I
nte
rna
t
io
na
l J
o
urna
l o
f
Rec
o
nfig
ura
ble a
nd
E
m
be
dd
e
d Sy
s
t
e
m
s
(
I
J
R
E
S)
Vo
l.
15
,
No
.
1
,
Ma
r
ch
20
26
,
p
p
.
2
2
4
~
2
3
5
I
SS
N:
2089
-
4864
,
DOI
:
1
0
.
1
1
5
9
1
/i
j
r
es
.
v
1
5
.
i
1
.
pp
2
2
4
-
235
224
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
r
es.ia
esco
r
e.
co
m
Sy
na
ptic
shield
:
f
usio
n
o
f
ResN
ex
t
–
5
0
and
lo
ng
sho
rt
-
ter
m
m
e
m
o
ry
for enha
na
ced deep
fa
k
e d
etec
tion
A
m
it
M
is
hra
1
,
P
ra
j
w
a
l C
hin
ch
m
a
la
t
pu
re
2
,
G
o
v
ind
a
B
.
S
a
m
ba
re
3
,
Vio
m
e
s
h K
u
m
a
r
Si
ng
h
4
,
At
ul
G
ula
bra
o
P
a
w
a
r
3
,
Ra
hu
l P
r
a
k
a
s
h M
ira
j
k
a
r
5
,
P
riy
a
nk
a
K
.
T
a
k
a
l
k
a
r
6
,
K
uld
ee
p Va
y
a
da
nd
e
4
1
D
r
.
V
i
s
h
w
a
n
a
t
h
K
a
r
a
d
M
I
T
W
o
r
l
d
P
e
a
c
e
U
n
i
v
e
r
si
t
y
,
P
u
n
e
,
I
n
d
i
a
2
N
o
r
t
h
e
a
st
e
r
n
U
n
i
v
e
r
si
t
y
,
B
o
st
o
n
,
U
n
i
t
e
d
S
t
a
t
e
s
3
P
i
mp
r
i
C
h
i
n
c
h
w
a
d
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
P
u
n
e
,
I
n
d
i
a
4
V
i
sh
w
a
k
a
r
ma
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
P
u
n
e
,
I
n
d
i
a
5
B
h
a
r
a
t
i
V
i
d
y
a
p
e
e
t
h
’
s
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
K
o
l
h
a
p
u
r
,
I
n
d
i
a
6
B
h
a
r
a
t
i
V
i
d
y
a
p
e
e
t
h
’
s
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
1
,
2
0
2
5
R
ev
i
s
ed
J
an
1
4
,
2
0
2
6
A
cc
ep
ted
Feb
1
7
,
2
0
2
6
Re
c
e
n
t
d
e
v
e
lo
p
m
e
n
ts
in
d
e
e
p
f
a
k
e
s
h
a
v
e
c
re
a
ted
m
u
c
h
a
n
x
iet
y
a
b
o
u
t
t
h
e
a
u
th
e
n
ti
c
it
y
o
f
a
n
y
d
ig
it
a
l
c
o
n
ten
t
a
n
d
th
u
s,
c
a
ll
s
f
o
r
i
m
p
le
m
e
n
ti
n
g
d
e
tec
ti
o
n
m
e
c
h
a
n
is
m
s
th
a
t
w
il
l
w
o
rk
a
c
c
o
rd
in
g
ly
.
T
h
is
p
a
p
e
r
u
se
s
S
y
n
a
p
ti
c
S
h
iel
d
,
a
in
n
o
v
a
ti
v
e
d
e
e
p
lea
rn
in
g
(DL
)
fra
m
e
w
o
rk
w
h
ich
is
c
u
sto
m
ize
d
to
d
e
tec
t
a
lt
e
ra
ti
o
n
s
b
y
d
e
e
p
fa
k
e
s
w
it
h
h
ig
h
p
re
c
isio
n
lev
e
ls.
It
e
m
p
lo
y
s
b
o
th
c
o
n
v
o
lu
ti
o
n
n
e
u
ra
l
n
e
tw
o
rk
s
(CNN
s)
a
s
we
ll
a
s
m
o
d
u
les
f
o
r
ti
m
e
f
e
a
tu
re
e
x
trac
ti
o
n
s
to
tes
t
sp
a
ti
a
l
a
n
d
m
o
ti
o
n
in
d
ica
to
rs
f
ro
m
v
id
e
o
d
a
ta.
Hig
h
-
lev
e
l
p
re
p
ro
c
e
ss
in
g
p
ip
e
li
n
e
s
in
c
o
m
b
in
a
ti
o
n
w
it
h
c
o
n
f
id
e
n
c
e
sc
o
rin
g
m
e
c
h
a
n
is
m
h
e
lp
m
a
k
e
S
y
n
a
p
ti
c
S
h
ield
a
d
a
p
t
i
v
e
to
w
a
rd
m
a
n
ip
u
latio
n
tec
h
n
iq
u
e
s
su
c
h
a
s
F
a
c
e
S
wa
p
a
n
d
De
e
p
F
a
k
e
.
T
h
e
a
c
c
u
ra
c
y
o
f
o
u
r
m
o
d
e
l
su
rp
a
ss
e
s
o
th
e
r
d
e
e
p
f
a
k
e
d
e
tec
ti
o
n
m
o
d
e
ls
w
it
h
a
h
ig
h
a
c
c
u
ra
c
y
o
f
9
8
.
3
%
.
T
h
e
a
b
o
v
e
re
su
lt
s
a
re
b
a
se
d
o
n
e
x
h
a
u
stiv
e
e
x
p
e
rime
n
tatio
n
o
n
sta
n
d
a
rd
d
a
tas
e
ts
li
k
e
F
a
c
e
F
o
re
n
sic
s+
+
,
De
e
p
F
a
k
e
d
e
tec
ti
o
n
c
h
a
ll
e
n
g
e
(DFDC),
a
n
d
Ce
leb
De
e
p
F
a
k
e
(Ce
leb
-
DF).
S
y
n
a
p
ti
c
S
h
iel
d
is
sh
o
w
n
to
b
e
t
h
e
b
e
st
w
it
h
o
u
tstan
d
in
g
re
su
lt
s
th
a
t
m
a
in
tain
a
h
ig
h
e
r
c
o
n
f
id
e
n
c
e
sc
o
re
e
q
u
iv
a
len
t
to
it
s
p
re
c
isio
n
a
n
d
re
li
a
b
i
li
ty
.
S
c
a
lab
il
it
y
in
h
a
v
in
g
th
e
c
a
p
a
c
it
y
to
a
c
c
o
m
m
o
d
a
te
v
a
rio
u
s
m
a
n
ip
u
latio
n
tec
h
n
i
q
u
e
s
a
n
d
lev
e
ls
o
f
v
id
e
o
q
u
a
li
ty
in
d
ica
tes
ro
b
u
stn
e
ss
i
n
o
f
f
e
rin
g
a
n
e
ff
e
c
ti
v
e
m
e
th
o
d
to
w
a
rd
e
n
s
u
rin
g
in
teg
rit
y
in
d
ig
it
a
l
m
e
d
ia.
T
h
e
w
o
rk
is
a
n
im
p
o
rtan
t
m
o
v
e
f
o
rwa
rd
in
a
d
d
r
e
ss
in
g
th
e
p
ro
b
lem
s c
re
a
ted
b
y
De
e
p
F
a
k
e
te
c
h
n
o
l
o
g
ies
.
K
ey
w
o
r
d
s
:
A
r
ti
f
icial
i
n
tel
lig
e
n
ce
C
o
m
p
u
ter
v
is
io
n
Dee
p
f
ak
e
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
R
ec
u
r
r
en
t n
e
u
r
al
n
et
w
o
r
k
R
es
-
n
e
x
t b
ased
co
n
v
o
lu
tio
n
n
eu
r
al
n
et
w
o
r
k
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
R
ah
u
l P
r
ak
as
h
Mir
aj
k
ar
B
h
ar
ati
Vid
y
ap
ee
th
’
s
C
o
lle
g
e
o
f
E
n
g
in
ee
r
i
n
g
Ko
lh
ap
u
r
,
I
n
d
ia
E
m
ail:
r
ah
u
l
m
ir
aj
k
ar
9
8
2
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
is
is
a
n
e
w
s
y
n
t
h
etic
m
ed
ia
cr
ea
tio
n
to
o
l
b
ased
o
n
b
r
ea
k
t
h
r
o
u
g
h
s
i
n
d
ee
p
lear
n
i
n
g
(
DL
)
t
h
at
i
s
D
L
an
d
g
e
n
er
ati
v
e
m
o
d
el
s
,
esp
ec
i
all
y
i
n
g
e
n
er
ativ
e
ad
v
er
s
ar
ial
n
et
w
o
r
k
(
G
AN)
[
1
]
,
w
h
ic
h
ca
n
i
m
p
lan
t
n
e
w
f
ac
es
an
d
v
o
ices
o
n
to
v
id
eo
s
in
v
er
y
r
ea
li
s
tic
w
a
y
s
.
Su
c
h
ar
tif
ici
al
in
telli
g
en
ce
(
A
I
)
-
g
en
er
ated
v
id
eo
s
,
k
n
o
w
n
as
Dee
p
Fak
es,
ca
n
s
er
io
u
s
l
y
co
m
p
r
o
m
is
e
th
e
v
er
ac
it
y
o
f
d
i
g
ital
m
ed
ia
b
ec
a
u
s
e
it
ca
n
b
e
b
en
ef
icia
l
f
o
r
eit
h
er
s
p
r
ea
d
in
g
m
is
in
f
o
r
m
at
io
n
o
r
m
an
ip
u
lati
n
g
p
u
b
lic
o
p
in
io
n
t
h
r
o
u
g
h
d
a
m
a
g
i
n
g
r
ep
u
tat
io
n
s
[
2
]
,
[
3
]
.
T
h
e
in
cr
ea
s
i
n
g
a
v
ailab
ilit
y
an
d
s
o
p
h
is
ticatio
n
o
f
Dee
p
Fak
e
g
en
er
atio
n
to
o
ls
r
aise
th
e
s
ta
k
es,
s
o
it
is
clea
r
l
y
n
ec
es
s
ar
y
th
at
d
etec
tio
n
m
ec
h
an
is
m
s
b
e
d
ev
elo
p
ed
th
at
ar
e
r
eliab
le
[
4
]
.
No
w
d
a
y
s
m
et
h
o
d
s
u
s
u
a
ll
y
r
el
y
o
n
p
ix
el
-
b
ased
a
n
al
y
s
i
s
o
r
s
i
m
p
l
e
s
tati
s
tical
m
o
d
el
s
t
h
at
h
a
v
e
lo
s
t
m
o
m
e
n
tu
m
d
u
e
to
m
o
d
er
n
tech
n
iq
u
e
s
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
S
yn
a
p
tic
s
h
ield
:
fu
s
io
n
o
f R
es
N
ex
t
–
5
0
a
n
d
lo
n
g
s
h
o
r
t
-
term me
mo
r
y
fo
r
en
h
a
n
a
ce
d
…
(
A
mit Mis
h
r
a
)
225
p
r
o
d
u
cin
g
f
e
w
er
id
e
n
ti
f
iab
le
ar
tif
ac
ts
[
5
]
.
W
e
ap
p
r
o
ac
h
th
e
ch
alle
n
g
e
u
s
i
n
g
o
u
r
n
o
v
el
D
L
-
b
ased
f
r
a
m
e
w
o
r
k
th
at
m
a
k
es
a
a
m
al
g
a
m
atio
n
o
f
co
n
v
o
lu
tio
n
n
e
u
r
al
n
et
w
o
r
k
(
C
NN)
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
e
t
w
o
r
k
(
R
NN)
to
r
o
b
u
s
tl
y
cla
s
s
i
f
y
if
t
h
e
g
iv
e
n
v
id
eo
is
a
Dee
p
Fak
e
[
6
]
.
Ou
r
ap
p
r
o
ac
h
u
s
es
a
R
esNe
x
t
C
NN,
w
h
ic
h
ex
tr
ac
t
s
f
r
a
m
ele
v
el
f
ea
t
u
r
es
a
n
d
d
etec
ts
m
i
n
o
r
ar
tef
ac
ts
lef
t
b
y
Dee
p
Fak
e
g
e
n
er
atio
n
p
r
o
ce
s
s
es;
th
ese
ar
e
th
en
p
r
o
ce
s
s
ed
b
y
t
h
e
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
ST
M)
n
et
w
o
r
k
-
a
v
ar
ian
t
o
f
an
R
NN,
ca
p
tu
r
in
g
te
m
p
o
r
al
in
co
n
s
i
s
te
n
cies
in
v
id
eo
f
r
a
m
es,
w
h
ic
h
i
s
t
h
e
m
o
s
t c
o
m
m
o
n
in
d
icato
r
o
f
m
an
ip
u
latio
n
[
7
]
.
T
h
e
in
teg
r
atio
n
o
f
s
p
atial
a
n
d
te
m
p
o
r
al
an
al
y
s
is
b
r
in
g
s
o
u
r
m
o
d
el
to
h
ig
h
ac
cu
r
ac
y
a
n
d
r
o
b
u
s
t
n
es
s
to
m
an
y
D
ee
p
Fak
e
s
ce
n
ar
io
s
[
8
]
,
[
9
]
.
W
e
ex
am
in
e
o
u
r
m
e
th
o
d
o
n
a
lar
g
e
d
ataset
co
m
p
o
s
ed
o
f
b
o
th
r
ea
l
an
d
m
an
ip
u
lated
v
id
eo
s
.
I
t
s
u
r
p
ass
e
s
p
r
ev
io
u
s
m
et
h
o
d
s
,
s
o
it
co
u
ld
b
e
a
g
o
o
d
to
o
l
f
o
r
th
e
d
etec
tio
n
o
f
d
ee
p
f
a
k
es
i
n
v
ar
io
u
s
r
ea
l
-
w
o
r
ld
tas
k
s
,
s
u
c
h
as
s
o
cial
p
lat
f
o
r
m
m
o
n
ito
r
in
g
a
n
d
d
ig
ita
l
f
o
r
en
s
ic
s
[
1
0
]
.
A
I
an
d
D
L
h
a
v
e
led
to
tr
em
e
n
d
o
u
s
i
n
n
o
v
atio
n
s
in
t
h
e
g
e
n
er
atio
n
an
d
m
an
ip
u
latio
n
o
f
m
ed
ia
[
1
1
]
.
Am
id
s
t
all
t
h
ese
d
ev
e
lo
p
m
en
ts
,
o
n
e
s
u
c
h
d
o
u
b
le
-
ed
g
ed
to
o
l
is
Dee
p
Fak
e
tech
n
o
lo
g
y
,
s
u
p
p
o
r
ted
b
y
GANs
b
esid
es
o
th
er
f
r
a
m
e
w
o
r
k
s
o
f
g
en
er
atio
n
.
Dis
tin
g
u
is
h
i
n
g
b
et
w
ee
n
r
ea
l
a
n
d
f
a
k
e
v
id
eo
s
,
ch
an
g
i
n
g
f
ac
e
s
,
o
r
ev
en
v
o
ice
i
m
i
tatio
n
is
f
ab
u
l
o
u
s
i
n
e
n
ter
tai
n
m
e
n
t,
ed
u
ca
ti
o
n
,
an
d
ar
tis
tr
y
,
b
u
t
i
m
p
r
o
p
er
u
s
e
h
as
b
r
o
u
g
h
t
cr
itical
p
ea
k
s
in
et
h
ical
an
d
s
e
cu
r
it
y
i
s
s
u
es
[
1
2
]
,
[
1
3
]
.
Dee
p
Fak
es
h
av
e
o
cc
u
r
r
ed
as
o
n
e
o
f
th
e
m
o
s
t
i
m
p
o
r
tan
t
th
r
ea
ts
w
i
th
i
n
t
h
e
d
ig
ita
l
s
p
a
ce
,
p
u
ttin
g
co
n
ce
r
n
s
o
n
w
h
e
th
er
tech
n
o
lo
g
y
ca
n
b
r
ea
k
t
h
e
tr
u
s
t
d
ev
elo
p
ed
b
et
w
ee
n
v
i
s
u
a
l a
n
d
au
d
io
co
n
t
en
t [
1
4
]
.
Dee
p
Fak
e
v
id
eo
s
ar
e
es
s
en
t
i
all
y
h
i
g
h
l
y
r
ea
li
s
tic
m
a
n
ip
u
la
ted
o
r
ev
en
f
a
k
ed
ev
e
n
ts
th
a
t
th
r
ea
te
n
p
r
iv
ac
y
,
p
u
b
lic
s
ec
u
r
it
y
,
a
n
d
s
o
cial
co
h
esio
n
.
S
u
c
h
v
id
eo
s
m
a
y
b
e
e
m
p
lo
y
ed
in
a
p
r
o
p
ag
an
d
a
ca
m
p
ai
g
n
o
f
m
is
in
f
o
r
m
at
io
n
,
p
o
liti
ca
l
co
er
cio
n
,
an
d
id
en
tit
y
th
e
f
t
a
m
o
n
g
s
t
o
th
er
n
ef
ar
io
u
s
g
o
al
s
[
1
5
]
,
[
1
6
]
.
T
h
er
ef
o
r
e,
a
Dee
p
Fak
e
v
id
eo
o
f
a
p
r
o
m
in
e
n
t
p
u
b
lic
f
ig
u
r
e
m
a
y
b
e
u
s
e
f
u
l
f
o
r
s
p
r
ea
d
in
g
f
a
ls
e
i
n
f
o
r
m
ati
o
n
o
r
in
citin
g
ci
v
il
u
n
r
e
s
t,
w
h
er
ea
s
an
alter
ed
co
r
p
o
r
ate
v
id
eo
m
a
y
d
a
m
a
g
e
r
ep
u
tatio
n
o
r
m
is
lead
in
v
e
s
to
r
s
[
1
7
]
.
T
h
is
b
if
u
r
ca
ted
n
atu
r
e
o
f
d
ee
p
f
a
k
es
ca
lls
f
o
r
d
etec
tio
n
m
ec
h
a
n
is
m
s
t
h
at
ar
e
b
o
th
r
o
b
u
s
t
a
n
d
s
ca
lab
le,
y
et
s
p
ec
if
ic
en
o
u
g
h
to
m
atc
h
th
e
r
ap
id
ev
o
lu
t
io
n
o
f
t
h
ese
tec
h
n
o
lo
g
ies.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
R
esear
ch
b
y
Oa
k
[
1
8
]
p
r
esen
t
s
an
i
m
p
r
o
v
ed
d
ee
p
f
a
k
e
d
etec
t
io
n
tech
n
iq
u
e
o
v
er
Face
-
Xr
a
y
,
b
ased
o
n
co
n
tin
u
o
u
s
f
r
a
m
e
f
ac
e
-
s
w
ap
p
i
n
g
.
I
t
g
en
er
ate
s
m
as
k
s
t
h
at
ad
d
f
u
s
io
n
f
ea
t
u
r
es
to
v
id
eo
s
u
s
in
g
a
U
-
Net
-
b
ased
GAN
an
d
p
er
f
o
r
m
s
f
ac
e
-
s
w
a
p
p
in
g
u
s
i
n
g
Dela
u
n
a
y
tr
ia
n
g
u
latio
n
an
d
p
iece
w
is
e
a
f
f
in
e
tr
an
s
f
o
r
m
atio
n
.
W
ith
th
is
m
et
h
o
d
,
in
tr
a
-
f
r
a
m
e
f
u
s
i
o
n
an
d
in
ter
-
f
r
a
m
e
te
m
p
o
r
al
f
ea
tu
r
e
s
ar
e
g
u
ar
a
n
teed
to
b
e
p
r
esen
t
in
t
h
e
p
r
o
d
u
ce
d
v
id
eo
s
.
T
h
ese
f
ea
tu
r
es
ar
e
th
en
e
x
tr
ac
ted
u
s
i
n
g
an
E
f
f
icie
n
tNe
t
-
L
ST
M
m
o
d
el,
w
h
er
e
L
ST
M
co
n
ce
n
tr
ates
o
n
te
m
p
o
r
al
p
att
er
n
s
an
d
E
f
f
icie
n
tNet
ca
p
tu
r
e
s
s
p
atial
f
ea
t
u
r
es.
T
h
is
co
m
b
i
n
atio
n
h
elp
s
to
d
etec
t
d
ee
p
f
ak
e
ev
id
e
n
ce
ef
f
icie
n
tl
y
.
I
n
s
it
u
atio
n
s
in
v
o
lv
in
g
cr
o
s
s
-
d
ataset
d
etec
tio
n
,
th
e
ap
p
r
o
ac
h
s
h
o
w
s
en
h
a
n
ce
d
g
en
er
aliza
tio
n
an
d
attai
n
s
an
a
r
ea
u
n
d
er
th
e
cu
r
v
e
(
A
UC
)
o
f
0
.
8
4.
R
esear
ch
b
y
I
q
b
al
et
a
l.
[
1
9
]
ad
d
r
ess
es
th
e
lo
w
d
etec
t
io
n
a
cc
u
r
ac
y
o
f
ex
i
s
ti
n
g
Dee
p
Fak
e
d
etec
tio
n
m
et
h
o
d
s
in
cr
o
s
s
-
ar
c
h
i
v
e
s
ce
n
ar
io
s
an
d
lo
w
-
q
u
alit
y
v
id
eo
s
e
ts
.
Usi
n
g
a
n
et
w
o
r
k
s
tr
u
ct
u
r
e
t
h
at
co
m
b
i
n
es
v
id
eo
an
d
s
in
g
le
-
b
r
an
c
h
d
o
u
b
le
-
b
r
an
ch
d
etec
tio
n
to
g
at
h
er
s
p
ati
al
an
d
te
m
p
o
r
al
d
ata,
it
s
u
g
g
est
s
a
t
w
o
-
b
r
an
ch
d
ee
p
f
ak
e
d
etec
tio
n
m
et
h
o
d
.
A
lo
n
g
w
i
th
ap
p
l
y
in
g
d
if
f
er
en
t
d
ata
au
g
m
en
tatio
n
tec
h
n
iq
u
es,
th
e
m
eth
o
d
also
u
s
e
s
th
e
co
n
v
o
lu
t
io
n
al
b
lo
ck
a
tten
tio
n
m
o
d
u
le
(
C
B
A
M)
to
im
p
r
o
v
e
th
e
Xce
p
tio
n
n
et
w
o
r
k
.
C
o
m
p
ar
ati
v
e
test
s
u
s
i
n
g
v
ar
io
u
s
d
atasets
d
e
m
o
n
s
tr
ate
th
at
t
h
e
s
u
g
g
es
ted
n
et
w
o
r
k
m
o
d
el
p
er
f
o
r
m
s
ex
is
ti
n
g
id
ea
s
in
ter
m
s
o
f
d
etec
tio
n
p
er
f
o
r
m
an
ce
a
n
d
g
e
n
er
aliza
tio
n
ab
ilit
ies.
L
ai
et
a
l.
[
2
0
]
Dee
p
f
ak
es,
w
h
ic
h
p
o
s
e
s
er
io
u
s
r
is
k
s
li
k
e
m
a
n
ip
u
la
tin
g
p
u
b
lic
v
ie
w
,
cr
ea
tin
g
g
eo
p
o
liti
ca
l
ten
s
io
n
s
,
u
n
s
tab
l
e
f
in
a
n
cial
m
ar
k
ets,
s
ca
m
s
,
d
ef
a
m
at
io
n
,
an
d
f
in
d
t
h
e
f
t,
ar
e
a
o
u
tco
m
e
s
o
f
ad
v
an
ce
s
i
n
D
L
,
b
ig
d
ata,
an
d
i
m
ag
e
p
r
o
ce
s
s
i
n
g
.
T
h
is
s
tu
d
y
e
x
a
m
in
e
s
b
o
th
r
ec
en
t
a
n
d
e
m
er
g
i
n
g
tr
en
d
s
i
n
d
ee
p
f
ak
e
tech
n
o
lo
g
y
.
A
tea
m
o
f
b
r
ain
y
f
o
l
k
s
u
s
ed
t
w
o
s
m
ar
t
to
o
ls
to
ca
tch
r
o
g
u
e
ac
tiv
ities
o
n
co
m
p
u
ter
n
et
wo
r
k
s
[
2
1
]
.
On
e
to
o
l,
ca
lled
C
NN,
is
g
r
ea
t
at
f
in
d
in
g
th
e
p
atter
n
o
f
i
n
f
o
r
m
atio
n
i
n
ti
m
e.
T
h
e
o
th
er
,
k
n
o
w
n
as
L
ST
M,
is
a
w
h
izz
at
ca
p
tu
r
i
n
g
ch
a
n
g
e
s
o
v
er
ti
m
e.
T
o
g
eth
er
,
th
e
y
m
a
k
e
a
k
iller
co
m
b
o
f
o
r
s
n
i
f
f
i
n
g
o
u
t t
h
e
b
ad
g
u
y
s
.
R
esear
ch
b
y
A
ab
ito
v
a
et
a
l.
[
2
2
]
p
r
o
p
o
s
es
a
n
o
v
el
D
L
m
o
d
el
f
o
r
f
a
k
e
f
ac
e
d
etec
tio
n
in
m
ed
i
a
f
o
r
en
s
ic
s
,
w
h
ic
h
s
i
m
u
ltan
eo
u
s
l
y
ex
tr
ac
ts
co
n
te
n
t
an
d
tr
ac
e
ch
ar
ac
ter
is
tic
s
to
d
etec
t
m
an
ip
u
lated
f
ac
es.
R
ec
en
t
w
o
r
k
s
h
av
e
a
ls
o
ex
p
lo
r
ed
g
r
ap
h
n
e
u
r
al
n
et
w
o
r
k
-
b
a
s
ed
ap
p
r
o
ac
h
es
f
o
r
f
r
a
u
d
an
d
an
o
m
a
l
y
d
etec
t
io
n
i
n
co
m
p
le
x
r
elatio
n
al
d
ataset
s
[
2
3
]
.
C
o
m
p
ar
ativ
e
ev
al
u
atio
n
s
b
et
w
ee
n
g
r
ap
h
-
b
ased
lear
n
i
n
g
a
n
d
tr
ad
itio
n
al
ar
ch
itect
u
r
es
h
i
g
h
li
g
h
t
i
m
p
r
o
v
e
m
e
n
t
s
in
s
tr
u
ct
u
r
ed
d
etec
tio
n
s
ce
n
ar
io
s
[
2
4
]
.
Fu
r
t
h
er
m
o
r
e,
h
eter
o
g
en
eo
u
s
g
r
ap
h
tr
an
s
f
o
r
m
er
m
o
d
els
h
a
v
e
d
e
m
o
n
s
tr
ated
en
h
an
ce
d
ca
p
ab
ilit
y
in
m
o
d
elin
g
m
u
lt
i
-
s
o
u
r
ce
r
elatio
n
al
d
ata
f
o
r
d
etec
tio
n
tas
k
s
[
2
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
1
,
Ma
r
c
h
202
6
:
2
2
4
-
235
226
3.
M
E
T
H
O
D
3
.
1
.
Desig
n
3.
1
.
1.
I
np
ut
v
ideo
f
ra
m
es
T
h
is
p
ar
t
o
f
o
u
r
p
ap
er
ap
p
r
o
ac
h
es
a
n
o
v
er
v
ie
w
o
f
t
h
e
m
e
th
o
d
o
lo
g
y
w
e
ar
e
g
o
i
n
g
to
i
m
p
le
m
e
n
t.
Fo
llo
w
i
n
g
ar
e
th
e
p
h
ase
s
:
P
r
ep
r
o
ce
s
s
in
g
a
v
id
eo
:
t
ak
e
o
u
t in
d
iv
id
u
al
f
r
a
m
e
s
an
d
r
esize
th
e
m
to
2
2
4
b
y
2
2
4
p
ix
els.
No
r
m
a
lizatio
n
:
t
o
co
n
f
ir
m
co
n
s
is
te
n
c
y
ac
r
o
s
s
th
e
m
o
d
el,
n
o
r
m
alize
t
h
e
p
ix
el
v
al
u
es to
a
s
ta
n
d
ar
d
r
a
n
g
e.
R
esNe
x
t
f
o
r
ex
tr
ac
tio
n
o
f
f
ea
t
u
r
es:
−
Fra
m
e
-
by
-
f
r
a
m
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
: Re
s
Nex
t CNN p
r
o
ce
s
s
e
s
ea
ch
f
r
a
m
e,
ex
tr
ac
ti
n
g
2
0
4
8
-
d
im
en
s
io
n
a
l
f
ea
t
u
r
e
v
ec
to
r
s
.
−
T
h
e
g
o
al
o
f
R
esNex
t
is
to
r
e
co
r
d
ea
ch
f
r
a
m
e
'
s
s
p
atial
i
n
f
o
r
m
atio
n
,
s
u
ch
as
w
r
in
k
le
s
,
f
a
cial
lan
d
m
ar
k
s
,
u
n
e
v
e
n
lig
h
ti
n
g
,
an
d
ex
p
r
es
s
io
n
s
.
Fi
g
u
r
e
1
s
h
o
w
s
R
esNe
x
t F
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
Fig
u
r
e
1
.
R
es_
Nex
t5
0
_
3
2
x
4
d
(
f
ea
tu
r
e
e
x
tr
ac
tio
n
)
T
h
e
n
ex
t step
is
p
r
o
p
o
s
ed
s
y
s
t
e
m
h
a
s
b
ee
n
,
te
m
p
o
r
al
an
al
y
s
i
s
w
ith
L
ST
M:
−
Seq
u
en
ce
f
o
r
m
at
io
n
:
a
s
eq
u
en
ce
is
cr
ea
ted
b
y
o
r
g
a
n
izi
n
g
t
h
e
f
ea
tu
r
e
v
ec
to
r
s
t
h
at
wer
e
tak
en
f
r
o
m
s
u
cc
e
s
s
i
v
e
f
r
a
m
es.
−
L
ST
M
p
r
o
ce
s
s
in
g
:
t
h
e
L
ST
M
la
y
er
,
w
h
ic
h
h
as 2
0
4
8
laten
t d
i
m
e
n
s
io
n
s
a
n
d
is
i
n
te
n
d
ed
to
ca
p
tu
r
e
te
m
p
o
r
al
ch
an
g
es
b
et
w
ee
n
f
r
a
m
es,
i
n
cl
u
d
in
g
p
o
s
itio
n
s
h
i
f
ts
,
f
a
ce
m
o
v
e
m
e
n
t
s
,
an
d
b
lin
k
in
g
p
atter
n
s
,
is
ap
p
lied
to
th
ese
s
eq
u
en
ce
s
.
−
Dr
o
p
o
u
t:
t
o
av
o
id
o
v
er
f
itti
n
g
,
a
4
0
% d
r
o
p
o
u
t is i
m
p
le
m
en
ted
d
u
r
in
g
tr
ai
n
i
n
g
.
Af
ter
p
er
f
o
r
m
i
n
g
T
em
p
o
r
al
an
al
y
s
i
s
n
e
x
t s
tep
is
;
co
m
p
letel
y
n
et
w
o
r
k
ed
an
d
o
u
tp
u
t la
y
er
s
:
−
Fu
ll
y
co
n
n
ec
ted
la
y
er
:
an
o
u
tp
u
t
is
r
o
u
ted
th
r
o
u
g
h
a
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
u
b
s
eq
u
en
t
to
th
e
L
ST
M'
s
p
r
o
ce
s
s
in
g
o
f
t
h
e
s
eq
u
en
ce
s
.
T
h
is
lay
er
d
eter
m
i
n
es
i
f
th
e
v
id
eo
is
r
ea
l
o
r
p
h
o
n
y
b
y
m
ap
p
in
g
th
e
i
n
ter
n
a
l
L
ST
M
s
tate.
−
So
f
t
M
a
x
la
y
er
:
t
h
e
L
ST
M
o
u
tp
u
t
i
s
tr
an
s
f
o
r
m
ed
in
to
p
r
o
b
ab
ili
ties
f
o
r
ea
ch
clas
s
(
r
ea
l
v
s
.
f
a
k
e)
af
te
r
p
ass
in
g
t
h
r
o
u
g
h
a
So
f
t
Ma
x
la
y
er
w
it
h
t
h
e
f
in
al
o
u
tp
u
t.
T
h
e
m
o
d
el
p
r
ed
icts
th
e
cla
s
s
w
i
t
h
t
h
e
m
a
x
i
m
u
m
p
r
o
b
a
b
ilit
y
.
Fi
g
u
r
e
2
s
h
o
w
s
f
r
a
m
es
e
x
tr
ac
ted
f
r
o
m
v
id
eo
s
an
d
Fig
u
r
e
3
ex
p
lain
s
f
ea
t
u
r
e
e
x
tr
ac
tio
n
u
s
i
n
g
r
esn
e
x
t5
0
.
Fig
u
r
e
4
s
h
o
w
s
Se
q
u
en
ce
lear
n
i
n
g
an
d
v
id
eo
cl
ass
i
f
icatio
n
u
s
in
g
L
ST
M
lay
e
r
an
d
Fig
u
r
e
5
s
h
o
w
s
s
y
s
te
m
d
esig
n
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
S
yn
a
p
tic
s
h
ield
:
fu
s
io
n
o
f R
es
N
ex
t
–
5
0
a
n
d
lo
n
g
s
h
o
r
t
-
term me
mo
r
y
fo
r
en
h
a
n
a
ce
d
…
(
A
mit Mis
h
r
a
)
227
Fig
u
r
e
2
.
Fra
m
es e
x
tr
ac
ted
f
r
o
m
v
id
eo
s
Fig
u
r
e
3
.
Featu
r
e
ex
tr
ac
tio
n
u
s
in
g
r
e
s
n
e
x
t5
0
_
3
2
x
4
d
Fig
u
r
e
4
.
Seq
u
en
ce
lear
n
in
g
a
n
d
v
id
eo
class
i
f
icat
io
n
u
s
i
n
g
L
ST
M
lay
er
Fig
u
r
e
5
.
S
y
s
te
m
d
esi
g
n
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
3
.
2
.
Dev
el
o
p
m
ent
Dev
elo
p
m
e
n
t p
h
a
s
e
d
o
es in
s
ta
llatio
n
o
f
t
h
e
r
eq
u
ir
ed
lib
r
ar
ies an
d
P
y
t
h
o
n
3
as
f
o
llo
w
s
:
−
Fo
r
th
is
p
r
o
j
ec
t,
u
tili
ze
P
y
t
h
o
n
3
,
s
i
n
ce
it
h
a
s
a
lar
g
e
co
m
m
u
n
i
t
y
,
is
ea
s
y
to
u
s
e,
an
d
s
u
p
p
o
r
ts
a
ex
te
n
s
iv
e
r
an
g
e
o
f
m
ac
h
in
e
lear
n
i
n
g
(
M
L
)
lib
r
ar
ies.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
1
,
Ma
r
c
h
202
6
:
2
2
4
-
235
228
−
P
y
T
o
r
ch
f
r
a
m
e
w
o
r
k
:
P
y
T
o
r
ch
i
s
s
elec
ted
d
u
e
to
i
ts
ad
ap
t
ab
ilit
y
,
s
ca
lab
ilit
y
,
an
d
s
m
o
o
t
h
GP
U
s
u
p
p
o
r
t
w
it
h
C
UD
A
,
w
h
ich
e
n
ab
les q
u
ick
er
tr
ai
n
i
n
g
o
n
b
i
g
d
atasets
.
−
D
y
n
a
m
ic
co
m
p
u
tatio
n
g
r
ap
h
s
,
w
h
ich
P
y
T
o
r
ch
o
f
f
er
s
,
p
r
o
v
id
e
v
er
s
at
ilit
y
w
h
en
d
ev
el
o
p
in
g
m
o
d
els,
p
ar
ticu
lar
l
y
w
h
en
cr
ea
ti
n
g
b
es
p
o
k
e
la
y
er
s
(
s
u
ch
as
C
NN
m
i
x
ed
w
it
h
L
ST
M)
.
3
.
3
.
E
v
a
lua
t
i
o
n
3
.
3
.
1
.
Va
lid
a
t
io
n set
Fo
r
v
alid
atio
n
p
u
r
p
o
s
e
w
e
p
er
f
o
r
m
f
o
llo
w
in
g
s
tep
s
:
−
Go
al:
t
o
d
ev
elo
p
e
th
e
m
o
d
el
s
i
m
p
li
f
ies
s
u
cc
es
s
f
u
ll
y
to
n
e
w
,
u
n
s
ee
n
d
ata,
y
o
u
m
u
s
t
e
v
alu
a
te
it
o
n
a
v
alid
atio
n
s
et
o
n
ce
it
h
a
s
b
ee
n
tr
ain
ed
.
R
ea
l
an
d
f
ic
titi
o
u
s
v
i
d
eo
s
th
at
th
e
m
o
d
el
h
as
n
ev
er
v
ie
w
ed
b
ef
o
r
e
s
h
o
u
ld
b
e
in
clu
d
ed
in
t
h
e
v
a
li
d
atio
n
s
et.
−
B
alan
ce
d
d
ataset:
t
o
p
r
ev
en
t b
ias in
p
er
f
o
r
m
a
n
ce
o
u
tco
m
es,
m
ak
e
s
u
r
e
t
h
e
v
al
id
atio
n
s
et
c
o
n
tain
s
an
eq
u
a
l
a
m
o
u
n
t o
f
r
ea
l a
n
d
f
alse
v
id
eo
s
.
3
.
3
.
2
.
E
v
a
lua
t
io
n
m
et
rice
s
W
e
em
p
lo
y
a
n
u
m
b
er
o
f
m
etr
ics
to
esti
m
a
te
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el,
f
o
llo
w
i
n
g
ar
e
th
e
m
etr
ics
.
−
A
cc
u
r
ac
y
:
q
u
a
n
ti
f
ie
s
th
e
a
m
o
u
n
t
o
f
tr
u
e
o
r
f
alse
p
r
ed
ictio
n
s
th
e
m
o
d
el
co
r
r
ec
tly
p
r
ed
icts
.
I
t
is
th
e
p
r
o
p
o
r
tio
n
o
f
ac
cu
r
ate
f
o
r
ec
as
ts
to
all
f
o
r
ec
asts
.
−
P
r
ec
is
io
n
:
c
alcu
late
s
th
e
p
r
o
p
o
r
tio
n
o
f
"
f
ak
e"
v
id
eo
s
th
at
w
er
e
g
en
u
i
n
el
y
p
r
ed
icted
to
b
e
p
h
o
n
y
.
−
T
h
e
r
ec
all
(
s
en
s
iti
v
it
y
)
m
etr
ic
q
u
an
ti
f
ie
s
th
e
p
r
o
p
o
r
tio
n
o
f
r
ea
l "
f
ak
e"
f
il
m
s
th
at
w
er
e
p
r
o
p
er
l
y
d
etec
ted
.
−
T
h
e
F1
-
s
co
r
e
o
f
f
er
s
a
b
alan
ce
d
ass
ess
m
en
t
o
f
a
m
o
d
el's
p
er
f
o
r
m
a
n
ce
b
y
co
m
p
u
ti
n
g
t
h
e
h
a
r
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all.
3
.
3
.
3
.
Co
nfusi
o
n
m
a
t
ri
x
W
e
ca
n
ex
a
m
in
e
th
e
tr
u
e
p
o
s
itiv
e
s
(
T
P),
tr
u
e
n
eg
ati
v
es
(
T
N)
,
f
alse
p
o
s
itiv
e
s
(
FP
)
,
an
d
f
al
s
e
n
eg
at
iv
e
s
(
FN)
b
y
u
s
i
n
g
a
co
n
f
u
s
io
n
m
atr
i
x
.
W
e
ca
n
s
ee
h
o
w
t
h
e
m
o
d
el
is
d
o
i
n
g
f
o
r
ea
c
h
les
s
o
n
(
w
i
th
b
o
th
ac
tu
al
an
d
f
alse
v
id
eo
s
)
.
−
T
P
:
a
n
au
th
e
n
tic
v
id
eo
w
a
s
ac
cu
r
atel
y
p
r
ed
icted
b
y
t
h
e
m
o
d
el.
−
FN:
w
h
e
n
a
p
h
o
n
y
v
id
eo
w
as
m
is
tak
e
n
l
y
p
r
ed
icted
b
y
th
e
m
o
d
el
to
b
e
r
ea
l.
−
FP
:
w
h
en
a
r
ea
l v
id
eo
w
as
m
is
tak
en
l
y
p
r
ed
icted
b
y
t
h
e
m
o
d
e
l to
b
e
f
a
ls
e.
−
T
N:
a
n
ac
tu
al
v
id
eo
w
as a
cc
u
r
atel
y
p
r
ed
icted
b
y
t
h
e
m
o
d
el.
An
i
n
-
d
ep
th
an
a
l
y
s
is
o
f
FP
an
d
FN
w
as
co
n
d
u
cted
to
clas
s
if
y
co
m
m
o
n
is
s
u
e
s
.
FP
o
f
te
n
o
cc
u
r
r
ed
i
n
v
id
eo
s
w
it
h
s
u
d
d
en
li
g
h
tin
g
c
h
an
g
es
o
r
h
ig
h
l
y
d
y
n
a
m
ic
b
ac
k
g
r
o
u
n
d
s
,
w
h
ic
h
w
er
e
m
is
cl
ass
i
f
ied
d
u
e
to
th
e
ir
r
ese
m
b
lan
ce
to
d
ee
p
f
a
k
e
ar
ti
f
ac
ts
.
FN
w
er
e
m
o
r
e
p
r
ev
a
len
t
in
ca
s
es
w
it
h
m
i
n
i
m
al
m
o
ti
o
n
o
r
s
u
b
tle
f
ac
ial
ex
p
r
ess
io
n
s
,
as
th
e
te
m
p
o
r
al
in
co
n
s
is
te
n
cie
s
w
er
e
h
ar
d
er
f
o
r
th
e
L
ST
M
to
d
etec
t.
T
ab
l
e
1
s
h
o
w
s
d
ataset
ch
ar
ac
ter
is
tic
s
.
T
ab
le
1
.
Data
s
et
ch
ar
ac
ter
is
tic
s
D
a
t
a
se
t
n
a
me
T
o
t
a
l
v
i
d
e
o
s
R
e
a
l
v
i
d
e
o
s
F
a
k
e
v
i
d
e
o
s
R
e
so
l
u
t
i
o
n
P
r
i
mary
man
i
p
u
l
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
N
o
t
e
s
F
a
c
e
F
o
r
e
n
si
c
s++
2
,
0
0
0
1
,
0
0
0
1
,
0
0
0
7
2
0
p
D
e
e
p
f
a
k
e
,
F
a
c
e
S
w
a
p
W
i
d
e
l
y
u
se
d
i
n
d
e
t
e
c
t
i
o
n
b
e
n
c
h
mar
k
s
D
F
D
C
3
,
0
0
0
1
,
5
0
0
1
,
5
0
0
V
a
r
i
o
u
s
G
A
N
-
b
a
se
d
ma
n
i
p
u
l
a
t
i
o
n
s
I
n
c
l
u
d
e
s c
h
a
l
l
e
n
g
i
n
g
r
e
a
l
-
w
o
r
l
d
e
x
a
mp
l
e
s
C
e
l
e
b
-
DF
1
,
0
0
0
5
0
0
5
0
0
1
0
8
0
p
G
A
N
-
b
a
se
d
r
e
e
n
a
c
t
me
n
t
F
e
a
t
u
r
e
s h
i
g
h
-
q
u
a
l
i
t
y
f
a
k
e
s
4.
AL
G
O
RI
T
H
M
:
DE
E
P
F
AK
E
DE
T
E
CT
I
O
N
USI
NG
R
E
SNE
X
T
-
5
0
AND
L
S
T
M
4
.
1
.
P
re
pro
ce
s
s
ing
−
R
esize
ea
c
h
f
r
a
m
e
Xi
to
2
2
4
*
2
2
3
p
ix
els.
−
No
r
m
a
lize
th
e
p
i
x
el
v
al
u
e
s
to
r
an
g
e
[
0
,
1
]
:
′
=
255
(
1
)
4
.
2
.
Sp
a
t
ia
l
f
ea
t
ure
ex
t
ra
ct
i
o
n w
it
h Re
s
NeX
t
-
50
Fo
r
ea
ch
p
r
ep
r
o
ce
s
s
ed
f
r
am
e
′
:
−
P
ass
′
th
r
o
u
g
h
la
y
er
s
o
f
co
n
v
o
l
u
tio
n
al
w
i
th
g
r
o
u
p
ed
co
n
v
o
l
u
t
io
n
s
(
ca
r
d
in
alit
y
C
)
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
S
yn
a
p
tic
s
h
ield
:
fu
s
io
n
o
f R
es
N
ex
t
–
5
0
a
n
d
lo
n
g
s
h
o
r
t
-
term me
mo
r
y
fo
r
en
h
a
n
a
ce
d
…
(
A
mit Mis
h
r
a
)
229
=
(
∗
′
+
)
(
2
)
W
h
er
e,
W
k
an
d
b
k
ar
e
w
ei
g
h
t
s
an
d
b
iases
o
f
o
f
co
n
v
o
l
u
tio
n
al
k
er
n
el,
a
n
d
g
is
t
h
e
ac
ti
v
atio
n
f
u
n
ctio
n
(
R
e
L
U)
.
−
Ag
g
r
eg
ate
f
ea
tu
r
e
m
ap
s
f
r
o
m
m
u
ltip
le
p
ath
s
in
R
esNe
x
t:
=
∑
=
1
(
∗
′
+
)
(
3
)
−
Ou
tp
u
t a
2
0
4
8
-
d
im
e
n
s
io
n
al
f
e
atu
r
e
v
ec
to
r
FiF
_
iF
i f
o
r
ea
ch
f
r
a
m
e.
4
.
3
.
F
o
r
m
a
t
io
n
o
f
s
equence
C
o
m
b
i
n
e
th
e
f
ea
t
u
r
e
v
ec
to
r
s
{
F1
,
F2
,
…,
Fn
}
in
to
a
s
eq
u
e
n
c
e
F:
[
1
,
2
,
.
.
.
,
]
,
2048
(
4
)
4
.
4
.
T
e
m
po
ra
l
f
ea
t
ure
a
na
ly
s
is
u
s
ing
lo
ng
s
ho
rt
-
t
er
m
m
e
m
o
ry
Fo
r
th
e
s
eq
u
e
n
ce
F:
−
I
n
itializatio
n
o
f
L
ST
M
s
tates:
ℎ
=
0
,
=
0
(
5
)
−
Fo
r
ea
ch
ti
m
e
s
tep
t=1
,
2
,
…,
n
:
C
o
m
p
u
te
th
e
i
n
p
u
t
g
ate
(
6
)
.
=
(
∗
[
ℎ
−
1
,
]
+
)
(
6
)
C
o
m
p
u
te
th
e
f
o
r
g
et
g
ate
(
7
)
.
=
(
∗
[
ℎ
−
1
,
]
+
)
(
7
)
C
o
m
p
u
te
th
e
ce
ll
s
tate
(
8
)
.
t
=
t
⊙
t
-
1
+
t
⊙
ℎ
(
c
∗
[
ℎ
t
-
1
,
t
]
+
c
)
(
8
)
C
o
m
p
u
te
th
e
o
u
tp
u
t
g
ate
(
9
)
.
t
=
(
o
∗
[
ℎ
t
-
1
,
t
]
+
o
)
(
9
)
Up
d
ate
th
e
h
id
d
en
s
ta
te
(
1
0
)
.
ℎ
t
=
t
⊙
ℎ
(
t
)
(
1
0
)
4
.
5
.
Cla
s
s
if
ica
t
io
n
P
ass
th
e
f
in
al
h
id
d
en
s
ta
te
h
_
"
n
"
to
a
f
u
ll
y
co
n
n
ec
ted
la
y
er
(
1
1
)
.
=
∗
ℎ
+
(
1
1
)
A
p
p
l
y
t
h
e
So
f
t
M
a
x
f
u
n
c
tio
n
f
o
r
co
m
p
u
ti
n
g
clas
s
p
r
o
b
ab
ilit
ies
(
1
2
)
.
(
=
|
)
=
(
)
=
/
∑
=
1
,
=
1
,
2
,
…
,
(
1
2
)
Ass
i
g
n
th
e
cla
s
s
w
it
h
t
h
e
h
i
g
h
est p
r
io
r
ity
(
1
3
)
.
=
(
|
)
(
1
3
)
4
.
6
.
O
utput
R
etu
r
n
y
,
d
e
m
o
n
s
tr
atin
g
w
h
e
th
er
th
e
v
id
eo
is
r
ea
l
o
r
f
ak
e
.
T
h
is
alg
o
r
ith
m
co
m
b
in
e
s
s
p
atial
an
d
te
m
p
o
r
al
f
ea
tu
r
e
ex
tr
ac
tio
n
s
e
a
m
les
s
l
y
f
o
r
ac
cu
r
ate
d
etec
tio
n
.
T
a
b
le
2
s
h
o
w
s
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
v
s
ep
o
ch
s
.
Fi
g
u
r
e
6
s
h
o
w
s
m
o
d
el
ac
cu
r
ac
y
.
T
ab
le
3
s
h
o
w
s
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
an
d
Fig
u
r
e
7
s
h
o
ws
m
o
d
el
lo
s
s
.
T
ab
le
4
s
h
o
w
s
tr
ai
n
in
g
a
n
d
v
alid
atio
n
ac
c
u
r
ac
y
a
n
d
lo
s
s
v
s
ep
o
ch
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
1
,
Ma
r
c
h
202
6
:
2
2
4
-
235
230
T
ab
le
2
.
T
r
ain
in
g
an
d
v
alid
ati
o
n
ac
cu
r
ac
y
v
s
.
ep
o
ch
s
Ep
o
c
h
T
r
a
i
n
i
n
g
a
c
c
u
r
a
c
y
V
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
Ep
o
c
h
T
r
a
i
n
i
n
g
a
c
c
u
r
a
c
y
V
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
1
0
.
6
0
.
6
11
0
.
9
5
0
.
9
4
2
0
.
6
8
0
.
6
7
12
0
.
9
6
0
.
9
4
5
3
0
.
7
5
0
.
7
3
13
0
.
9
7
0
.
9
5
5
4
0
.
8
0
.
7
8
14
0
.
9
7
5
0
.
9
6
5
0
.
8
4
0
.
8
3
15
0
.
9
8
0
.
9
6
5
6
0
.
8
7
0
.
8
6
16
0
.
9
8
5
0
.
9
7
7
0
.
9
0
.
8
8
17
0
.
9
9
0
.
9
7
5
8
0
.
9
2
0
.
9
1
18
0
.
9
9
5
0
.
9
8
9
0
.
9
3
5
0
.
9
2
19
0
.
9
9
8
0
.
9
8
5
10
0
.
9
4
0
.
9
3
20
1
.
0
0
.
9
9
Fig
u
r
e
6
.
Mo
d
el
ac
cu
r
ac
y
T
ab
le
3
.
T
r
ain
in
g
an
d
v
alid
ati
o
n
lo
s
s
v
s
.
ep
o
ch
s
Ep
o
c
h
T
r
a
i
n
i
n
g
l
o
ss
V
a
l
i
d
a
t
i
o
n
l
o
ss
Ep
o
c
h
T
r
a
i
n
i
n
g
l
o
ss
V
a
l
i
d
a
t
i
o
n
l
o
ss
1
0
.
6
5
0
.
6
7
11
0
.
2
1
0
.
2
3
2
0
.
6
0
.
6
1
12
0
.
1
9
0
.
2
2
3
0
.
5
4
0
.
5
6
13
0
.
1
7
0
.
2
4
0
.
4
8
0
.
5
14
0
.
1
5
0
.
1
8
5
0
.
4
2
0
.
4
5
15
0
.
1
3
0
.
1
6
6
0
.
3
7
0
.
3
9
16
0
.
1
2
0
.
1
4
7
0
.
3
3
0
.
3
5
17
0
.
1
1
0
.
1
3
8
0
.
2
9
0
.
3
2
18
0
.
1
0
.
1
1
9
0
.
2
6
0
.
2
8
19
0
.
0
9
5
0
.
1
0
5
10
0
.
2
3
0
.
2
6
20
0
.
0
9
0
.
1
Fig
u
r
e
7
.
Mo
d
el
lo
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
S
yn
a
p
tic
s
h
ield
:
fu
s
io
n
o
f R
es
N
ex
t
–
5
0
a
n
d
lo
n
g
s
h
o
r
t
-
term me
mo
r
y
fo
r
en
h
a
n
a
ce
d
…
(
A
mit Mis
h
r
a
)
231
T
ab
le
4
.
T
r
ain
in
g
an
d
v
alid
ati
o
n
ac
cu
r
ac
y
a
n
d
lo
s
s
v
s
.
ep
o
ch
s
Ep
o
c
h
T
r
a
i
n
i
n
g
a
c
c
u
r
a
c
y
V
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
T
r
a
i
n
i
n
g
l
o
ss
V
a
l
i
d
a
t
i
o
n
l
o
ss
1
0
.
6
0
.
6
0
.
6
5
0
.
6
7
2
0
.
6
8
0
.
6
7
0
.
6
0
.
6
1
3
0
.
7
5
0
.
7
3
0
.
5
4
0
.
5
6
4
0
.
8
0
.
7
8
0
.
4
8
0
.
5
5
0
.
8
4
0
.
8
3
0
.
4
2
0
.
4
5
6
0
.
8
7
0
.
8
6
0
.
3
7
0
.
3
9
7
0
.
9
0
.
8
8
0
.
3
3
0
.
3
5
8
0
.
9
2
0
.
9
1
0
.
2
9
0
.
3
2
9
0
.
9
3
5
0
.
9
2
0
.
2
6
0
.
2
8
10
0
.
9
4
0
.
9
3
0
.
2
3
0
.
2
6
11
0
.
9
5
0
.
9
4
0
.
2
1
0
.
2
3
12
0
.
9
6
0
.
9
4
5
0
.
1
9
0
.
2
2
13
0
.
9
7
0
.
9
5
5
0
.
1
7
0
.
2
14
0
.
9
7
5
0
.
9
6
0
.
1
5
0
.
1
8
15
0
.
9
8
0
.
9
6
5
0
.
1
3
0
.
1
6
16
0
.
9
8
5
0
.
9
7
0
.
1
2
0
.
1
4
17
0
.
9
9
0
.
9
7
5
0
.
1
1
0
.
1
3
18
0
.
9
9
5
0
.
9
8
0
.
1
0
.
1
1
19
0
.
9
9
8
0
.
9
8
5
0
.
0
9
5
0
.
1
0
5
20
1
.
0
0
.
9
9
0
.
0
9
0
.
1
5.
CO
M
P
ARATI
VE
ANA
L
YS
I
S
5
.
1
.
Co
m
pa
ri
s
o
n
o
f
Sy
na
ptic
Sh
ield w
it
h
v
is
io
n
t
ra
ns
f
o
rm
er
s
a
nd
g
ra
ph
neura
l net
w
o
rk
s
T
h
e
Sy
n
ap
tic
Sh
ield
m
o
d
el
in
co
r
p
o
r
ates
R
esNeXt
–
5
0
f
o
r
th
e
ex
tr
ac
tio
n
o
f
s
p
atial
f
ea
tu
r
es
an
d
L
ST
M
f
o
r
t
h
e
an
a
l
y
s
is
o
f
ti
m
e
,
th
u
s
r
es
u
lti
n
g
i
n
h
ig
h
e
f
f
icie
n
c
y
in
v
id
eo
f
r
a
m
e
s
d
ee
p
f
ak
e
d
e
tectio
n
,
ca
p
tu
r
in
g
s
tatic
d
etails a
s
w
ell
as a
n
o
m
a
lies
b
ased
o
n
m
o
tio
n
.
T
h
is
h
a
s
g
ain
ed
s
o
m
u
c
h
i
n
ter
est
o
f
late
b
ec
au
s
e
o
f
it
s
n
ati
v
e
m
ec
h
an
is
m
o
f
atte
n
tio
n
f
o
r
an
i
m
a
g
e,
atten
tio
n
is
f
o
c
u
s
ed
o
n
o
t
h
er
p
ar
ts
th
at
ca
p
t
u
r
e
g
lo
b
al
d
ep
en
d
en
cies
in
s
tead
o
f
f
o
cu
s
i
n
g
o
n
lo
ca
l
f
ea
t
u
r
e
ex
tr
ac
tio
n
t
h
r
o
u
g
h
co
n
v
o
lu
t
io
n
al
la
y
er
s
t
h
at
is
p
la
y
ed
b
y
tr
a
d
itio
n
al
C
NN
s
lik
e
R
esNeX
t
–
5
0
.
T
h
e
s
tr
en
g
t
h
o
f
ViT
s
in
clu
d
es
g
o
o
d
ca
p
tu
r
in
g
lo
n
g
-
r
an
g
e
d
ep
en
d
en
cie
s
in
i
m
ag
e
s
,
an
d
it
ca
n
m
o
d
el
t
h
e
e
n
tire
i
m
ag
e
co
n
tex
t
m
o
r
e
h
o
li
s
ticall
y
co
m
p
ar
ed
to
C
NN
s
.
T
h
is
ac
t
u
all
y
h
elp
s
t
h
e
m
d
etec
t
s
u
b
tle
g
lo
b
al
i
n
co
n
s
is
ten
c
ies
i
n
i
m
a
g
es.
Gen
er
all
y
s
p
ea
k
i
n
g
,
ViT
s
r
eq
u
ir
e
m
u
c
h
m
o
r
e
d
ata
an
d
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
to
b
e
u
s
ed
f
o
r
h
ig
h
-
p
er
f
o
r
m
a
n
ce
tr
ain
i
n
g
.
T
h
ey
ar
e
n
o
t
n
ec
ess
ar
il
y
m
o
r
e
co
m
p
u
tatio
n
a
ll
y
e
f
f
ic
ien
t
i
n
p
r
o
ce
s
s
in
g
te
m
p
o
r
al
d
ata
th
an
C
NN
-
L
ST
M
m
o
d
els,
u
n
l
ess
t
h
e
y
ar
e
s
p
ec
if
ical
l
y
d
es
ig
n
ed
o
r
u
s
ed
to
g
et
h
er
w
it
h
R
N
Ns
to
p
r
o
ce
s
s
v
id
eo
s
eq
u
en
ce
s
.
T
ab
le
5
s
h
o
w
d
if
f
e
r
en
t CNN a
r
c
h
itect
u
r
es c
o
m
p
a
r
is
o
n
.
T
ab
le
5
.
Dif
f
er
en
t
C
NN
ar
ch
i
t
ec
tu
r
e
co
m
p
ar
is
o
n
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
n
a
me
P
r
e
c
i
si
o
n
[
P
]
R
e
c
a
l
l
[
R
]
A
U
C
[
A
U
C
]
A
c
c
u
r
a
c
y
[
A
]
F1
-
sco
r
e
[
F
1
]
V
6
1
9
0
.
9
1
0
.
9
7
0
.
9
8
7
0
.
9
4
0
.
9
4
V
G
G
F
A
C
E
0
.
9
9
0
.
9
8
0
.
9
9
8
0
.
9
9
0
.
9
9
D
EN
S
EN
ET
2
0
1
0
.
9
6
0
.
9
7
0
.
9
9
4
0
.
9
6
0
.
9
6
D
EN
S
EN
ET
1
2
1
0
.
9
9
0
.
7
0
0
.
9
7
1
0
.
9
7
0
.
8
2
6.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
NS
6
.
1
.
Ana
ly
s
is
o
f
m
o
del
perf
o
r
m
a
nce
o
v
er
co
m
pet
it
o
rs
T
h
e
co
u
p
lin
g
o
f
R
esNeXt
–
5
0
w
it
h
L
ST
M
en
s
u
r
e
s
th
e
ca
p
tu
r
e
o
f
f
in
e
-
g
r
ain
ed
s
p
atial
f
e
atu
r
es
an
d
te
m
p
o
r
al
in
co
n
s
i
s
te
n
cies
s
i
m
u
ltan
eo
u
s
l
y
.
R
e
s
NeX
t
–
5
0
ef
f
ici
en
tl
y
id
en
ti
f
ies
th
e
s
u
b
tle
ch
a
n
g
e
i
n
th
e
lig
h
ti
n
g
co
n
d
itio
n
s
,
s
k
i
n
to
n
e
tr
an
s
it
io
n
s
alo
n
g
w
it
h
f
ac
e
tex
t
u
r
es,
al
l
o
f
w
h
ich
ar
e
cr
itical
f
o
r
th
e
q
u
alit
y
o
f
d
ee
p
f
a
k
e
d
etec
tio
n
.
L
ST
M
tr
ac
k
s
th
e
ca
s
e
o
f
f
r
a
m
e
-
to
-
f
r
a
m
e
co
n
s
i
s
t
en
c
y
.
I
t
ca
p
tu
r
es
in
ap
p
r
o
p
r
iate
f
ac
ial
m
o
v
e
m
e
n
ts
,
in
co
n
s
is
te
n
t
ex
p
r
ess
io
n
s
,
an
d
m
is
m
atc
h
ed
o
n
e’
s
in
d
icati
v
e
o
f
m
an
ip
u
latio
n
s
.
T
h
is
d
u
al
an
al
y
s
i
s
g
iv
e
s
m
o
r
e
r
o
b
u
s
t d
etec
tio
n
ca
p
ab
ilit
y
t
h
a
n
a
C
NN
-
b
ased
m
o
d
el
t
h
at
is
o
n
l
y
s
p
atia
l f
ea
t
u
r
e
-
b
a
s
ed
.
Si
m
i
lar
to
m
o
s
t
o
th
er
d
ee
p
f
a
k
e
-
d
etec
t
io
n
m
o
d
el
s
,
it
o
f
te
n
lo
s
es
p
er
f
o
r
m
a
n
ce
o
n
ce
d
ea
lin
g
w
it
h
co
m
p
r
es
s
ed
o
r
o
f
lo
w
v
id
eo
q
u
alit
y
.
T
h
e
p
r
o
b
lem
is
o
v
er
co
m
e
b
y
t
h
e
S
y
n
ap
tic
S
h
ie
ld
s
in
ce
its
ad
v
a
n
ce
d
f
ea
t
u
r
e
ex
tr
ac
tio
n
an
d
s
eq
u
e
n
t
ial
an
al
y
s
is
e
n
s
u
r
e
th
e
p
er
f
o
r
m
an
ce
at
g
r
ea
t
ac
cu
r
ac
y
.
T
h
u
s
,
it
ca
n
b
e
ap
p
lied
f
o
r
r
ea
l
ap
p
licatio
n
s
co
n
ce
r
n
in
g
th
e
v
id
eo
q
u
alit
y
li
k
e
u
p
lo
ad
s
f
o
r
s
o
cial
m
ed
ia
o
r
s
u
r
v
eil
lan
ce
v
id
eo
s
.
Stan
d
ar
d
ev
alu
a
tio
n
d
ataset
s
o
f
th
e
Dee
p
f
a
k
e
d
etec
tio
n
c
h
all
en
g
e,
i
n
clu
d
i
n
g
DFD
C
an
d
C
e
leb
-
DF,
co
n
tai
n
i
n
g
d
ee
p
f
ak
es
o
f
d
if
f
er
en
t
lev
els
o
f
co
m
p
le
x
it
y
,
ar
e
u
s
ed
f
o
r
test
in
g
th
e
m
o
d
el.
T
h
e
ac
cu
r
ac
y
o
f
Sy
n
ap
tic
S
h
ield
r
e
m
ain
s
w
e
ll
o
v
er
9
5
%,
esp
e
ciall
y
th
r
o
u
g
h
th
e
ab
ilit
y
to
g
en
e
r
alize
o
v
er
th
e
v
ar
io
u
s
m
e
th
o
d
s
o
f
d
ee
p
f
ak
e
g
en
er
atio
n
.
He
n
ce
,
its
p
er
f
o
r
m
an
ce
d
if
f
er
s
f
r
o
m
th
e
o
v
er
-
s
p
ec
ialized
m
o
d
els
t
h
at
ca
n
o
n
l
y
o
v
er
f
it
a
s
p
ec
i
f
ic
t
y
p
e
o
f
d
ee
p
f
ak
e
a
n
d
n
o
t b
e
ef
f
ec
ti
v
e
w
h
e
n
f
ac
in
g
o
th
er
s
c
h
e
m
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
1
,
Ma
r
c
h
202
6
:
2
2
4
-
235
232
7.
F
UT
UR
E
SCO
P
E
T
h
e
f
u
t
u
r
e
s
co
p
e
o
f
r
esea
r
ch
in
d
ee
p
f
a
k
e
d
etec
tio
n
w
o
u
ld
b
e
to
ex
ten
d
th
e
ca
p
ab
ilit
ies
o
f
cu
r
r
en
t
m
o
d
el
s
i
n
h
a
n
d
li
n
g
al
l t
y
p
e
s
o
f
d
ee
p
f
a
k
es,
i
n
clu
d
i
n
g
au
d
io
-
v
is
u
al
o
n
es.
T
h
e
ap
p
r
o
ac
h
d
ev
elo
p
ed
h
er
e
is
b
ased
o
n
v
i
s
u
a
l i
n
co
n
s
is
te
n
cies,
w
h
e
r
ea
s
th
e
in
te
g
r
atio
n
o
f
a
u
d
io
an
al
y
s
i
s
i
n
to
th
e
s
y
s
te
m
w
o
u
ld
g
r
ea
tl
y
en
h
a
n
ce
t
h
e
s
ch
e
m
e
f
o
r
id
en
ti
f
y
in
g
m
a
n
i
p
u
lated
m
ater
ial.
T
h
is
m
a
y
i
n
clu
d
e
s
p
ee
ch
r
ec
o
g
n
itio
n
an
d
NL
P
m
et
h
o
d
s
to
d
etec
t
lip
m
o
tio
n
-
au
d
io
as
y
n
c
h
r
o
n
is
m
o
r
s
y
n
t
h
etic
v
o
ice
p
atter
n
s
.
F
u
t
u
r
e
m
o
d
els
w
ill
t
h
en
in
te
g
r
ate
m
u
lti
-
m
o
d
al
a
n
al
y
s
i
s
,
en
ab
li
n
g
e
f
f
e
ctiv
e
d
etec
tio
n
o
f
d
ee
p
f
a
k
es
w
h
ic
h
r
esu
l
t
in
m
a
n
ip
u
lat
io
n
o
f
v
id
eo
an
d
au
d
io
,
h
en
ce
g
en
er
all
y
i
n
cr
ea
s
i
n
g
t
h
e
ac
cu
r
ac
y
an
d
r
o
b
u
s
t
n
es
s
.
I
m
p
r
o
v
ed
ar
ch
itect
u
r
es
o
f
n
e
u
r
al
n
et
w
o
r
k
s
,
s
u
c
h
as
T
r
an
s
f
o
r
m
er
s
,
m
a
y
o
u
tp
er
f
o
r
m
tr
ad
itio
n
al
R
NN
s
in
ca
p
tu
r
i
n
g
m
o
r
e
s
u
b
tle
lo
n
g
-
r
an
g
e
d
ep
en
d
en
cie
s
an
d
s
lig
h
t
te
m
p
o
r
al
d
ep
en
d
en
cies
b
et
w
ee
n
f
r
a
m
es.
L
ig
h
t
w
ei
g
h
t
v
er
s
io
n
s
o
f
s
u
c
h
m
o
d
els
ca
n
b
e
d
er
iv
ed
f
o
r
ea
s
y
d
ep
lo
y
m
en
t
as
b
r
o
w
s
er
p
lu
g
i
n
s
o
r
m
o
b
ile
ap
p
licatio
n
s
an
d
t
h
r
o
u
g
h
e
m
b
ed
d
in
g
w
it
h
i
n
s
o
ci
al
m
ed
ia
p
lat
f
o
r
m
s
fo
r
r
ea
l
-
ti
m
e
d
ee
p
f
ak
e
d
etec
ti
o
n
at
t
h
e
u
s
er
le
v
el.
C
o
n
tin
u
o
u
s
lear
n
i
n
g
an
d
ad
j
u
s
t
in
g
to
n
e
w
d
ee
p
f
ak
e
s
w
il
l
also
b
e
in
ev
itab
le
o
v
er
ti
m
e
as
d
ee
p
f
ak
e
s
g
e
n
er
atio
n
te
ch
n
iq
u
es
i
m
p
r
o
v
e.
Fo
r
ex
a
m
p
le,
w
o
r
k
m
a
y
b
e
in
v
e
s
ti
g
ated
in
ad
v
er
s
ar
ial
tr
ain
in
g
an
d
r
ein
f
o
r
ce
m
e
n
t
lear
n
i
n
g
ap
p
licatio
n
s
i
n
d
esig
n
i
n
g
m
o
d
el
s
th
at
ar
e
n
o
t
o
n
l
y
r
esil
ien
t
to
c
u
r
r
en
t
t
h
r
ea
ts
b
u
t
also
ad
ap
tiv
e
w
it
h
r
esp
e
ct
to
b
r
ea
k
th
r
o
u
g
h
s
i
n
f
u
t
u
r
e
d
ee
p
f
ak
e
g
e
n
er
atio
n
tech
n
o
lo
g
ies.
8.
CO
NCLU
SI
O
N
W
ith
t
h
e
i
n
tr
o
d
u
ctio
n
o
f
t
h
is
r
o
b
u
s
t
D
L
f
r
a
m
e
w
o
r
k
,
t
h
e
ar
ticle
co
n
c
lu
d
es
b
y
p
r
o
p
o
s
in
g
a
n
in
v
e
s
ti
g
atio
n
b
ased
o
n
th
e
co
m
b
in
at
io
n
o
f
C
NNs
an
d
R
NN
s
f
o
r
h
ig
h
er
ac
cu
r
ac
y
i
n
d
ee
p
f
ak
e
v
id
eo
d
etec
tio
n
.
Ou
r
p
r
o
p
o
s
ed
s
y
s
te
m
ca
p
tu
r
e
s
s
u
b
tle
s
p
atial
a
n
d
te
m
p
o
r
al
in
co
n
s
is
te
n
cie
s
s
p
ec
i
f
ic
to
a
d
ee
p
f
ak
e
v
id
eo
b
y
co
m
b
i
n
i
n
g
t
h
e
u
s
e
o
f
th
e
R
e
s
Nex
t
C
NN
f
o
r
f
r
a
m
e
-
lev
el
f
e
atu
r
e
ex
tr
ac
tio
n
co
u
p
led
w
it
h
an
L
ST
M
n
et
w
o
r
k
f
o
r
th
e
a
n
al
y
s
is
o
f
th
e
te
m
p
o
r
al
in
s
ta
n
ce
s
.
T
h
is
is
co
n
f
ir
m
ed
b
y
w
id
e
test
i
n
g
o
f
o
u
r
m
et
h
o
d
o
n
v
ar
io
u
s
d
atasets
;
i
n
f
ac
t,
o
u
r
m
et
h
o
d
o
u
tp
er
f
o
r
m
ed
tr
ad
itio
n
al
d
et
ec
tio
n
m
o
d
els
a
n
d
ca
n
b
ec
o
m
e
a
r
eliab
le
an
d
s
ca
lab
le
s
o
l
u
tio
n
f
o
r
d
etec
ti
n
g
m
a
n
ip
u
lated
m
ed
ia.
T
h
is
w
o
r
k
d
ea
ls
w
ith
t
h
e
u
r
g
en
t n
ee
d
f
o
r
cr
ea
tin
g
e
f
f
ec
ti
v
e
d
ee
p
f
ak
e
-
d
etec
t
io
n
to
o
ls
th
a
t a
r
e
v
er
y
i
m
p
o
r
tan
t to
th
e
au
t
h
e
n
ticit
y
an
d
tr
u
s
t
w
o
r
t
h
in
e
s
s
o
f
d
ig
ital c
o
n
te
n
t.
F
UNDIN
G
I
NF
O
RM
AT
I
O
N
T
h
is
r
esear
ch
d
id
n
o
t
r
ec
eiv
e
an
y
s
p
ec
if
ic
g
r
an
t
f
r
o
m
f
u
n
d
in
g
ag
e
n
cie
s
in
t
h
e
p
u
b
lic,
co
m
m
er
cial,
o
r
n
o
t
-
f
o
r
-
p
r
o
f
it
s
ec
to
r
s
.
AUTHO
R
CO
NT
RIB
UT
I
O
NS ST
A
T
E
M
E
NT
T
h
is
jo
u
r
n
al
u
s
e
s
t
h
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
t
o
r
ec
o
g
n
ize
in
d
i
v
id
u
al
au
t
h
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
t
h
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
lla
b
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
Am
it M
i
s
h
r
a
✓
✓
✓
✓
✓
✓
P
r
a
j
w
a
l
C
h
i
n
c
h
m
alatp
u
r
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
Go
v
in
d
a
B
.
Sa
m
b
ar
e
✓
✓
✓
✓
✓
✓
Vio
m
es
h
K
u
m
ar
S
in
g
h
✓
✓
✓
✓
✓
✓
✓
A
t
u
l G
u
lab
r
ao
P
aw
ar
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
ah
u
l P
r
ak
as
h
Mir
aj
k
ar
✓
✓
✓
✓
✓
✓
✓
✓
P
r
iy
a
n
k
a
K.
T
ak
alk
ar
✓
✓
✓
✓
✓
✓
✓
✓
Ku
ld
ee
p
Va
y
ad
an
d
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
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
si
s
I
:
I
n
v
e
st
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
si
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
t
h
o
r
s
s
tate
n
o
co
n
f
lic
t o
f
i
n
t
er
est.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
S
yn
a
p
tic
s
h
ield
:
fu
s
io
n
o
f R
es
N
ex
t
–
5
0
a
n
d
lo
n
g
s
h
o
r
t
-
term me
mo
r
y
fo
r
en
h
a
n
a
ce
d
…
(
A
mit Mis
h
r
a
)
233
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
atasets
u
s
ed
in
th
is
s
t
u
d
y
ar
e
p
u
b
licl
y
av
ailab
le
an
d
ca
n
b
e
ac
ce
s
s
ed
f
r
o
m
t
h
eir
r
esp
ec
tiv
e
s
o
u
r
ce
s
.
RE
F
E
R
E
NC
E
S
[
1
]
D
.
L
i
u
,
Z
.
Y
a
n
g
,
R
.
Z
h
a
n
g
,
a
n
d
J
.
L
i
u
,
“
A
R
o
b
u
st
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
M
e
t
h
o
d
b
a
se
d
o
n
C
o
n
t
i
n
u
o
u
s
F
r
a
me
F
a
c
e
-
sw
a
p
p
i
n
g
,
”
i
n
2
0
2
2
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ss
i
n
g
a
n
d
C
l
o
u
d
C
o
m
p
u
t
i
n
g
(
AI
I
P
C
C
)
,
K
u
n
mi
n
g
,
C
h
i
n
a
:
I
EEE,
A
u
g
.
2
0
2
2
,
p
p
.
1
8
8
–
1
9
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
I
I
P
C
C
5
7
2
9
1
.
2
0
2
2
.
0
0
0
4
8
.
[
2
]
R
.
Z
h
a
n
g
,
Z
.
Ji
a
n
g
,
a
n
d
C
.
S
u
n
,
“
Tw
o
-
B
r
a
n
c
h
D
e
e
p
f
a
k
e
D
e
t
e
c
t
i
o
n
N
e
t
w
o
r
k
B
a
se
d
o
n
I
mp
r
o
v
e
d
X
c
e
p
t
i
o
n
,
”
i
n
2
0
2
3
I
EEE
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
E
l
e
c
t
ri
c
a
l
,
A
u
t
o
m
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r
En
g
i
n
e
e
ri
n
g
(
I
C
EA
C
E)
,
C
h
a
n
g
c
h
u
n
,
C
h
i
n
a
:
I
EEE,
D
e
c
.
2
0
2
3
,
p
p
.
2
2
7
–
231
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
E
A
C
E6
0
6
7
3
.
2
0
2
3
.
1
0
4
4
2
7
1
6
.
[
3
]
Á
.
F
.
G
a
mb
í
n
,
A
.
Y
a
z
i
d
i
,
A
.
V
a
si
l
a
k
o
s,
H
.
H
a
u
g
e
r
u
d
,
a
n
d
Y
.
D
j
e
n
o
u
r
i
,
“
D
e
e
p
f
a
k
e
s:
c
u
r
r
e
n
t
a
n
d
f
u
t
u
r
e
t
r
e
n
d
s,”
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
v
i
e
w
,
v
o
l
.
5
7
,
n
o
.
3
,
p
.
6
4
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
4
6
2
-
0
2
3
-
1
0
6
7
9
-
x.
[
4
]
E.
K
i
m
a
n
d
S
.
C
h
o
,
“
Ex
p
o
si
n
g
F
a
k
e
F
a
c
e
s
t
h
r
o
u
g
h
D
e
e
p
N
e
u
r
a
l
N
e
t
w
o
r
k
s
C
o
mb
i
n
i
n
g
C
o
n
t
e
n
t
a
n
d
T
r
a
c
e
F
e
a
t
u
r
e
Ex
t
r
a
c
t
o
r
s,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
9
,
p
p
.
1
2
3
4
9
3
–
1
2
3
5
0
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
1
1
0
8
5
9
.
[
5
]
Y
.
L
i
a
n
d
S
.
L
y
u
,
“
Ex
p
o
si
n
g
D
e
e
p
F
a
k
e
V
i
d
e
o
s
B
y
D
e
t
e
c
t
i
n
g
F
a
c
e
W
a
r
p
i
n
g
A
r
t
i
f
a
c
t
s,
”
i
n
Pr
o
c
.
I
EE
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
W
o
rk
sh
o
p
s (
C
VP
RW
)
,
L
o
n
g
B
e
a
c
h
,
C
A
,
U
S
A
,
2
0
1
9
,
p
p
.
4
6
–
5
2
.
[
6
]
S
.
T
h
a
se
e
n
I
k
r
a
m,
P
.
V
,
S
.
C
h
a
mb
i
a
l
,
D
.
S
o
o
d
,
a
n
d
A
.
V
,
“
A
P
e
r
f
o
r
man
c
e
E
n
h
a
n
c
e
me
n
t
o
f
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
t
h
r
o
u
g
h
t
h
e
u
se
o
f
a
H
y
b
r
i
d
C
N
N
D
e
e
p
L
e
a
r
n
i
n
g
M
o
d
e
l
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
E
l
e
c
t
ri
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
S
y
st
e
m
s
,
v
o
l
.
1
4
,
n
o
.
2
,
p
p
.
1
6
9
–
1
7
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
2
9
8
5
/
i
j
e
c
e
s.1
4
.
2
.
6
.
[
7
]
D
.
M
.
M
o
n
t
se
r
r
a
t
e
t
a
l
.
,
“
D
e
e
p
f
a
k
e
s
d
e
t
e
c
t
i
o
n
w
i
t
h
a
u
t
o
m
a
t
i
c
f
a
c
e
w
e
i
g
h
t
i
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
Wo
r
k
sh
o
p
s
,
S
e
a
t
t
l
e
,
W
A
,
U
S
A
,
2
0
2
0
,
p
p
.
2
8
5
1
–
2
8
5
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
W
5
0
4
9
8
.
2
0
2
0
.
0
0
3
4
2
.
[
8
]
D
.
W
o
d
a
j
o
a
n
d
S
.
A
t
n
a
f
u
,
“
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
U
s
i
n
g
C
o
n
v
o
l
u
t
i
o
n
a
l
V
i
si
o
n
T
r
a
n
sf
o
r
me
r
,
”
a
rX
i
v
p
r
e
p
ri
n
t
,
2
0
2
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
1
0
2
.
1
1
1
2
6
.
[
9
]
D
.
W
o
d
a
j
o
,
S
.
A
t
n
a
f
u
,
a
n
d
Z
.
A
k
h
t
a
r
,
“
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
U
si
n
g
G
e
n
e
r
a
t
i
v
e
C
o
n
v
o
l
u
t
i
o
n
a
l
V
i
si
o
n
T
r
a
n
sf
o
r
me
r
,
”
a
r
Xi
v
p
re
p
ri
n
t
,
2
0
2
3
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
3
0
7
.
0
7
0
3
6
.
[
1
0
]
Y
.
L
i
,
M
.
-
C
.
C
h
a
n
g
,
a
n
d
S
.
L
y
u
,
“
I
n
I
c
t
u
O
c
u
l
i
:
Ex
p
o
si
n
g
A
I
C
r
e
a
t
e
d
F
a
k
e
V
i
d
e
o
s
b
y
D
e
t
e
c
t
i
n
g
Ey
e
B
l
i
n
k
i
n
g
,
”
i
n
2
0
1
8
I
EE
E
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
rks
h
o
p
o
n
I
n
f
o
rm
a
t
i
o
n
Fo
r
e
n
s
i
c
s
a
n
d
S
e
c
u
r
i
t
y
(
WIFS
)
,
I
EEE,
D
e
c
.
2
0
1
8
,
p
p
.
1
–
7
,
d
o
i
:
1
0
.
1
1
0
9
/
W
I
F
S
.
2
0
1
8
.
8
6
3
0
7
8
7
.
[
1
1
]
L
.
L
i
e
t
a
l
.
,
“
F
a
c
e
X
-
r
a
y
f
o
r
mo
r
e
g
e
n
e
r
a
l
f
a
c
e
f
o
r
g
e
r
y
d
e
t
e
c
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
S
e
a
t
t
l
e
,
W
A
,
U
S
A
,
2
0
2
0
,
p
p
.
5
0
0
1
–
5
0
1
0
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
2
6
0
0
.
2
0
2
0
.
0
0
5
0
5
.
[
1
2
]
X
.
Y
a
n
g
,
Y
.
L
i
,
a
n
d
S
.
L
y
u
,
“
Ex
p
o
si
n
g
D
e
e
p
F
a
k
e
s
U
si
n
g
I
n
c
o
n
si
s
t
e
n
t
H
e
a
d
P
o
se
s,”
i
n
I
C
A
S
S
P
2
0
1
9
-
2
0
1
9
I
E
EE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ac
o
u
st
i
c
s
,
S
p
e
e
c
h
a
n
d
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
(
I
C
AS
S
P)
,
B
r
i
g
h
t
o
n
,
U
n
i
t
e
d
K
i
n
g
d
o
m,
2
0
1
9
,
p
p
.
8
2
6
1
–
8
2
6
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
S
S
P
.
2
0
1
9
.
8
6
8
3
1
6
4
.
[
1
3
]
D
.
A
f
c
h
a
r
,
V
.
N
o
z
i
c
k
,
J.
Y
a
m
a
g
i
sh
i
,
a
n
d
I
.
Ec
h
i
z
e
n
,
“
M
e
so
N
e
t
:
A
c
o
mp
a
c
t
f
a
c
i
a
l
v
i
d
e
o
f
o
r
g
e
r
y
d
e
t
e
c
t
i
o
n
n
e
t
w
o
r
k
,
”
i
n
2
0
1
8
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
Wo
r
k
s
h
o
p
o
n
I
n
f
o
r
m
a
t
i
o
n
F
o
re
n
s
i
c
s
a
n
d
S
e
c
u
r
i
t
y
(
WIFS
)
,
H
o
n
g
K
o
n
g
,
C
h
i
n
a
,
2
0
1
8
,
p
p
.
1
–
7
,
d
o
i
:
1
0
.
1
1
0
9
/
W
I
F
S
.
2
0
1
8
.
8
6
3
0
7
6
1
.
[
1
4
]
D
.
G
u
e
r
a
a
n
d
E
.
J
.
D
e
l
p
,
“
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
U
s
i
n
g
R
e
c
u
r
r
e
n
t
N
e
u
r
a
l
N
e
t
w
o
r
k
s,”
i
n
2
0
1
8
1
5
t
h
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
A
d
v
a
n
c
e
d
Vi
d
e
o
a
n
d
S
i
g
n
a
l
B
a
se
d
S
u
r
v
e
i
l
l
a
n
c
e
(
AV
S
S
)
,
A
u
c
k
l
a
n
d
,
N
e
w
Ze
a
l
a
n
d
,
2
0
1
8
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
A
V
S
S
.
2
0
1
8
.
8
6
3
9
1
6
3
.
[
1
5
]
S
.
B
e
n
Ja
b
e
u
r
,
H
.
B
a
l
l
o
u
k
,
W
.
B
e
n
A
r
f
i
,
a
n
d
J.
M
.
S
a
h
u
t
,
“
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
a
p
p
l
i
c
a
t
i
o
n
s
i
n
f
a
k
e
r
e
v
i
e
w
d
e
t
e
c
t
i
o
n
:
B
i
b
l
i
o
me
t
r
i
c
a
n
a
l
y
si
s
a
n
d
f
u
t
u
r
e
a
v
e
n
u
e
s
f
o
r
r
e
s
e
a
r
c
h
,
”
J
o
u
r
n
a
l
o
f
Bu
si
n
e
ss
R
e
se
a
rc
h
,
v
o
l
.
1
5
8
,
p
.
1
1
3
6
3
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
b
u
sr
e
s.
2
0
2
2
.
1
1
3
6
3
1
.
[
1
6
]
S
.
S
h
e
i
k
h
i
,
“
A
n
e
f
f
e
c
t
i
v
e
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
me
t
h
o
d
u
si
n
g
W
O
A
-
x
g
b
T
r
e
e
a
l
g
o
r
i
t
h
m
a
n
d
c
o
n
t
e
n
t
-
b
a
se
d
f
e
a
t
u
r
e
s,”
Ap
p
l
i
e
d
S
o
f
t
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
0
9
,
p
.
1
0
7
5
5
9
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
so
c
.
2
0
2
1
.
1
0
7
5
5
9
.
[
1
7
]
M
.
J.
S
o
n
a
w
a
n
e
,
S
.
S
.
M
u
n
d
h
e
,
K
.
S
.
G
a
i
k
w
a
d
,
S
.
M
.
S
a
t
h
e
,
a
n
d
Y
.
D
.
Ja
d
h
a
v
,
“
D
e
e
p
f
a
k
e
V
i
d
e
o
D
e
t
e
c
t
i
o
n
u
s
i
n
g
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Ad
v
a
n
c
e
d
Re
se
a
r
c
h
i
n
S
c
i
e
n
c
e
C
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
4
,
n
o
.
2,
p
p
.
2
0
–
2
4
,
2
0
2
4
,
d
o
i
:
1
0
.
4
8
1
7
5
/
i
j
a
r
sct
-
2
2
1
0
3
.
[
1
8
]
R
.
O
a
k
,
“
D
e
t
e
c
t
i
n
g
r
e
v
i
e
w
f
r
a
u
d
u
si
n
g
me
t
a
h
e
u
r
i
s
t
i
c
g
r
a
p
h
n
e
u
r
a
l
n
e
t
w
o
r
k
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
(
S
i
n
g
a
p
o
r
e
)
,
v
o
l
.
1
6
,
n
o
.
7
,
p
p
.
4
0
1
9
–
4
0
2
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
8
7
0
-
0
2
4
-
0
1
8
9
6
-
w.
[
1
9
]
A
.
I
q
b
a
l
,
M
.
A
.
R
a
u
f
,
M
.
Z
u
b
a
i
r
,
a
n
d
T
.
Y
o
u
n
i
s
,
“
A
n
Ef
f
i
c
i
e
n
t
E
n
se
mb
l
e
a
p
p
r
o
a
c
h
f
o
r
F
a
k
e
R
e
v
i
e
w
s
D
e
t
e
c
t
i
o
n
,
”
i
n
2
0
2
3
3
r
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
C
AI
)
,
I
s
l
a
mab
a
d
,
P
a
k
i
st
a
n
,
2
0
2
3
,
p
p
.
7
0
–
75
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
I
5
8
4
0
7
.
2
0
2
3
.
1
0
1
3
6
6
5
2
.
[
2
0
]
S
.
L
a
i
,
J.
W
u
,
C
.
Y
e
,
a
n
d
Z
.
M
a
,
“
U
C
F
-
P
K
S
:
U
n
f
o
r
e
see
n
C
o
n
s
u
me
r
F
r
a
u
d
D
e
t
e
c
t
i
o
n
w
i
t
h
P
r
i
o
r
K
n
o
w
l
e
d
g
e
a
n
d
S
e
ma
n
t
i
c
F
e
a
t
u
r
e
s,”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
o
c
i
a
l
S
y
st
e
m
s
,
v
o
l
.
1
1
,
n
o
.
4
,
p
p
.
5
4
5
4
–
5
4
6
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
S
.
2
0
2
4
.
3
3
7
2
5
1
9
.
[
2
1
]
I
.
D
.
M
i
e
n
y
e
a
n
d
N
.
Je
r
e
,
“
D
e
e
p
L
e
a
r
n
i
n
g
f
o
r
C
r
e
d
i
t
C
a
r
d
F
r
a
u
d
D
e
t
e
c
t
i
o
n
:
A
R
e
v
i
e
w
o
f
A
l
g
o
r
i
t
h
ms,
C
h
a
l
l
e
n
g
e
s
,
a
n
d
S
o
l
u
t
i
o
n
s
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
9
6
8
9
3
–
9
6
9
1
0
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
4
2
6
9
5
5
.
[
2
2
]
G
.
A
a
b
i
t
o
v
a
,
M
.
A
b
a
l
k
a
n
o
v
,
G
.
A
.
A
b
i
t
o
v
a
,
G
.
S
h
u
t
e
y
e
v
a
,
E.
A
i
t
m
u
k
h
a
n
b
e
t
o
v
a
,
a
n
d
K
.
K
u
l
n
i
y
a
z
o
v
a
,
“
R
e
v
i
e
w
o
f
C
l
o
u
d
A
I
f
o
r
R
e
a
l
-
T
i
me
F
r
a
u
d
D
e
t
e
c
t
i
o
n
,
”
i
n
2
0
2
4
1
0
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
A
u
t
o
m
a
t
i
o
n
,
R
o
b
o
t
i
c
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
(
I
C
ARA)
,
A
t
h
e
n
s
,
G
r
e
e
c
e
,
2
0
2
4
,
p
p
.
4
5
4
–
4
6
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
R
A
6
0
7
3
6
.
2
0
2
4
.
1
0
5
5
3
1
4
9
.
[
2
3
]
W
.
H
a
o
,
W
.
Z
h
a
n
g
,
a
n
d
H
.
J
i
n
,
“
S
A
G
E
-
N
e
t
:
Emp
l
o
y
i
n
g
S
p
a
t
i
a
l
A
t
t
e
n
t
i
o
n
a
n
d
G
e
o
m
e
t
r
i
c
En
c
o
d
i
n
g
f
o
r
P
o
i
n
t
C
l
o
u
d
B
a
se
d
P
l
a
c
e
R
e
c
o
g
n
i
t
i
o
n
,
”
I
E
EE
R
o
b
o
t
i
c
s
a
n
d
Au
t
o
m
a
t
i
o
n
L
e
t
t
e
rs
,
v
o
l
.
9
,
n
o
.
6
,
p
p
.
4
9
5
8
–
4
9
6
5
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
L
R
A
.
2
0
2
4
.
3
3
8
7
1
1
2
.
[
2
4
]
Z
.
W
u
a
n
d
I
.
S
a
v
i
d
i
s,
“
C
o
mp
a
r
a
t
i
v
e
A
n
a
l
y
si
s
o
f
G
r
a
p
h
I
so
mo
r
p
h
i
sm
a
n
d
G
r
a
p
h
N
e
u
r
a
l
N
e
t
w
o
r
k
s
f
o
r
A
n
a
l
o
g
H
i
e
r
a
r
c
h
y
L
a
b
e
l
i
n
g
,
”
i
n
2
0
2
4
2
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
S
y
m
p
o
s
i
u
m
o
n
Q
u
a
l
i
t
y
El
e
c
t
r
o
n
i
c
D
e
si
g
n
(
I
S
Q
ED
)
,
S
a
n
F
r
a
n
c
i
sco
,
C
A
,
U
S
A
,
2
0
2
4
,
p
p
.
1
–
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
S
Q
ED
6
0
7
0
6
.
2
0
2
4
.
1
0
5
2
8
7
4
9
.
[
2
5
]
S
.
T
a
n
g
,
L
.
Ji
n
,
a
n
d
F
.
C
h
e
n
g
,
“
F
r
a
u
d
D
e
t
e
c
t
i
o
n
i
n
O
n
l
i
n
e
P
r
o
d
u
c
t
R
e
v
i
e
w
S
y
st
e
ms v
i
a
H
e
t
e
r
o
g
e
n
e
o
u
s G
r
a
p
h
T
r
a
n
sf
o
r
me
r
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
9
,
p
p
.
1
6
7
3
6
4
–
1
6
7
3
7
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
0
8
4
9
2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.