)
∶
=
∏
(
|
−
1
)
=
1
)
(
2
)
(
|
−
1
)
∶
=
(
√
1
−
,
)
)
(
3
)
wh
er
e
is
th
e
to
tal
n
u
m
b
er
o
f
s
tep
s
.
T
h
e
f
o
r
war
d
p
r
o
ce
s
s
h
as a
n
o
tab
le
p
r
o
p
er
ty
o
f
ad
m
itti
n
g
,
in
clo
s
ed
f
o
r
m
,
o
f
s
am
p
lin
g
at
tim
estep
.
Hen
ce
,
m
ay
b
e
ex
p
r
ess
ed
in
a
clo
s
ed
f
o
r
m
as:
=
√
̅
0
+
√
1
−
̅
)
(
4
)
wh
er
e
∶
=
1
−
,
̅
∶
=
∏
=
1
,
an
d
ϵ
~
(
0
,
Ι
)
.
I
n
o
r
d
er
to
cr
ea
te
im
ag
es
f
r
o
m
r
an
d
o
m
n
o
is
e,
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
ab
o
v
e
n
ee
d
s
to
b
e
in
v
er
ted
.
W
h
en
is
s
m
all
en
o
u
g
h
,
th
is
c
an
b
e
d
o
n
e
b
y
lear
n
i
n
g
t
h
e
g
a
u
s
s
ian
(
−
1
|
)
.
Nev
er
th
eless
,
d
u
e
to
th
e
tr
u
e
d
is
tr
ib
u
tio
n
o
f
0
b
ein
g
u
n
ab
le
t
o
ac
ce
s
s
th
e
g
au
s
s
ian
,
(
−
1
|
)
is
n
o
t
k
n
o
wn
.
T
h
e
r
ef
o
r
e,
a
n
eu
r
al
n
e
two
r
k
is
th
en
tr
ain
ed
to
esti
m
ate
th
e
co
n
d
itio
n
al
d
is
tr
ib
u
tio
n
:
(
−
1
|
)
∶
=
(
(
,
)
,
Σ
(
,
)
)
)
(
5
)
wh
er
e,
tak
en
f
r
o
m
(
4
)
,
−
1
=
1
√
(
−
1
−
√
1
−
̅
)
is
o
b
tain
ed
th
r
o
u
g
h
tr
ain
in
g
to
p
r
ed
ict
it.
T
h
e
o
b
jectiv
e
o
f
tr
ain
i
n
g
,
co
n
s
id
er
in
g
a
s
im
p
le
m
ea
n
-
s
q
u
a
r
e
d
er
r
o
r
lo
s
s
b
etwe
en
th
e
ac
tu
al
n
o
is
e
a
n
d
th
e
an
ticip
ated
n
o
is
e
as
an
ex
am
p
le,
will
b
e
‖
(
,
)
−
‖
2
.
B
y
u
s
in
g
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
emo
ve
g
la
s
s
es d
iffu
s
io
n
mo
d
el
a
n
in
n
o
va
tive
co
n
d
itio
n
ed
o
f e
ye
g
la
s
s
es remo
va
l wi
th
…
(
Yu
liz
a
)
1507
opt
im
ize
r
an
d
o
m
ter
m
s
o
f
,
tr
ain
in
g
is
p
o
s
s
ib
le
to
b
e
co
m
p
leted
in
an
ef
f
icie
n
t
m
an
n
e
r
.
I
n
(
−
1
|
)
,
g
iv
en
a
g
au
s
s
ian
tr
an
s
itio
n
(
−
1
;
(
,
)
,
∑
(
,
)
)
,
th
e
d
en
o
is
in
g
d
is
tr
ib
u
tio
n
ca
n
b
e
m
o
d
elled
with
th
e
m
ea
n
(
,
)
co
m
p
u
ted
as a
p
ar
t o
f
(
,
)
.
T
h
is
lead
s
to
th
e
f
o
llo
win
g
p
ar
am
etr
izatio
n
:
(
,
)
∶
=
1
√
(
−
√
1
−
̅
(
,
)
)
(
6
)
A
s
im
p
lifie
d
tr
ain
in
g
o
b
jectiv
e
is
d
is
co
v
er
ed
b
y
Ho
et
a
l.
[
4
2
]
to
b
e
m
o
r
e
b
en
ef
icial:
ℒ
=
,
0
,
[
‖
−
(
,
)
‖
2
]
)
(
7
)
T
h
is
tr
ain
in
g
o
b
jectiv
e
is
o
b
t
ain
ed
b
y
p
r
ed
ictin
g
th
e
cu
m
u
lativ
e
n
o
is
e
th
at
is
ad
d
ed
to
th
e
in
ter
m
ed
iate
im
ag
e
.
I
n
s
h
o
r
t,
it
r
em
o
v
es
th
e
n
ee
d
f
o
r
th
e
weig
h
tin
g
in
th
e
L
an
g
ev
i
n
d
y
n
am
ics,
tr
an
s
f
o
r
m
in
g
it
in
to
a
weig
h
ted
v
ar
iatio
n
al
b
o
u
n
d
wh
ich
lead
s
to
d
o
wn
-
weig
h
t
in
g
th
e
lo
s
s
ter
m
s
o
f
th
e
s
i
m
p
lifie
d
o
b
jectiv
e
co
r
r
esp
o
n
d
in
g
t
o
s
m
all
.
4.
M
E
T
H
O
D
R
GDM
co
m
b
in
es
th
e
p
r
in
ci
p
les
o
f
DDPM
with
in
n
o
v
at
iv
e
m
o
d
if
icatio
n
s
to
ac
h
iev
e
au
to
m
atic
g
lass
es
r
em
o
v
al.
W
e
f
ir
s
t
p
r
es
en
t
o
u
r
ap
p
r
o
ac
h
f
o
r
co
n
d
itio
n
in
g
th
e
DDPM
o
n
g
lass
es
r
em
o
v
al.
T
h
is
will
b
e
f
o
llo
wed
b
y
an
in
tr
o
d
u
ctio
n
o
n
an
ap
p
r
o
ac
h
to
im
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
th
e
in
-
p
ai
n
ted
e
y
e
s
’
p
o
s
itio
n
a
f
ter
t
h
e
r
em
o
v
al
o
f
g
lass
es.
T
h
e
m
o
d
el
o
p
er
ates
th
r
o
u
g
h
a
s
er
ies
o
f
ca
r
ef
u
lly
o
r
ch
es
tr
ated
s
tep
s
th
at
in
teg
r
ate
co
n
d
itio
n
in
g
m
o
d
u
les,
d
if
f
u
s
io
n
p
r
o
ce
s
s
es,
an
d
lan
d
m
ar
k
-
b
ased
p
o
s
itio
n
in
g
to
ac
h
iev
e
h
ig
h
-
q
u
ality
,
r
ea
lis
tic
r
esu
lt
s
.
T
h
e
p
r
o
ce
s
s
b
eg
in
s
with
an
in
p
u
t
f
ac
ial
im
ag
e
th
at
co
n
tain
s
g
lass
es.
T
h
is
im
ag
e
s
er
v
es
as
th
e
s
t
ar
tin
g
p
o
in
t
f
o
r
th
e
g
lass
es
r
em
o
v
al
p
r
o
ce
d
u
r
e.
A
m
o
d
if
icatio
n
o
f
th
e
s
eg
m
en
tatio
n
m
o
d
el,
th
e
SAM
[
1
9
]
,
is
a
p
p
lied
to
th
e
in
p
u
t
im
ag
e.
T
h
is
s
eg
m
en
tatio
n
m
o
d
el
is
s
p
ec
if
ically
tr
ain
ed
t
o
d
etec
t
an
d
is
o
late
th
e
g
lass
es
f
r
o
m
th
e
r
est
o
f
t
h
e
f
ac
ial
f
ea
tu
r
es,
r
esu
ltin
g
in
a
b
in
ar
y
m
ask
th
at
h
ig
h
lig
h
ts
th
e
g
lass
es r
eg
io
n
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
f
lo
w
o
f
m
o
d
el
ex
p
lain
ed
.
T
h
e
b
in
ar
y
m
ask
g
en
er
ated
in
th
e
p
r
ev
i
o
u
s
s
tep
is
cr
u
cial
f
o
r
t
h
e
co
n
d
itio
n
i
n
g
p
r
o
ce
s
s
.
I
t
id
en
tifie
s
t
h
e
e
x
ac
t
r
e
g
io
n
s
o
f
th
e
im
ag
e
wh
er
e
th
e
g
lass
es
ar
e
lo
ca
ted
,
en
s
u
r
in
g
th
at
t
h
ese
ar
ea
s
r
ec
e
iv
e
f
o
cu
s
ed
atten
tio
n
d
u
r
in
g
t
h
e
s
u
b
s
eq
u
en
t
d
en
o
is
in
g
p
r
o
c
ess
.
I
n
p
ar
allel,
th
e
m
o
d
el
ex
tr
ac
ts
an
d
in
c
o
r
p
o
r
at
es
lan
d
m
ar
k
d
ata
f
r
o
m
th
e
in
p
u
t
im
ag
e.
T
h
is
d
ata
in
clu
d
es
p
r
ec
is
e
co
o
r
d
in
ates
f
o
r
t
h
e
ey
es
an
d
o
th
e
r
k
ey
f
a
cial
f
ea
tu
r
es,
wh
ic
h
ar
e
ess
en
tial
f
o
r
m
ain
tain
in
g
th
e
co
r
r
ec
t
p
o
s
itio
n
in
g
o
f
th
e
ey
es d
u
r
in
g
th
e
in
p
ain
tin
g
p
r
o
ce
s
s
.
Fig
u
r
e
2
.
T
h
e
f
lo
w
of
th
e
R
G
DM
m
o
d
el
T
o
in
itiate
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
,
g
au
s
s
ian
n
o
is
e
is
s
tr
ateg
ically
in
jecte
d
in
t
o
th
e
g
lass
es
r
eg
io
n
o
f
t
h
e
im
ag
e,
g
u
id
ed
b
y
th
e
b
in
ar
y
m
ask
.
T
h
e
n
o
is
e
lev
els
ar
e
co
n
tr
o
lled
b
y
a
v
ar
ian
ce
s
ch
ed
u
le,
en
s
u
r
in
g
th
at
th
e
n
o
is
e
is
g
r
ad
u
ally
ad
d
e
d
to
th
e
id
en
tifie
d
r
eg
io
n
wh
ile
p
r
es
er
v
in
g
th
e
s
tr
u
ctu
r
e
o
f
th
e
s
u
r
r
o
u
n
d
in
g
u
n
m
ask
ed
ar
ea
s
.
T
h
e
m
o
d
el
th
en
p
e
r
f
o
r
m
s
th
e
f
o
r
war
d
d
if
f
u
s
io
n
p
r
o
c
ess
,
wh
er
e
th
e
n
o
is
y
im
ag
e
u
n
d
er
g
o
es
a
s
er
ies
o
f
tr
an
s
f
o
r
m
atio
n
s
.
E
ac
h
s
tep
o
f
th
e
Ma
r
k
o
v
c
h
ain
i
n
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
in
cr
em
e
n
tally
ad
d
s
m
o
r
e
n
o
is
e
to
th
e
id
en
tifie
d
r
eg
i
o
n
,
lead
i
n
g
to
a
p
r
o
g
r
ess
iv
ely
n
o
is
ier
v
er
s
io
n
o
f
th
e
o
r
ig
in
al
im
ag
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
:
1
5
0
3
-
1
5
1
6
1508
On
ce
th
e
f
o
r
war
d
d
if
f
u
s
io
n
p
r
o
ce
s
s
r
ea
ch
es its
p
ea
k
n
o
is
e
lev
el,
th
e
r
ev
er
s
e
d
i
f
f
u
s
io
n
p
r
o
c
ess
b
eg
in
s
.
T
h
e
m
o
d
el,
co
n
d
itio
n
ed
b
y
b
o
th
th
e
b
in
ar
y
m
ask
an
d
th
e
lan
d
m
ar
k
d
ata,
s
y
s
tem
atica
lly
r
em
o
v
es
th
e
n
o
is
e
f
r
o
m
th
e
im
a
g
e,
ef
f
ec
tiv
ely
r
e
co
n
s
tr
u
ctin
g
th
e
p
ix
els
in
th
e
g
lass
es
r
eg
io
n
.
Du
r
in
g
t
h
is
r
ev
er
s
e
p
r
o
ce
s
s
,
th
e
m
o
d
el
f
o
cu
s
es
o
n
in
p
ain
tin
g
th
e
g
lass
es
r
eg
io
n
.
T
h
e
b
in
ar
y
m
ask
e
n
s
u
r
es
th
at
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
is
co
n
ce
n
tr
ated
o
n
th
e
ar
ea
s
t
h
at
o
r
ig
i
n
ally
c
o
n
tain
ed
th
e
g
lass
es,
wh
ile
th
e
lan
d
m
ar
k
d
ata
g
u
id
es
th
e
ac
cu
r
at
e
p
lace
m
en
t o
f
t
h
e
ey
es a
n
d
o
th
er
f
ac
ial
f
ea
tu
r
es.
T
h
e
o
u
t
p
u
t
f
r
o
m
ea
ch
s
tep
o
f
th
e
r
ev
er
s
e
d
if
f
u
s
io
n
is
g
r
ad
u
a
lly
r
ef
in
ed
,
with
th
e
m
o
d
el
f
illi
n
g
in
t
h
e
g
lass
es
r
eg
io
n
with
p
lau
s
ib
le
f
ac
ial
d
etails.
T
h
e
r
ev
er
s
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
co
n
tin
u
es
u
n
til
th
e
m
o
d
el
g
en
e
r
ates
a
f
in
al,
n
o
is
e
-
f
r
ee
im
ag
e
.
T
h
i
s
o
u
tp
u
t
im
ag
e
h
as
th
e
g
lass
es
co
m
p
letely
r
em
o
v
ed
,
with
th
e
ey
es
an
d
o
th
e
r
f
ac
ial
f
ea
tu
r
es a
cc
u
r
ately
r
ec
o
n
s
tr
u
cted
an
d
alig
n
ed
with
t
h
e
ir
o
r
ig
in
al
p
o
s
itio
n
s
.
4
.
1
.
Co
nd
it
io
nin
g
m
o
du
le
T
h
e
co
n
d
itio
n
i
n
g
m
o
d
u
le
in
R
GDM
is
a
cr
u
cial
n
o
v
el
ad
d
itio
n
,
f
o
r
in
c
o
r
p
o
r
atin
g
in
f
o
r
m
at
io
n
ab
o
u
t
g
lass
es
co
n
d
itio
n
s
in
to
t
h
e
d
e
n
o
is
in
g
p
r
o
ce
s
s
.
I
t
aim
s
t
o
in
t
eg
r
ate
in
f
o
r
m
atio
n
ab
o
u
t
th
e
p
r
esen
ce
o
r
ab
s
en
ce
o
f
g
lass
es
in
to
th
e
d
en
o
is
in
g
p
r
o
ce
s
s
,
allo
win
g
th
e
m
o
d
el
to
ad
a
p
t
its
d
en
o
is
in
g
s
tr
ateg
y
b
ased
o
n
th
is
co
n
d
itio
n
.
T
h
e
m
o
d
el’
s
g
lass
es
r
em
o
v
al
m
eth
o
d
r
elies
o
n
p
r
ed
ictin
g
th
e
m
is
s
in
g
p
ix
el
v
alu
es
o
f
th
e
f
ac
ial
im
ag
e.
T
h
is
is
d
o
n
e
b
y
th
e
h
e
lp
o
f
th
e
c
o
n
d
itio
n
in
g
m
o
d
u
le
b
y
m
ask
in
g
th
e
g
lass
es
r
eg
io
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
o
v
er
v
iew
o
f
th
e
R
GDM
ap
p
r
o
ac
h
.
Fig
u
r
e
3
.
R
GDM
:
g
lass
es r
em
o
v
al
p
r
o
ce
s
s
o
v
er
v
iew
I
n
th
e
co
n
tex
t o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
co
n
s
id
er
in
g
a
n
in
p
u
t im
ag
e
d
en
o
ted
as '
'
,
a
g
las
s
es
m
ask
as '
'
r
ep
r
esen
tin
g
th
e
ar
ea
s
o
f
'
'
to
b
e
in
-
p
ain
ted
,
an
d
ey
es'
lan
d
m
ar
k
attr
ib
u
te
r
ep
r
esen
ted
b
y
'
'
,
th
e
p
r
im
ar
y
o
b
jectiv
e
is
to
ef
f
ec
tiv
el
y
in
-
p
ain
t
th
e
g
lass
es
m
ask
an
d
g
en
er
ate
a
n
o
u
tp
u
t
im
a
g
e
d
en
o
ted
as
'
̂
'
.
T
h
e
g
en
er
ated
r
eg
io
n
s
ar
e
r
ep
r
ese
n
ted
as
'
⨀
̂
'
,
wh
ile
th
e
k
n
o
wn
f
a
cial
r
eg
io
n
s
a
r
e
ex
p
r
ess
ed
as
'
(
1
−
)
⨀
'
.
Giv
en
th
at
n
o
is
e
is
ex
cl
u
s
iv
ely
in
tr
o
d
u
ce
d
to
th
e
m
ask
ed
g
l
ass
es
r
eg
io
n
,
wh
er
e
'
=
0
'
,
th
e
f
o
r
war
d
p
r
o
ce
s
s
at
tim
estep
'
'
ca
n
b
e
m
ath
em
atica
lly
ex
p
r
ess
ed
as
(
8
)
.
̂
=
√
1
−
̅
+
√
̅
0
)
=
(
1
−
)
⨀
0
+
⨀
̂
(
8
)
T
h
e
in
p
u
t
v
a
r
iab
les,
,
,
an
d
,
ar
e
f
ed
in
to
th
e
c
o
n
d
itio
n
in
g
m
o
d
u
le,
wh
er
e
th
e
m
o
d
u
le
d
i
s
ce
r
n
s
th
e
p
r
o
ce
s
s
to
r
esto
r
e
th
e
m
ask
ed
g
lass
es
r
eg
io
n
,
0
⨀
,
wh
ile
u
tili
zin
g
th
e
clea
r
an
d
u
n
m
ask
ed
f
ac
ial
in
f
o
r
m
atio
n
.
T
h
is
en
s
u
r
es
th
at
th
e
in
-
p
ain
ted
e
y
es
with
in
th
e
m
ask
ed
g
lass
es,
,
h
ar
m
o
n
iz
e
with
th
e
o
v
er
all
f
ac
ial
im
ag
e.
In
(
9
)
en
ca
p
s
u
lat
es th
e
tr
ain
in
g
o
b
jectiv
e
d
er
i
v
ed
f
r
o
m
th
is
p
r
o
ce
s
s
.
ℒ
−
=
~
(
0
,
)
[
‖
−
(
,
,
,
)
‖
2
]
)
(
9
)
R
an
d
o
m
g
au
s
s
ian
n
o
is
e
is
g
en
er
ated
ac
r
o
s
s
th
e
m
ask
ed
ar
ea
=
(
1
−
)
⨀
0
+
⨀
0
.
T
h
er
ef
o
r
e,
to
o
b
tain
th
e
g
en
er
ated
o
u
tp
u
t
0
,
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
is
r
ev
er
s
ed
d
u
r
in
g
th
e
i
n
f
er
en
c
e
s
tag
e.
W
ith
th
e
r
ev
er
s
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
s
u
cc
ess
f
u
lly
co
n
d
itio
n
ed
to
f
o
cu
s
o
n
th
e
g
lass
es
ar
ea
,
we
s
u
cc
ess
f
u
lly
in
-
p
ain
t
th
e
g
lass
es
ar
ea
with
th
e
n
ec
ess
ar
y
f
ac
ial
f
ea
tu
r
es
an
d
ey
es,
wh
ich
in
r
etu
r
n
r
em
o
v
es
th
e
g
lass
es
f
r
o
m
th
e
f
ac
ial
im
ag
es.
T
h
is
ca
n
b
e
s
ee
n
in
F
ig
u
r
e
3
,
wh
er
e
,
wh
ich
is
th
e
s
o
u
r
ce
im
ag
e
0
co
m
b
in
ed
with
b
in
ar
y
m
ask
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
emo
ve
g
la
s
s
es d
iffu
s
io
n
mo
d
el
a
n
in
n
o
va
tive
co
n
d
itio
n
ed
o
f e
ye
g
la
s
s
es remo
va
l wi
th
…
(
Yu
liz
a
)
1509
u
n
d
er
g
o
es
th
e
d
if
f
u
s
io
n
m
o
d
el
to
p
r
o
d
u
ce
−
1
an
d
s
o
o
n
,
wh
ich
will
b
e
iter
ated
u
n
til
o
u
tp
u
t
im
ag
e
̂
is
f
in
ally
p
r
o
d
u
ce
d
wh
er
e
th
e
ey
eg
lass
is
s
u
cc
es
s
f
u
lly
r
em
o
v
ed
.
4
.
2
.
E
y
es
prec
is
io
n a
cc
ura
cy
T
o
e
n
s
u
r
e
t
h
e
g
e
n
e
r
a
t
i
o
n
o
f
a
c
c
u
r
a
t
e
a
n
d
f
a
i
r
g
l
a
s
s
es
m
a
s
k
s
,
o
u
r
m
e
t
h
o
d
o
l
o
g
y
i
n
c
o
r
p
o
r
a
t
e
s
a
m
e
t
i
c
u
l
o
u
s
l
y
c
r
a
f
t
e
d
a
p
p
r
o
a
c
h
l
e
v
e
r
a
g
i
n
g
a
m
o
d
i
f
i
e
d
S
A
M
[
1
9
]
.
T
h
i
s
a
d
a
p
t
a
t
i
o
n
o
f
S
A
M
is
p
u
r
p
o
s
e
f
u
l
l
y
t
r
a
i
n
e
d
t
o
a
d
e
p
t
l
y
d
e
t
e
c
t
a
n
d
s
e
g
m
e
n
t
e
y
e
g
l
as
s
es
w
i
t
h
i
n
f
a
c
i
al
i
m
a
g
e
s
,
t
h
u
s
e
n
h
a
n
ci
n
g
t
h
e
p
r
ec
i
s
io
n
o
f
t
h
e
g
e
n
e
r
a
te
d
g
l
a
s
s
e
s
'
m
as
k
s
.
T
h
e
p
r
o
c
e
s
s
i
n
i
t
i
a
t
es
w
it
h
i
n
p
u
t
t
i
n
g
t
h
e
s
o
u
r
c
e
i
m
a
g
e
s
i
n
t
o
t
h
e
m
o
d
i
f
i
e
d
S
A
M
,
s
t
r
a
t
e
g
ic
a
l
ly
t
a
i
l
o
r
e
d
t
o
f
o
c
u
s
s
o
l
e
l
y
o
n
t
h
e
s
e
g
m
e
n
t
at
i
o
n
o
f
e
y
e
g
l
as
s
es
.
T
h
e
o
u
t
c
o
m
e
o
f
t
h
i
s
s
e
g
m
e
n
tat
i
o
n
s
te
p
y
i
e
l
d
s
t
h
e
g
l
a
s
s
e
s
'
m
as
k
s
d
e
n
o
t
e
d
as
‘
’
a
s
v
i
s
u
a
l
l
y
r
e
p
r
e
s
e
n
t
e
d
i
n
F
i
g
u
r
e
3
.
T
h
e
s
e
m
a
s
k
s
b
e
c
o
m
e
a
p
i
v
o
t
a
l
c
o
m
p
o
n
e
n
t
i
n
t
h
e
s
u
b
s
e
q
u
e
n
t
s
t
a
g
es
o
f
o
u
r
m
e
t
h
o
d
o
l
o
g
y
,
s
e
a
m
l
es
s
l
y
i
n
t
e
g
r
a
t
i
n
g
i
n
t
o
t
h
e
p
r
o
c
es
s
es
o
u
t
lin
e
d
i
n
S
e
c
t
i
o
n
A
f
o
r
f
u
r
t
h
e
r
a
n
a
l
y
s
is
a
n
d
g
l
a
s
s
e
s
r
em
o
v
a
l
.
T
h
i
s
n
u
a
n
c
e
d
a
p
p
r
o
a
c
h
e
n
s
u
r
e
s
t
h
e
r
e
li
a
b
i
l
it
y
a
n
d
a
c
c
u
r
a
c
y
o
f
t
h
e
m
a
s
k
s
,
l
a
y
i
n
g
a
r
o
b
u
s
t
f
o
u
n
d
a
t
i
o
n
f
o
r
s
u
b
s
e
q
u
e
n
t
p
h
a
s
es
o
f
t
h
e
r
e
m
o
v
e
g
l
a
s
s
e
s
d
i
f
f
u
s
i
o
n
m
o
d
e
l
.
Desp
ite
its
p
r
o
f
icien
cy
in
g
e
n
er
atin
g
an
ac
c
u
r
ate
m
ask
f
o
r
d
if
f
er
en
t
ty
p
es
o
f
g
lass
es,
we
f
o
u
n
d
th
at
th
e
s
eg
m
en
tatio
n
m
o
d
el
te
n
d
s
to
m
ask
th
e
en
tire
g
lass
es,
in
clu
d
in
g
th
eir
len
s
es,
ir
r
esp
ec
ti
v
e
o
f
wh
et
h
er
t
h
ey
r
esem
b
le
s
u
n
g
lass
es
with
co
l
o
r
ed
len
s
es
o
r
co
n
v
en
tio
n
al
e
y
eg
lass
es
with
clea
r
len
s
e
s
.
T
h
is
is
ev
id
en
t
in
th
e
b
in
ar
y
m
ask
'
'
an
d
th
e
m
ask
ed
im
ag
e
'
'
p
o
r
tr
ay
ed
in
Fig
u
r
e
3
.
T
h
is
co
m
p
r
e
h
en
s
iv
e
m
ask
in
g
lead
s
to
th
e
p
o
s
s
ib
ilit
y
o
f
th
e
g
e
n
er
ated
ey
es
b
ein
g
p
o
s
itio
n
ed
d
if
f
er
en
tly
th
an
in
t
h
e
o
r
ig
in
al
i
m
ag
e,
in
p
ar
ticu
la
r
co
n
ce
r
n
in
g
t
h
e
ty
p
es
o
f
g
lass
es
th
at
ar
e
d
esig
n
ed
f
o
r
v
is
io
n
co
r
r
ec
ti
o
n
(
th
o
s
e
with
clea
r
len
s
es).
T
o
ad
d
r
ess
th
is
p
o
ten
tial
m
is
alig
n
m
en
t
an
d
en
s
u
r
e
th
e
ac
c
u
r
ate
p
o
s
iti
o
n
in
g
o
f
th
e
g
e
n
er
ated
ey
es,
th
e
m
o
d
el
is
also
eq
u
ip
p
e
d
with
in
f
o
r
m
atio
n
c
o
n
ce
r
n
in
g
th
e
ey
es'
lan
d
m
ar
k
s
d
en
o
ted
as
'
'
,
wh
ich
is
le
v
er
ag
ed
f
r
o
m
th
e
d
ataset'
s
in
tr
in
s
ic
lan
d
m
ar
k
att
r
ib
u
tes:
=
(
,
|
ℎ
,
ℎ
)
)
(
1
0
)
T
h
r
o
u
g
h
th
e
in
co
r
p
o
r
atio
n
o
f
ey
es'
lan
d
m
ar
k
s
d
en
o
ted
as
'
'
,
th
e
m
o
d
el
en
s
u
r
es
th
e
p
r
ec
is
e
p
o
s
itio
n
in
g
o
f
th
e
g
en
e
r
ated
e
y
es,
m
ain
tain
in
g
th
em
i
n
th
e
e
x
ac
t
s
am
e
p
o
s
itio
n
as
in
th
e
o
r
ig
in
al
im
ag
e.
T
h
is
s
tr
ateg
ic
u
s
e
o
f
lan
d
m
ar
k
in
f
o
r
m
atio
n
ac
ts
as
a
s
af
eg
u
ar
d
ag
ain
s
t
p
o
ten
tial
d
is
to
r
tio
n
s
th
at
co
u
ld
ar
is
e
d
u
r
in
g
t
h
e
g
lass
es
r
em
o
v
al
p
r
o
ce
s
s
.
Fig
u
r
e
4
p
r
o
v
id
es
a
v
is
u
al
r
ep
r
esen
tatio
n
o
f
th
e
im
p
ac
t
o
f
in
cl
u
d
in
g
ey
es'
lan
d
m
ar
k
s
,
s
h
o
wca
s
in
g
s
am
p
les
o
f
g
en
er
ated
ey
es
b
o
t
h
with
an
d
with
o
u
t
th
e
u
tili
za
tio
n
o
f
lan
d
m
a
r
k
in
f
o
r
m
atio
n
.
T
h
is
n
o
t
o
n
ly
e
x
em
p
lifie
s
th
e
ef
f
ec
tiv
e
n
es
s
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
in
p
r
es
er
v
in
g
t
h
e
o
r
ig
in
al
ey
e
p
o
s
itio
n
b
u
t
also
s
er
v
es
as
a
tes
tam
en
t
to
th
e
m
o
d
el'
s
a
b
ilit
y
to
p
r
o
d
u
ce
r
ea
lis
tic
an
d
ac
cu
r
ate
r
esu
lts
in
th
e
ab
s
en
ce
o
f
g
lass
es.
T
h
e
p
s
eu
d
o
co
d
e
f
o
r
th
e
g
lass
es r
em
o
v
al
p
r
o
ce
s
s
ca
n
b
e
s
ee
n
in
Alg
o
r
ith
m
1
.
Alg
o
r
ith
m
1.
Glass
es r
em
o
v
al
1:
~
(
,
)
2:
for
=
,
…
,
1
do
3:
for
=
1
,
…
,
do
4:
~
(
,
)
if
>
1
, else
=
0
5:
̂
=
√
1
−
̅
+
√
̅
0
6:
~
(
,
)
if
>
1
, else
=
0
7:
=
(
,
|
ℎ
,
ℎ
)
8:
=
(
1
−
)
⨀
0
+
⨀
̂
9:
end for
10:
end for
11:
return
0
4
.
3
.
E
x
perim
ent
a
l
s
et
up
4
.
3
.
1
.
Da
t
a
s
et
W
e
co
n
d
u
cted
o
u
r
ex
p
e
r
im
e
n
ts
u
s
in
g
th
e
C
eleb
A
d
ataset
[
4
8
]
,
wh
ich
is
a
f
ac
e
attr
ib
u
te
d
ataset
d
er
iv
ed
f
r
o
m
lab
ellin
g
im
a
g
e
s
s
elec
ted
f
r
o
m
th
e
ch
allen
g
i
n
g
f
ac
e
d
atasets
,
C
eleb
Face
s
.
I
t
f
ea
tu
r
es
1
0
,
1
7
7
id
en
titi
es,
ea
ch
with
a
r
o
u
n
d
2
0
im
ag
es,
f
o
r
a
t
o
tal
o
f
2
0
2
,
5
9
9
ce
leb
r
ity
im
a
g
es.
I
t
h
as
ea
ch
im
ag
e
an
n
o
tated
with
f
o
r
ty
f
ac
ial
f
ea
tu
r
es
an
d
f
iv
e
k
ey
ch
ar
ac
ter
is
tics
,
in
clu
d
in
g
th
e
p
r
esen
ce
o
r
ab
s
en
c
e
o
f
g
lass
es
an
d
th
e
lan
d
m
ar
k
s
o
f
b
o
th
th
e
lef
t
an
d
r
ig
h
t
ey
es.
Fro
m
th
e
an
n
o
ta
tio
n
s
,
we
s
ep
ar
ated
th
e
C
eleb
A
d
ataset
in
to
two
s
u
b
s
ets,
o
n
e
wh
er
e
th
e
im
ag
e
s
co
n
s
is
t
o
f
g
lass
es
an
d
th
e
o
th
er
wh
er
e
th
e
im
ag
es
h
a
v
e
n
o
g
lass
es.
I
n
to
tal,
th
er
e
ar
e
1
8
9
,
4
0
6
n
o
n
-
g
lass
es
im
ag
es
an
d
1
3
,
1
9
3
im
ag
es
wea
r
in
g
g
lass
es.
All
o
f
th
e
f
ac
ial
im
ag
es
ar
e
cr
o
p
p
e
d
to
1
7
8
×
1
7
8
,
b
ef
o
r
e
b
ein
g
r
esized
to
2
5
6
×
2
5
6
ea
c
h
.
4
.
3
.
2
.
I
m
plem
ent
a
t
i
o
n
det
a
ils
T
h
e
a
r
c
h
i
t
e
ct
u
r
e
o
f
t
h
e
r
e
m
o
v
e
g
l
a
s
s
e
s
d
i
f
f
u
s
i
o
n
m
o
d
e
l
is
r
o
o
t
e
d
i
n
t
h
e
D
D
PM
f
r
a
m
ew
o
r
k
.
I
t
is
c
o
m
p
o
s
e
d
o
f
a
g
e
n
e
r
at
o
r
r
e
s
p
o
n
s
i
b
l
e
f
o
r
t
h
e
d
e
n
o
is
i
n
g
p
r
o
c
ess
a
n
d
a
c
o
n
d
i
t
i
o
n
i
n
g
m
o
d
u
l
e
t
h
a
t
e
m
p
l
o
y
s
a
b
i
n
a
r
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
:
1
5
0
3
-
1
5
1
6
1510
m
a
s
k
t
o
i
d
e
n
t
i
f
y
t
h
e
g
l
a
s
s
es
in
t
h
e
i
n
p
u
t
i
m
a
g
e
s
.
W
e
h
a
v
e
i
m
p
l
e
m
e
n
t
e
d
t
h
e
R
GD
M
m
e
th
o
d
u
s
i
n
g
P
y
T
o
r
c
h
2
.
0
.
0
,
P
y
t
h
o
n
3
.
8
(
U
b
u
n
t
u
2
0
.
0
4
)
,
a
n
d
C
u
d
a
1
1
.
8
,
r
u
n
n
i
n
g
o
n
a
n
R
T
X
3
0
9
0
(
2
4
G
B
)
G
P
U
.
T
h
e
m
o
d
e
l
u
n
d
e
r
g
o
e
s
t
r
a
i
n
i
n
g
o
n
t
h
e
d
a
t
as
e
t
f
o
r
3
0
0
,
0
0
0
i
t
e
r
a
t
i
o
n
s
,
w
i
t
h
T
s
e
t
t
o
2
5
0
.
R
GDM
is
b
ased
o
n
th
e
DDPM,
wh
ich
is
a
ty
p
e
o
f
p
r
o
b
a
b
ilis
tic
g
en
er
ativ
e
m
o
d
el
th
at
r
ev
er
s
es
a
d
if
f
u
s
io
n
p
r
o
ce
s
s
to
g
en
er
at
e
h
ig
h
-
q
u
ality
im
ag
es.
T
h
e
n
etwo
r
k
is
co
n
d
itio
n
ed
o
n
a
b
in
ar
y
m
ask
an
d
lan
d
m
ar
k
d
ata,
wh
ic
h
g
u
i
d
e
th
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
to
id
en
tify
th
e
g
lass
es
r
eg
io
n
in
th
e
in
p
u
t
im
ag
es.
W
e
h
av
e
im
p
lem
en
ted
th
e
R
GDM
m
eth
o
d
u
s
in
g
Py
T
o
r
c
h
2
.
0
.
0
,
Py
th
o
n
3
.
8
(
U
b
u
n
tu
2
0
.
0
4
)
,
an
d
C
u
d
a
1
1
.
8
,
r
u
n
n
in
g
o
n
an
R
T
X
3
0
9
0
(
2
4
GB
)
GPU.
T
h
e
C
e
l
e
b
A
d
a
ta
s
et
w
as
u
t
i
l
ize
d
f
o
r
t
r
a
i
n
i
n
g
a
n
d
t
e
s
t
i
n
g
t
h
e
R
G
DM
m
o
d
e
l
.
T
h
i
s
d
a
t
as
e
t
c
o
n
t
a
i
n
s
o
v
e
r
2
0
0
,
0
0
0
c
e
l
e
b
r
i
t
y
i
m
a
g
e
s
w
i
t
h
a
v
a
r
i
e
t
y
o
f
a
t
t
r
i
b
u
t
e
s
,
i
n
c
l
u
d
i
n
g
w
h
e
t
h
e
r
t
h
e
p
e
r
s
o
n
is
w
ea
r
i
n
g
g
l
a
s
s
e
s
.
S
i
n
c
e
t
h
e
C
e
l
e
b
A
d
a
t
as
e
t
d
o
e
s
n
o
t
c
o
m
e
w
it
h
d
e
t
ai
l
e
d
s
e
g
m
e
n
t
at
i
o
n
m
a
s
k
s
,
a
m
o
d
i
f
i
e
d
v
e
r
s
i
o
n
o
f
t
h
e
S
A
M
[
1
9
]
w
as
e
m
p
l
o
y
e
d
t
o
g
e
n
e
r
a
t
e
b
i
n
a
r
y
m
a
s
k
s
.
T
h
es
e
m
as
k
s
i
s
o
l
at
e
t
h
e
g
l
a
s
s
e
s
r
e
g
i
o
n
,
h
e
l
p
i
n
g
t
h
e
m
o
d
e
l
f
o
c
u
s
o
n
t
h
e
c
o
r
r
e
c
t
a
r
e
a
d
u
r
i
n
g
t
h
e
d
i
f
f
u
s
io
n
p
r
o
c
e
s
s
.
7
0
%
o
f
t
h
e
d
a
t
as
e
t
(
a
r
o
u
n
d
1
4
0
,
0
0
0
i
m
a
g
e
s
)
i
s
a
ll
o
c
a
t
e
d
f
o
r
t
r
a
i
n
i
n
g
t
h
e
R
G
D
M
m
o
d
e
l
.
T
h
e
t
r
a
i
n
in
g
s
e
t
i
n
c
l
u
d
e
s
i
m
a
g
es
w
i
t
h
an
d
w
i
t
h
o
u
t
g
l
ass
e
s
,
e
n
s
u
r
i
n
g
th
e
m
o
d
e
l
l
e
a
r
n
s
t
o
a
c
c
u
r
a
t
e
l
y
r
e
m
o
v
e
g
l
as
s
es
w
h
il
e
p
r
e
s
e
r
v
i
n
g
t
h
e
u
n
d
e
r
l
y
i
n
g
f
a
c
i
a
l
f
e
at
u
r
e
s
.
T
h
e
r
e
m
ai
n
i
n
g
3
0
%
o
f
t
h
e
d
a
t
a
s
et
(
a
r
o
u
n
d
6
0
,
0
0
0
i
m
a
g
e
s
)
i
s
r
e
s
er
v
e
d
f
o
r
t
e
s
t
i
n
g
.
T
h
is
s
e
t
i
s
u
s
ed
t
o
e
v
a
l
u
a
te
t
h
e
m
o
d
e
l
’
s
p
e
r
f
o
r
m
a
n
c
e
,
p
a
r
t
i
c
u
la
r
l
y
i
t
s
a
b
i
l
it
y
t
o
r
e
c
o
n
s
t
r
u
ct
t
h
e
f
a
ce
w
i
t
h
o
u
t
g
l
as
s
e
s
w
h
i
l
e
m
ai
n
t
a
in
i
n
g
n
a
t
u
r
a
l
i
s
m
a
n
d
r
e
al
is
m
.
T
h
e
lear
n
i
n
g
r
ate
was
s
et
t
o
0
.
0
0
0
1
,
wh
ich
is
ty
p
ical
f
o
r
tr
ai
n
in
g
d
if
f
u
s
io
n
m
o
d
els.
T
h
is
r
a
te
en
s
u
r
es
th
at
th
e
m
o
d
el
lear
n
s
at
a
p
a
ce
th
at
b
alan
ce
s
co
n
v
er
g
en
ce
s
p
ee
d
an
d
s
tab
ilit
y
.
T
h
e
m
o
d
el
was
tr
ain
ed
f
o
r
3
0
0
,
0
0
0
iter
atio
n
s
,
wh
ich
is
a
co
m
m
o
n
ly
u
s
ed
m
ea
s
u
r
e
i
n
d
if
f
u
s
io
n
m
o
d
els
d
u
e
to
th
e
h
ig
h
n
u
m
b
er
o
f
tim
estep
s
in
v
o
lv
ed
,
with
T
s
et
to
2
5
0
.
T
h
is
ex
ten
s
iv
e
tr
ain
in
g
allo
wed
th
e
m
o
d
el
to
lear
n
th
e
co
m
p
lex
task
o
f
g
lass
es
r
em
o
v
al
an
d
e
n
s
u
r
ed
th
at
it
g
en
e
r
alize
d
well
t
o
n
ew
im
ag
es.
T
h
e
Ad
a
m
o
p
ti
m
izer
was
u
s
ed
f
o
r
tr
ain
in
g
,
with
b
eta
1
s
et
to
0
.
9
an
d
b
eta2
s
et
t
o
0
.
9
9
9
.
A
d
am
was
ch
o
s
en
f
o
r
its
r
o
b
u
s
t
n
ess
an
d
a
b
ilit
y
to
h
an
d
le
th
e
s
p
a
r
s
e
g
r
ad
ien
ts
th
a
t a
r
e
o
f
ten
e
n
co
u
n
ter
ed
in
d
ee
p
g
en
er
ativ
e
m
o
d
els.
T
h
e
in
p
u
t
t
o
th
e
R
GDM
m
o
d
el
is
a
f
ac
ial
im
ag
e
with
g
lass
es,
alo
n
g
with
a
co
r
r
esp
o
n
d
i
n
g
b
in
a
r
y
m
ask
th
at
is
o
lates
th
e
g
lass
es
r
eg
io
n
.
T
h
e
im
ag
e
is
p
r
e
p
r
o
ce
s
s
ed
to
m
atch
th
e
in
p
u
t
r
eq
u
ir
em
en
ts
o
f
th
e
m
o
d
el,
ty
p
ically
r
esized
to
2
5
6
×
2
5
6
p
ix
els
an
d
n
o
r
m
alize
d
.
T
h
e
f
in
al
im
ag
e,
with
th
e
g
lass
es
r
em
o
v
ed
,
i
s
g
en
er
ated
t
h
r
o
u
g
h
th
e
r
e
v
er
s
e
d
if
f
u
s
io
n
p
r
o
ce
s
s
.
Star
tin
g
f
r
o
m
th
e
n
o
is
y
im
a
g
e
cr
ea
ted
d
u
r
in
g
th
e
f
o
r
war
d
p
r
o
ce
s
s
,
th
e
m
o
d
el
g
r
a
d
u
ally
r
ed
u
ce
s
th
e
n
o
is
e,
g
u
id
e
d
b
y
th
e
co
n
d
itio
n
in
g
m
o
d
u
le,
u
n
til
a
h
ig
h
-
q
u
ality
,
g
lass
es
-
r
em
o
v
ed
im
a
g
e
is
p
r
o
d
u
ce
d
.
T
h
e
o
u
t
p
u
t
is
a
2
5
6
×
2
5
6
p
ix
el
im
ag
e
with
th
e
g
la
s
s
es
r
em
o
v
ed
.
T
h
e
m
o
d
el’
s
g
o
al
is
to
g
en
er
ate
a
r
ea
lis
tic
im
ag
e
wh
er
e
th
e
g
lass
es h
av
e
b
ee
n
in
p
ain
ted
with
t
h
e
u
n
d
er
l
y
in
g
f
ac
ial
f
ea
tu
r
es,
p
ar
tic
u
lar
ly
f
o
cu
s
in
g
o
n
m
ain
tain
in
g
th
e
co
r
r
e
ct
p
l
ac
em
en
t
o
f
th
e
ey
es
an
d
e
n
s
u
r
in
g
th
at
th
e
o
u
tp
u
t
is
f
r
ee
o
f
ar
tifa
cts.
4
.
3
.
3
.
E
v
a
lua
t
io
n
m
et
rics
W
e
co
n
d
u
cte
d
t
h
e
ev
al
u
atio
n
o
f
im
a
g
e
q
u
ality
t
h
r
o
u
g
h
th
e
u
tili
za
tio
n
o
f
th
e
Fré
ch
et
in
ce
p
tio
n
d
is
tan
ce
(
FID
)
[
4
9
]
,
[
5
0
]
.
FI
D
s
er
v
es
as
a
m
etr
ic
f
o
r
q
u
an
tify
in
g
th
e
d
is
s
im
ilar
ity
b
etw
ee
n
r
ea
l
im
ag
es
a
n
d
g
en
er
ated
im
ag
es.
T
h
is
ca
lcu
l
atio
n
is
p
er
f
o
r
m
ed
u
s
in
g
th
e
in
ce
p
tio
n
n
etwo
r
k
to
e
x
tr
ac
t f
ea
tu
r
es a
n
d
ass
ess
th
e
d
is
tan
ce
b
etwe
en
th
e
d
is
tr
ib
u
tio
n
s
o
f
r
ea
l
an
d
g
en
er
ate
d
im
ag
es.
FID
h
as
estab
lis
h
ed
its
elf
as
a
wid
ely
ad
o
p
ted
e
v
alu
atio
n
m
e
tr
ic
in
v
ar
io
u
s
im
ag
e
g
e
n
er
atio
n
task
s
[
2
2
]
,
[
3
1
]
,
[
5
1
]
.
An
ad
d
itio
n
al
m
etr
ic
em
p
lo
y
ed
f
o
r
ass
ess
in
g
th
e
m
o
d
el
is
th
e
lear
n
e
d
p
e
r
ce
p
tu
al
im
ag
e
p
atch
s
im
ilar
ity
(
L
PIPS)
[
5
2
]
.
L
PI
PS
m
ea
s
u
r
es
th
e
lear
n
ed
d
is
tan
ce
with
in
Alex
Net
[
5
3
]
d
e
ep
f
ea
tu
r
e
s
p
ac
e
to
ev
alu
ate
th
e
d
is
tan
ce
b
etwe
en
im
ag
e
p
atch
es.
A
h
ig
h
e
r
L
PIPS
v
alu
e
in
d
icate
s
g
r
ea
ter
d
is
s
im
ilar
ity
,
s
u
g
g
esti
n
g
a
m
o
r
e
s
u
b
s
tan
tial
d
if
f
e
r
en
ce
b
etwe
en
i
m
a
g
e
p
a
tch
es.
C
o
n
v
er
s
ely
,
a
lo
we
r
L
P
I
PS
v
alu
e
s
ig
n
if
ies
in
cr
ea
s
ed
s
im
ilar
ity
,
in
d
icatin
g
a
clo
s
er
m
atch
b
etwe
en
im
ag
e
p
atch
es.
T
h
is
m
etr
ic
o
p
er
ates
o
n
th
e
p
r
em
is
e
th
at
a
lar
g
er
L
PIPS
d
is
tan
ce
im
p
lies
m
o
r
e
p
r
o
n
o
u
n
ce
d
d
is
tin
ctio
n
s
,
wh
ile
a
s
m
alle
r
d
is
tan
ce
s
ig
n
if
ies
h
eig
h
ten
ed
v
is
u
al
r
esem
b
lan
ce
b
etwe
en
im
ag
e
p
atch
es.
5.
E
XP
E
R
I
M
E
N
T
S AN
D
RE
S
UL
T
S
W
e
in
tr
o
d
u
ce
d
th
e
ex
ten
s
iv
e
ex
p
er
im
en
ts
c
o
n
d
u
cted
o
n
o
u
r
p
r
o
p
o
s
ed
R
GDM
m
o
d
el
in
th
is
ch
ap
ter
.
T
h
e
d
ataset
u
n
d
er
g
o
es
a
p
r
eli
m
in
ar
y
p
h
ase
wh
er
e
it
is
s
u
b
jecte
d
to
a
m
o
d
if
ied
s
eg
m
e
n
tatio
n
m
o
d
el,
SAM
[
1
9
]
,
r
esu
ltin
g
in
th
e
g
en
er
at
io
n
o
f
p
r
ec
is
e
g
lass
es
b
in
ar
y
m
ask
s
f
o
r
ea
c
h
im
ag
e
.
Su
b
s
eq
u
en
tly
,
b
o
th
th
e
b
in
ar
y
m
ask
s
an
d
th
e
o
r
ig
in
al
im
ag
es
ar
e
f
ed
i
n
to
th
e
c
o
n
d
itio
n
in
g
m
o
d
u
le,
as
ex
p
o
u
n
d
e
d
i
n
th
e
p
r
ec
ed
in
g
s
ec
tio
n
.
T
h
is
m
o
d
u
le
o
r
ch
estr
ates
th
e
co
m
b
in
atio
n
an
d
p
r
ep
ar
atio
n
o
f
im
ag
es
in
to
m
ask
ed
v
er
s
io
n
s
,
with
th
e
ju
d
icio
u
s
ad
d
itio
n
o
f
g
au
s
s
ian
n
o
is
e
tar
g
eted
at
th
e
m
ask
ed
ar
ea
s
.
T
h
is
p
r
ep
ar
ato
r
y
s
tep
s
ets
th
e
s
tag
e
f
o
r
th
e
s
u
b
s
eq
u
en
t
d
en
o
is
in
g
p
r
o
ce
d
u
r
e.
T
h
e
co
m
b
in
ed
d
ata,
n
o
w
en
r
ich
ed
with
g
au
s
s
ian
n
o
is
e,
is
th
en
p
r
o
ce
s
s
ed
th
r
o
u
g
h
th
e
d
if
f
u
s
io
n
m
o
d
el,
k
ick
s
tar
tin
g
th
e
i
n
-
p
ain
tin
g
p
r
o
ce
s
s
to
g
en
e
r
ate
v
er
s
io
n
s
o
f
th
e
f
ac
ial
im
ag
es
with
r
em
o
v
ed
g
lass
es.
C
r
u
ci
ally
,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
h
o
wca
s
es
its
v
er
s
atili
ty
b
y
n
o
t
o
n
ly
ef
f
ec
tiv
ely
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
R
emo
ve
g
la
s
s
es d
iffu
s
io
n
mo
d
el
a
n
in
n
o
va
tive
co
n
d
itio
n
ed
o
f e
ye
g
la
s
s
es remo
va
l wi
th
…
(
Yu
liz
a
)
1511
r
em
o
v
in
g
clea
r
-
len
s
ed
g
lass
es
b
u
t
also
d
em
o
n
s
tr
atin
g
p
r
o
f
ic
ien
cy
in
h
an
d
lin
g
d
ar
k
-
len
s
ed
v
ar
ieties,
in
clu
d
in
g
s
u
n
g
lass
es.
T
h
is
ad
ap
tab
ilit
y
u
n
d
er
s
co
r
es
th
e
r
o
b
u
s
tn
ess
o
f
th
e
R
GDM
m
eth
o
d
ac
r
o
s
s
a
s
p
ec
tr
u
m
o
f
ey
eg
lass
ty
p
es,
en
h
an
ci
n
g
its
ap
p
licab
il
ity
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
.
Fig
u
r
e
4
s
h
o
ws
th
e
o
v
er
v
iew
p
r
o
ce
s
s
o
f
h
o
w
th
e
ex
p
er
im
en
t
is
co
n
d
u
cted
in
o
r
d
er
to
o
b
tain
t
h
e
n
ec
ess
ar
y
r
esu
lts
.
As
ca
n
b
e
s
ee
n
in
t
h
e
f
i
g
u
r
e,
th
e
s
eg
m
en
tatio
n
p
r
o
ce
s
s
p
r
o
v
id
es
t
h
e
n
ec
ess
ar
y
g
lass
es
b
in
ar
y
m
ask
,
wh
ich
is
th
en
p
a
s
s
ed
th
r
o
u
g
h
th
e
co
n
d
itio
n
in
g
m
o
d
u
le
alo
n
g
with
th
e
o
r
ig
i
n
a
l
im
ag
e,
in
o
r
d
er
t
o
p
r
o
d
u
ce
th
e
n
ec
ess
ar
y
co
n
d
iti
o
n
f
o
r
th
e
DDPM
to
r
em
o
v
e
t
h
e
g
lass
es.
T
h
e
o
u
tp
u
t
o
f
th
e
c
o
n
d
itio
n
in
g
m
o
d
u
le
is
a
co
m
b
in
atio
n
o
f
th
e
o
r
i
g
in
al
im
ag
e
an
d
th
e
b
in
ar
y
m
ask
wh
ich
h
as
b
ee
n
ad
d
ed
with
g
a
u
s
s
ian
n
o
is
e
in
th
e
co
n
d
itio
n
in
g
ar
ea
.
E
y
es
p
o
s
itio
n
ac
cu
r
ac
y
,
in
th
is
ca
s
e
=
(
68
,
112
|
109
,
112
)
,
alo
n
g
with
th
e
o
u
tp
u
t
f
r
o
m
th
e
co
n
d
itio
n
in
g
m
o
d
u
le
is
th
en
p
ass
ed
th
r
o
u
g
h
th
e
d
en
o
is
in
g
p
r
o
ce
s
s
wh
ich
f
in
ally
g
en
er
ates
th
e
tar
g
ete
d
o
u
tp
u
t
im
a
g
e.
T
h
e
r
esu
lt
s
h
o
ws
th
at
th
e
p
r
o
p
o
s
ed
R
GDM
m
eth
o
d
s
u
cc
ess
f
u
lly
r
em
o
v
ed
th
e
g
lass
es
f
r
o
m
t
h
e
o
r
ig
in
al
im
ag
e.
Fig
u
r
e
4
.
R
GDM
:
g
lass
es r
em
o
v
al
p
ip
elin
e
I
n
th
is
d
etailed
an
aly
s
is
,
we
m
eticu
lo
u
s
ly
ass
ess
th
e
p
er
f
o
r
m
an
ce
im
p
ac
t
o
f
th
e
in
tr
o
d
u
ce
d
ey
es
p
o
s
itio
n
ac
cu
r
ac
y
in
th
e
p
r
o
p
o
s
ed
e
y
eg
lass
r
em
o
v
al
f
r
am
ewo
r
k
.
As
elu
cid
ated
in
Fig
u
r
e
5
,
th
e
im
p
lem
en
tatio
n
o
f
ey
e
s
p
o
s
itio
n
ac
cu
r
ac
y
p
lay
s
a
p
iv
o
tal
r
o
le
in
en
s
u
r
in
g
th
e
g
en
er
ated
e
y
es
alig
n
s
ea
m
less
ly
with
th
eir
in
ten
d
ed
p
o
s
itio
n
s
.
Par
ticu
lar
ly
,
f
o
r
clea
r
-
len
s
ed
g
lass
es,
th
e
ac
cu
r
ac
y
m
ec
h
an
is
m
g
u
ar
an
tees
th
at
th
e
g
en
er
ate
d
ey
es
p
r
ec
is
ely
o
cc
u
p
y
th
eir
o
r
ig
in
al
lo
ca
ti
o
n
s
.
T
h
is
m
eticu
lo
u
s
alig
n
m
e
n
t
b
ec
o
m
es
eq
u
all
y
cr
u
cial
f
o
r
d
a
r
k
-
len
s
ed
g
lass
es
o
r
s
u
n
g
lass
es,
wh
er
e
th
e
g
en
er
ate
d
ey
es
m
ain
tain
n
o
r
m
al
d
is
tan
ce
s
an
d
p
r
o
p
o
r
tio
n
s
c
o
n
s
is
ten
t
with
t
h
e
av
er
a
g
e
f
ac
e.
I
n
s
tar
k
co
n
tr
ast,
wh
en
th
e
m
o
d
el
o
p
er
at
es
with
o
u
t
th
e
ey
es
p
o
s
itio
n
ac
cu
r
ac
y
f
ea
tu
r
e,
t
h
e
g
en
er
ated
e
y
es
ex
h
ib
it
ir
r
eg
u
l
ar
p
lace
m
en
ts
,
ap
p
ea
r
in
g
h
i
g
h
er
th
an
th
e
o
r
ig
in
al
p
o
s
itio
n
an
d
u
n
co
m
f
o
r
tab
l
y
clo
s
e
to
th
e
p
er
s
o
n
'
s
ey
eb
r
o
ws.
T
h
is
d
iv
er
g
en
ce
r
esu
lts
in
u
n
n
at
u
r
al
a
n
d
o
cc
asio
n
ally
u
n
s
ettlin
g
im
ag
e
s
.
C
o
n
v
er
s
ely
,
wh
en
th
e
m
o
d
el
in
co
r
p
o
r
ates
th
e
ey
es
p
o
s
itio
n
ac
cu
r
ac
y
,
t
h
e
g
en
er
ated
e
y
es
r
etain
th
eir
a
u
th
en
ticity
in
th
e
o
r
ig
in
al
p
o
s
itio
n
,
cu
lm
in
atin
g
in
f
ac
ial
i
m
ag
es
th
at
ex
u
d
e
a
h
eig
h
ten
ed
s
en
s
e
o
f
n
at
u
r
alis
m
an
d
r
ea
lis
m
.
T
h
is
n
u
a
n
ce
d
ev
alu
atio
n
u
n
d
er
s
co
r
es
th
e
i
n
d
is
p
en
s
ab
le
r
o
le
o
f
ey
es
p
o
s
itio
n
ac
cu
r
ac
y
in
e
n
h
an
cin
g
th
e
v
is
u
al
f
id
elity
an
d
b
eliev
ab
ilit
y
o
f
t
h
e
g
en
e
r
ated
im
ag
es,
m
ar
k
in
g
it
as a
cr
itical
co
m
p
o
n
e
n
t in
th
e
o
v
er
all
s
u
cc
ess
o
f
th
e
p
r
o
p
o
s
ed
ey
eg
lass
r
em
o
v
al
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
8
,
No
.
3
,
J
u
n
e
20
2
5
:
1
5
0
3
-
1
5
1
6
1512
Fig
u
r
e
5
.
Ab
latio
n
s
tu
d
y
.
C
o
m
p
ar
is
o
n
o
n
g
lass
es r
em
o
v
al
r
es
u
lts
with
an
d
with
o
u
t e
y
es
p
o
s
itio
n
ac
cu
r
ac
y
,
o
n
th
e
C
eleb
A
d
ataset
6.
DIS
CU
SS
I
O
N
W
e
also
co
m
p
ar
e
d
o
u
r
m
o
d
el
to
o
th
er
cu
r
r
en
t
s
tate
-
of
-
th
e
-
a
r
t
m
o
d
els.
T
h
e
p
r
o
p
o
s
ed
R
GDM
m
eth
o
d
u
n
d
er
g
o
es
a
r
ig
o
r
o
u
s
co
m
p
a
r
a
tiv
e
an
aly
s
is
ag
ain
s
t
c
o
n
tem
p
o
r
a
r
y
s
tate
-
of
-
th
e
-
ar
t
g
lass
es
r
em
o
v
al
tec
h
n
iq
u
es,
in
clu
d
in
g
C
y
cleG
AN
[
5
4
]
,
E
R
GAN
[
3
5
]
,
an
d
3
D
Sy
n
t
h
etic
[
4
0
]
.
T
o
en
s
u
r
e
a
f
air
a
n
d
o
b
jectiv
e
co
m
p
a
r
is
o
n
,
ea
ch
m
eth
o
d
is
m
eticu
l
o
u
s
ly
tr
ain
ed
u
n
d
er
id
en
tical
e
x
p
er
i
m
en
tal
co
n
d
itio
n
s
,
m
ir
r
o
r
in
g
th
e
cir
cu
m
s
tan
ce
s
em
p
lo
y
ed
in
tr
ain
in
g
th
e
R
GDM
m
o
d
el.
T
h
is
m
eticu
lo
u
s
a
p
p
r
o
ac
h
to
ex
p
er
im
en
tatio
n
s
e
r
v
es
to
elu
cid
ate
th
e
r
el
ativ
e
s
tr
en
g
th
s
an
d
wea
k
n
e
s
s
es
o
f
ea
ch
m
eth
o
d
,
allo
win
g
f
o
r
a
c
o
m
p
r
e
h
en
s
iv
e
ass
ess
m
en
t
o
f
th
e
R
GDM
m
o
d
el'
s
p
er
f
o
r
m
an
ce
in
co
m
p
ar
is
o
n
to
its
co
u
n
ter
p
ar
ts
.
6
.
1
.
Q
ua
lit
a
t
iv
e
e
v
a
lua
t
io
ns
I
n
a
m
e
t
i
c
u
l
o
u
s
q
u
a
l
i
t
at
i
v
e
a
s
s
e
s
s
m
e
n
t
,
w
e
c
o
n
d
u
c
t
e
d
a
c
o
m
p
a
r
a
t
i
v
e
a
s
s
es
s
m
e
n
t
o
f
o
u
r
m
o
d
e
l
a
g
a
i
n
s
t
p
r
o
m
i
n
e
n
t
g
e
n
e
r
a
t
i
v
e
m
e
t
h
o
d
s
m
e
n
t
i
o
n
e
d
a
b
o
v
e
.
F
i
g
u
r
e
6
i
l
lu
s
t
r
a
t
es
t
h
e
e
v
a
l
u
at
i
o
n
o
f
t
h
e
q
u
a
l
i
t
y
o
f
g
e
n
e
r
at
e
d
g
l
a
s
s
e
s
-
r
e
m
o
v
e
d
i
m
a
g
e
s
o
n
t
h
e
C
el
e
b
A
d
at
a
s
et
.
T
h
r
o
u
g
h
l
e
v
e
r
a
g
i
n
g
t
h
e
o
p
e
n
-
s
o
u
r
c
e
c
o
d
es
,
t
h
e
s
t
a
t
e
-
of
-
t
h
e
-
a
r
t
m
e
t
h
o
d
s
—
C
y
c
l
eG
A
N
[
5
4
]
,
E
R
G
AN
[
3
5
]
,
a
n
d
3
D
S
y
n
t
h
e
t
i
c
[
4
0
]
—
a
r
e
r
e
i
m
p
l
e
m
e
n
t
e
d
f
o
r
b
e
n
c
h
m
a
r
k
i
n
g
.
A
s
d
e
p
i
c
t
e
d
i
n
Fi
g
u
r
e
6
,
c
o
m
p
a
r
e
d
t
o
o
u
r
a
p
p
r
o
a
c
h
,
C
y
c
l
e
G
AN
,
E
R
G
A
N
,
a
n
d
3
D
S
y
n
t
h
e
tic
a
l
l
h
a
v
e
t
h
ei
r
o
w
n
l
i
m
i
t
at
i
o
n
s
i
n
r
e
m
o
v
i
n
g
t
h
e
g
l
a
s
s
e
s
f
r
o
m
t
h
e
f
a
c
ia
l
i
m
a
g
e
s
.
B
o
t
h
C
y
c
l
eG
AN
a
n
d
3
D
S
y
n
t
h
e
t
i
c
e
x
h
i
b
it
l
i
m
i
t
at
i
o
n
s
,
f
ai
l
i
n
g
t
o
c
o
m
p
l
et
el
y
e
l
i
m
i
n
at
e
g
l
as
s
es
f
r
a
m
es
a
n
d
l
e
a
v
i
n
g
d
i
s
c
e
r
n
i
b
l
e
e
v
i
d
e
n
c
e
o
f
t
h
e
i
r
p
r
e
s
e
n
c
e
i
n
t
h
e
i
m
a
g
e
s
.
C
o
n
v
e
r
s
el
y
,
E
R
GA
N
f
a
l
ls
s
h
o
r
t
i
n
d
e
l
i
v
e
r
i
n
g
r
ea
l
i
s
t
i
c
a
n
d
c
le
a
r
i
m
a
g
es
w
h
il
e
a
t
t
e
m
p
t
i
n
g
r
e
m
o
v
a
l
,
o
c
c
a
s
i
o
n
a
ll
y
l
ea
v
i
n
g
s
u
b
t
l
e
t
r
ac
e
s
o
f
g
l
as
s
es
e
x
i
s
t
e
n
c
e
,
a
l
b
e
i
t
l
es
s
p
r
o
m
i
n
e
n
t
t
h
a
n
t
h
e
f
o
r
m
e
r
t
w
o
m
e
t
h
o
d
s
.
I
n
s
t
a
r
k
c
o
n
t
r
as
t
,
o
u
r
a
p
p
r
o
a
c
h
y
i
e
l
d
s
r
e
s
u
l
ts
t
h
a
t
a
r
e
m
a
r
k
e
d
l
y
t
r
u
e
t
o
l
i
f
e
a
n
d
a
u
t
h
e
n
t
i
c
.
T
h
e
g
l
a
s
s
e
s
f
r
a
m
es
a
r
e
i
m
p
e
c
c
a
b
l
y
r
e
m
o
v
e
d
w
i
t
h
o
u
t
l
e
a
v
i
n
g
a
n
y
t
r
a
c
e
s
o
r
a
n
o
m
a
l
o
u
s
i
n
p
a
i
n
t
i
n
g
a
r
t
i
f
a
c
ts
.
F
u
r
t
h
e
r
m
o
r
e
,
t
h
e
g
e
n
e
r
a
t
e
d
e
y
e
s
f
a
i
t
h
f
u
ll
y
r
e
p
l
i
c
a
te
t
h
e
o
r
ig
i
n
a
l
e
y
e
s
f
o
r
g
l
a
s
s
es
w
it
h
c
l
e
ar
l
e
n
s
e
s
.
F
o
r
g
l
a
s
s
es
w
it
h
c
o
l
o
u
r
e
d
o
r
d
a
r
k
l
e
n
s
e
s
,
t
h
e
g
e
n
e
r
a
t
e
d
e
y
e
s
h
a
r
m
o
n
i
ze
s
e
a
m
l
es
s
l
y
w
it
h
t
h
e
o
v
e
r
a
ll
f
a
c
i
al
c
o
n
t
e
x
t
,
e
n
s
u
r
i
n
g
a
n
a
t
u
r
a
l
a
n
d
r
e
al
is
t
i
c
a
p
p
e
a
r
a
n
c
e
i
n
t
h
e
g
e
n
e
r
a
t
e
d
i
m
a
g
e
s
.
T
h
i
s
p
r
o
n
o
u
n
c
e
d
s
u
p
e
r
i
o
r
i
t
y
i
n
r
e
a
l
is
m
a
n
d
p
r
e
c
i
s
io
n
u
n
d
e
r
s
c
o
r
e
s
t
h
e
e
f
f
i
c
a
c
y
o
f
t
h
e
p
r
o
p
o
s
e
d
R
G
DM
m
o
d
e
l
i
n
a
c
h
i
e
v
i
n
g
h
i
g
h
-
f
i
d
e
l
i
t
y
g
l
as
s
e
s
r
e
m
o
v
a
l
.
Fig
u
r
e
6
.
Qu
alitativ
e
r
esu
lts
.
C
o
m
p
ar
is
o
n
ag
ain
s
t o
th
e
r
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
f
o
r
g
lass
es r
em
o
v
al
o
n
th
e
C
eleb
A
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.