I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
6
7
2
~
3
6
8
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
6
7
2
-
3
6
8
2
3672
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
India
n
sig
n lang
ua
g
e video
genera
tion usin
g
atten
tion
-
enha
nced
g
enerativ
e adv
ersa
ria
l net
wo
rk
P
ra
chi P
ra
m
o
d Wa
g
hm
a
re
,
Ash
wi
ni M
a
ng
es
h De
s
hp
a
nd
e
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
,
M
a
h
a
r
s
h
i
K
a
r
v
e
S
t
r
e
e
S
h
i
k
sh
a
n
S
a
ms
t
h
a
’
s
C
u
m
mi
n
s
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
f
o
r
W
o
me
n
a
f
f
i
l
i
a
t
e
d
t
o
S
a
v
i
t
r
i
b
a
i
P
h
u
l
e
P
u
n
e
U
n
i
v
e
r
si
t
y
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
,
2
0
2
5
R
ev
is
ed
Ma
y
2
0
,
2
0
2
6
Acc
ep
ted
J
u
l 9
,
2
0
2
6
S
ig
n
lan
g
u
a
g
e
(S
L)
is
t
h
e
p
rim
a
ry
m
o
d
e
o
f
c
o
m
m
u
n
ica
ti
o
n
f
o
r
th
e
D
e
a
f
sig
n
e
rs.
De
sp
it
e
a
d
v
a
n
c
e
m
e
n
ts
in
d
e
e
p
lea
rn
in
g
,
S
L
re
c
o
g
n
i
ti
o
n
,
tr
a
n
sla
ti
o
n
,
a
n
d
v
id
e
o
g
e
n
e
ra
ti
o
n
fa
c
e
c
h
a
ll
e
n
g
e
s
li
k
e
b
lu
rri
n
e
ss
a
n
d
in
c
o
n
siste
n
c
ies
.
Th
is
re
se
a
rc
h
p
ro
p
o
se
s
a
n
o
v
e
l
si
g
n
lan
g
u
a
g
e
tran
sla
ti
o
n
(S
L
T)
a
p
p
ro
a
c
h
f
o
r
In
d
ian
sig
n
lan
g
u
a
g
e
(IS
L)
u
sin
g
a
n
a
tt
e
n
ti
o
n
-
d
ri
v
e
n
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
(G
AN
).
Th
e
p
re
p
r
o
c
e
ss
in
g
p
i
p
e
li
n
e
in
c
l
u
d
e
s v
id
e
o
fra
m
e
e
x
trac
ti
o
n
,
sk
e
leta
l
jo
in
t
c
o
o
rd
in
a
te
d
e
tec
ti
o
n
v
ia
Op
e
n
P
o
se
,
a
n
d
d
y
n
a
m
ic
ti
m
e
wa
rp
in
g
(DTW
)
fo
r
p
o
se
d
a
ta
re
fin
e
m
e
n
t.
Th
e
s
q
u
e
e
z
e
a
n
d
e
x
c
it
a
ti
o
n
(S
E)
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
e
n
h
a
n
c
e
s
2
D
c
o
n
v
o
lu
ti
o
n
a
l
lay
e
rs,
a
ll
o
win
g
t
h
e
g
e
n
e
ra
to
r
to
fo
c
u
s
o
n
re
lev
a
n
t
s
k
e
leta
l
p
o
se
se
q
u
e
n
c
e
s.
A
m
o
ti
o
n
d
isc
rimin
a
to
r
re
fin
e
s
m
o
ti
o
n
a
u
th
e
n
t
icity
.
P
e
rf
o
rm
a
n
c
e
e
v
a
lu
a
ti
o
n
u
si
n
g
tw
o
SL
d
a
tas
e
ts
d
e
m
o
n
stra
tes
sig
n
ifi
c
a
n
t
imp
r
o
v
e
m
e
n
ts
in
str
u
c
tu
ra
l
sim
il
a
rit
y
in
d
e
x
m
e
a
su
re
(
S
S
IM
)
,
p
e
a
k
sig
n
a
l
-
to
-
n
o
ise
ra
ti
o
(
P
S
NR
)
,
a
n
d
tem
p
o
ra
l
c
o
n
siste
n
c
y
m
e
tri
c
(
TCM
)
sc
o
re
s,
a
c
h
iev
in
g
9
9
.
6
0
(%
)
a
s S
S
I
M
,
3
1
.
1
0
d
B
a
s
P
S
NR,
a
n
d
0
.
9
1
1
1
a
s T
CM
.
Th
e
p
ro
p
o
se
d
m
o
d
e
l
o
u
tp
e
rfo
rm
s sta
n
d
a
rd
G
AN
a
n
d
d
y
n
a
m
ic
G
AN
in
S
L
v
id
e
o
g
e
n
e
ra
ti
o
n
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
Gen
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
I
n
d
ian
s
ig
n
la
n
g
u
a
g
e
Sig
n
lan
g
u
a
g
e
tr
an
s
latio
n
Sq
u
ee
ze
an
d
e
x
citatio
n
atten
tio
n
m
ec
h
an
is
m
Vid
eo
g
en
er
atio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Pra
ch
i Pr
am
o
d
W
ag
h
m
a
r
e
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
T
elec
o
m
m
u
n
icatio
n
Ma
h
ar
s
h
i K
ar
v
e
Stre
e
Sh
ik
s
h
a
n
Sam
s
th
a’
s
C
u
m
m
in
s
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
f
o
r
W
o
m
en
af
f
iliated
to
Sav
itrib
ai
Ph
u
le
P
u
n
e
Un
iv
er
s
ity
Kar
v
en
ag
ar
,
Pu
n
e
–
4
1
1
0
5
2
,
I
n
d
ia
E
m
ail: p
r
ac
h
i.w
ag
h
m
ar
e@
cu
m
m
in
s
co
lleg
e.
in
1.
I
NT
RO
D
UCT
I
O
N
C
o
m
m
u
n
icatio
n
is
f
u
n
d
am
en
tal
to
h
u
m
an
in
ter
ac
tio
n
,
p
r
i
m
ar
ily
f
ac
ilit
ated
t
h
r
o
u
g
h
s
p
ee
ch
an
d
g
estu
r
es.
Sig
n
lan
g
u
a
g
e
(
SL)
is
a
n
atu
r
al
v
is
u
al
lan
g
u
ag
e
th
at
r
elies
o
n
h
an
d
m
o
v
em
en
ts
,
f
ac
ial
ex
p
r
ess
io
n
s
,
an
d
b
o
d
y
lan
g
u
ag
e
to
co
n
v
ey
m
ea
n
in
g
.
I
t
is
th
e
p
r
i
m
ar
y
m
o
d
e
o
f
c
o
m
m
u
n
icatio
n
f
o
r
th
e
Dea
f
c
o
m
m
u
n
ity
,
th
o
u
g
h
s
o
m
e
h
ea
r
in
g
i
n
d
iv
id
u
als,
in
clu
d
in
g
in
ter
p
r
ete
r
s
an
d
f
am
ily
m
em
b
er
s
,
also
u
s
e
it.
E
ac
h
co
u
n
tr
y
h
as
its
o
wn
SL,
s
u
ch
as
I
n
d
ian
s
ig
n
lan
g
u
ag
e
(
I
SL)
,
wh
ich
h
as
a
d
is
tin
ct
g
r
am
m
ar
an
d
s
tr
u
c
tu
r
e.
Ad
d
itio
n
ally
,
r
eg
io
n
al
SL
s
lik
e
Ma
r
ath
i
SL
ex
is
t w
ith
in
I
n
d
ia,
ea
ch
ad
ap
te
d
to
lo
ca
l lin
g
u
is
tic
an
d
cu
ltu
r
al
co
n
tex
ts
.
T
r
ad
itio
n
al
s
ig
n
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
(
SLP)
tech
n
iq
u
es
s
tr
u
g
g
le
to
g
en
e
r
ate
n
at
u
r
al,
c
o
h
er
en
t
s
ig
n
s
eq
u
en
ce
s
.
Ma
n
y
d
ee
p
lear
n
in
g
-
b
ased
ap
p
r
o
ac
h
es
f
o
cu
s
p
r
im
ar
ily
o
n
m
an
u
al
g
estu
r
e
s
wh
ile
n
eg
lectin
g
non
-
m
a
n
u
al
f
ea
t
u
r
es
s
u
ch
as
f
ac
ial
ex
p
r
ess
io
n
s
an
d
b
o
d
y
p
o
s
tu
r
e,
lead
in
g
to
u
n
n
atu
r
al
-
lo
o
k
in
g
s
ig
n
av
atar
s
.
R
ec
en
t
ad
v
an
ce
m
en
ts
in
SL
P
h
av
e
lev
er
ag
e
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es
f
o
r
r
ec
o
g
n
itio
n
,
g
en
er
atio
n
,
an
d
tr
an
s
latio
n
.
Dh
an
jal
an
d
Sin
g
h
[
1
]
d
e
v
elo
p
e
d
an
au
to
m
atic
s
y
s
tem
to
co
n
v
er
t
Pu
n
jab
i
tex
t
in
to
I
SL
u
s
in
g
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P)
to
g
en
er
ate
I
SL
g
lo
s
s
es
an
d
co
r
r
esp
o
n
d
in
g
s
ig
n
r
ep
r
esen
tatio
n
s
.
T
o
en
h
an
ce
c
o
n
tin
u
o
u
s
SL
r
ec
o
g
n
itio
n
,
Pu
et
a
l
.
[
2
]
in
t
r
o
d
u
ce
d
a
cr
o
s
s
-
m
o
d
ality
au
g
m
en
tatio
n
f
r
a
m
ewo
r
k
in
teg
r
atin
g
m
u
ltim
o
d
al
lear
n
i
n
g
f
o
r
i
m
p
r
o
v
ed
ac
cu
r
ac
y
.
Si
m
ilar
ly
,
Hu
et
a
l
.
[
3
]
p
r
o
p
o
s
ed
th
e
g
lo
b
al
-
l
o
ca
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
I
n
d
ia
n
s
ig
n
la
n
g
u
a
g
e
vid
eo
g
e
n
era
tio
n
u
s
in
g
a
tten
tio
n
-
e
n
h
a
n
ce
d
g
e
n
era
tive
…
(
P
r
a
ch
i P
r
a
mo
d
Wa
g
h
ma
r
e)
3673
en
h
an
ce
m
e
n
t
n
etwo
r
k
(
GL
E
N
)
,
wh
ich
em
p
lo
y
s
non
-
n
eg
ativ
e
m
atr
ix
f
ac
to
r
izatio
n
(
NM
F)
-
awa
r
e
lear
n
in
g
to
ex
tr
ac
t
r
o
b
u
s
t
f
ea
tu
r
es,
en
h
a
n
cin
g
r
ec
o
g
n
itio
n
p
er
f
o
r
m
a
n
ce
.
Ad
d
r
ess
in
g
co
n
tin
u
o
u
s
3
D
SL
p
r
o
d
u
ctio
n
,
Sau
n
d
er
s
et
a
l
.
[
4
]
d
esig
n
ed
a
p
r
o
g
r
ess
iv
e
tr
an
s
f
o
r
m
e
r
-
b
ase
d
m
o
d
el
in
co
r
p
o
r
atin
g
m
ix
tu
r
e
d
en
s
ity
n
etwo
r
k
s
f
o
r
m
o
r
e
n
atu
r
al
s
ig
n
s
y
n
th
esi
s
.
B
r
ag
g
et
a
l
.
[
5
]
p
r
o
v
id
e
d
a
n
in
ter
d
is
cip
lin
ar
y
p
e
r
s
p
ec
tiv
e
o
n
SL
r
ec
o
g
n
itio
n
,
g
en
e
r
a
tio
n
,
an
d
tr
an
s
latio
n
,
h
ig
h
lig
h
tin
g
ch
al
len
g
es,
d
atasets
,
an
d
f
u
t
u
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
Sto
ll
et
a
l
.
[
6
]
in
t
r
o
d
u
ce
d
‘
S
i
g
n
S
y
n
t
h
’
,
a
d
a
ta
-
d
r
i
v
e
n
a
p
p
r
o
a
c
h
f
o
r
g
e
n
e
r
a
t
i
n
g
SL
v
i
d
e
o
s
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
f
r
o
m
t
e
x
t
u
a
l
a
n
d
g
l
o
s
s
i
n
p
u
ts
.
B
e
y
o
n
d
SL
,
g
e
n
e
r
a
t
i
v
e
m
o
d
e
l
s
h
a
v
e
b
e
e
n
i
n
s
t
r
u
m
e
n
t
a
l
i
n
i
m
a
g
e
a
n
d
v
i
d
e
o
s
y
n
t
h
e
s
is
.
X
u
e
t
a
l
.
[
7
]
p
r
o
p
o
s
e
d
ad
v
er
s
ar
ially
ap
p
r
o
x
im
ated
a
u
to
en
co
d
er
f
o
r
r
ea
lis
tic
im
ag
e
m
an
ip
u
latio
n
,
wh
ile
C
h
en
an
d
Gu
o
[
8
]
r
e
v
iewe
d
au
to
en
co
d
er
s
in
d
ee
p
lear
n
in
g
an
d
th
eir
a
p
p
licatio
n
s
in
r
e
p
r
e
s
en
tatio
n
lear
n
in
g
a
n
d
g
en
er
at
iv
e
m
o
d
elin
g
.
T
ex
t
-
to
-
im
ag
e
an
d
v
id
e
o
tr
a
n
s
f
o
r
m
atio
n
tech
n
iq
u
es
h
a
v
e
also
in
f
lu
e
n
ce
d
SL
r
esear
ch
.
J
u
n
g
et
a
l
.
[
9
]
d
ev
elo
p
e
d
a
lan
g
u
ag
e
-
a
g
n
o
s
ti
c
m
o
d
el
en
s
u
r
in
g
s
em
an
tic
co
n
s
is
ten
cy
in
g
en
er
ated
im
ag
es.
W
u
et
a
l
.
[
1
0
]
in
tr
o
d
u
ce
d
a
s
ec
u
r
e
g
e
n
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
(
Sec
GAN
)
,
a
cy
cle
-
co
n
s
is
ten
t
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
(
GAN)
f
o
r
s
ec
u
r
ely
r
ec
o
v
e
r
ab
le
v
id
eo
m
o
d
if
ica
tio
n
.
Me
n
et
a
l
.
[
1
1
]
p
r
o
p
o
s
ed
a
GAN
-
b
ased
m
o
d
el
f
o
r
c
o
n
tr
o
llab
le
p
e
r
s
o
n
im
ag
e
s
y
n
th
esis
,
wh
ile
Ng
u
y
en
et
a
l
.
[
1
2
]
ap
p
lied
GAN
-
b
ased
f
u
tu
r
e
f
r
a
m
e
p
r
ed
ictio
n
f
o
r
an
o
m
aly
d
etec
t
io
n
in
s
u
r
v
eillan
ce
v
id
eo
s
.
K
wo
n
an
d
Par
k
[
1
3
]
r
e
f
in
ed
t
h
is
ap
p
r
o
ac
h
u
s
in
g
r
etr
o
s
p
ec
tiv
e
c
y
cle
GANs
f
o
r
im
p
r
o
v
e
d
tem
p
o
r
al
c
o
n
s
is
ten
cy
,
an
d
W
an
g
et
a
l
.
[
1
4
]
in
tr
o
d
u
ce
d
‘
I
m
ag
in
at
o
r
’
,
a
s
p
atio
-
tem
p
o
r
al
GAN
f
o
r
r
ea
lis
tic
v
id
eo
g
e
n
er
atio
n
.
T
h
ese
ad
v
an
ce
m
e
n
ts
co
llectiv
el
y
d
em
o
n
s
tr
ate
th
e
im
p
ac
t
o
f
d
ee
p
lear
n
in
g
in
SL
P
an
d
v
id
eo
s
y
n
th
esis
.
T
h
e
ab
ilit
y
o
f
GANs
to
cr
ea
te
h
ig
h
-
q
u
ality
s
y
n
th
etic
im
ag
es a
n
d
v
id
e
o
s
m
ak
es th
e
m
a
p
r
o
m
is
in
g
s
o
lu
tio
n
f
o
r
SL
v
id
eo
g
en
er
atio
n
.
T
h
is
r
esear
ch
p
r
o
p
o
s
es
an
atten
tio
n
d
r
iv
en
GAN
-
b
a
s
ed
m
o
d
el
to
g
en
er
ate
h
i
g
h
-
q
u
ality
,
p
h
o
to
-
r
ea
lis
tic
SL
v
id
eo
s
f
r
o
m
s
k
eleta
l
p
o
s
itio
n
s
.
B
y
in
co
r
p
o
r
atin
g
lin
g
u
is
tic
co
n
tex
t
an
d
ex
p
r
ess
iv
e
f
ea
tu
r
es,
th
e
p
r
o
p
o
s
ed
m
o
d
el
aim
s
to
o
v
er
co
m
e
th
e
lim
itatio
n
s
o
f
tr
ad
itio
n
al
GANs.
T
h
e
ap
p
r
o
ac
h
en
s
u
r
es
th
e
g
en
er
atio
n
o
f
n
atu
r
al
an
d
co
n
tex
t
-
awa
r
e
s
ig
n
s
eq
u
en
ce
s
,
en
h
an
cin
g
ac
ce
s
s
ib
ilit
y
an
d
co
m
m
u
n
icatio
n
f
o
r
th
e
I
SL
s
ig
n
er
s
an
d
n
o
n
-
s
ig
n
in
g
in
d
iv
id
u
als.
T
h
e
ad
v
a
n
ce
m
e
n
ts
in
th
e
GAN
-
b
ased
SL
v
i
d
eo
g
e
n
er
atio
n
ca
n
s
ig
n
if
ican
tly
im
p
r
o
v
e
r
ea
l
-
tim
e
s
ig
n
lan
g
u
a
g
e
tr
an
s
latio
n
(
S
L
T
)
s
y
s
tem
s
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
R
ec
en
t
ad
v
an
ce
m
en
ts
in
SL
p
r
o
d
u
ctio
n
an
d
v
id
eo
s
y
n
t
h
esis
h
av
e
lev
er
ag
ed
n
e
u
r
al
n
et
wo
r
k
s
an
d
GANs
to
im
p
r
o
v
e
r
ea
lis
m
an
d
ac
cu
r
ac
y
.
Dh
an
jal
an
d
Sin
g
h
[
1
5
]
p
r
o
p
o
s
ed
a
p
r
o
ject
th
a
t
aim
s
to
cr
ea
te
an
au
to
m
atic
s
y
s
tem
ca
lled
‘
SIS
L
A’
th
at
ca
n
co
n
v
er
t
s
p
ee
ch
to
I
SL
au
to
m
atica
lly
.
T
h
r
ee
s
tep
s
co
m
p
r
is
e
th
e
s
y
s
tem
'
s
o
p
er
atio
n
:
tr
an
s
latio
n
o
f
th
e
s
o
u
r
ce
lan
g
u
a
g
e
in
to
I
SL
;
v
o
ice
r
ec
o
g
n
itio
n
(
s
p
ee
ch
r
ec
o
g
n
itio
n
(
SR
)
)
f
o
r
is
o
lated
wo
r
d
s
in
E
n
g
lis
h
,
Hin
d
i,
an
d
Pu
n
jab
i
;
an
d
a
Ham
No
Sy
s
-
b
ased
3
D
av
atar
d
ep
ictin
g
I
SL
m
o
tio
n
s
.
T
h
e
s
y
s
tem
's f
o
u
r
m
ain
im
p
lem
en
tatio
n
m
o
d
u
les co
m
p
r
is
e
th
e
r
eq
u
ir
em
e
n
t a
n
aly
s
is
,
d
ata
co
llectio
n
,
tech
n
ical
d
ev
elo
p
m
e
n
t,
an
d
ass
ess
m
en
t
m
o
d
u
les.
T
h
e
s
y
s
tem
is
m
o
r
e
ef
f
ec
tiv
e
b
ec
au
s
e
it
s
u
p
p
o
r
ts
m
u
ltip
le
lan
g
u
ag
es.
Sp
ee
ch
s
am
p
le
f
iles
f
o
r
E
n
g
lis
h
,
Hin
d
i,
an
d
Pu
n
jab
i
wer
e
tr
a
in
ed
an
d
test
ed
.
Acc
o
r
d
in
g
t
o
em
p
ir
ical
f
in
d
in
g
s
,
th
e
p
r
o
p
o
s
ed
m
o
d
els
h
av
e
m
in
im
u
m
ac
cu
r
ac
y
lev
els
f
o
r
E
n
g
lis
h
,
Pu
n
jab
i,
an
d
Hin
d
i
o
f
9
1
%,
8
9
%,
an
d
8
9
%,
r
esp
ec
tiv
ely
.
Usab
ilit
y
test
s
v
er
if
ied
th
e
s
y
s
tem
'
s
u
s
ef
u
ln
ess
in
h
elp
in
g
h
ea
r
in
g
-
im
p
air
ed
p
e
o
p
le
co
m
m
u
n
icate
an
d
lear
n
.
Ho
wev
er
,
th
is
s
y
s
tem
lack
s
t
h
e
in
co
r
p
o
r
atio
n
o
f
n
o
n
-
m
an
u
al
SL
ch
ar
ac
te
r
is
tics
an
d
s
en
t
en
ce
-
lev
el
tr
an
s
latio
n
.
Sto
ll
et
a
l
.
[
1
6
]
in
tr
o
d
u
ce
d
T
ex
t2
Sig
n
,
a
n
eu
r
al
m
ac
h
in
e
tr
an
s
latio
n
an
d
GAN
-
b
ased
f
r
a
m
ewo
r
k
f
o
r
SL
p
r
o
d
u
ctio
n
,
b
r
id
g
in
g
t
h
e
g
a
p
b
etwe
en
tex
t
u
al
in
p
u
t
an
d
s
ig
n
v
id
eo
s
.
Me
h
ta
et
a
l
.
[
1
7
]
f
o
cu
s
ed
o
n
au
to
m
ated
3
D
SL
ca
p
ti
o
n
g
en
er
atio
n
,
en
h
a
n
cin
g
ac
c
ess
ib
ilit
y
f
o
r
th
e
h
ea
r
in
g
-
i
m
p
air
ed
.
Similar
ly
,
Nata
r
ajan
an
d
R
ajasek
ar
[
1
8
]
p
r
o
p
o
s
ed
a
d
y
n
am
ic
GAN
m
o
d
el
f
o
r
g
en
er
atin
g
h
i
g
h
-
q
u
ali
ty
SL
v
id
e
o
s
f
r
o
m
s
k
eleta
l
p
o
s
es,
im
p
r
o
v
in
g
m
o
t
io
n
r
ea
lis
m
.
Z
elin
k
a
an
d
Ka
n
is
[
1
9
]
ex
p
l
o
r
ed
n
e
u
r
al
SL
s
y
n
t
h
esis
,
em
p
h
asizin
g
g
lo
s
s
-
b
ased
r
ep
r
esen
tatio
n
lea
r
n
in
g
f
o
r
s
ig
n
v
id
eo
g
en
e
r
atio
n
.
I
n
th
e
b
r
o
ad
er
d
o
m
ai
n
o
f
v
i
d
eo
g
e
n
er
atio
n
,
th
e
m
o
tio
n
a
n
d
co
n
ten
t
d
ec
o
m
p
o
s
ed
g
en
e
r
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
(
Mo
C
o
GAN
)
b
y
T
u
ly
ak
o
v
et
a
l.
[
2
0
]
in
tr
o
d
u
ce
d
a
m
o
tio
n
-
c
o
n
ten
t
d
ec
o
m
p
o
s
itio
n
f
r
am
ewo
r
k
,
ef
f
ec
tiv
ely
s
ep
ar
atin
g
m
o
tio
n
an
d
co
n
ten
t
in
v
i
d
eo
s
y
n
th
esis
.
Aig
n
er
an
d
Kö
r
n
er
[
2
1
]
d
e
v
elo
p
ed
Fu
tu
r
eGA
N,
lev
er
ag
in
g
s
p
atio
-
tem
p
o
r
al
3
D
co
n
v
o
lu
tio
n
s
to
p
r
ed
ict
f
u
t
u
r
e
f
r
a
m
es
in
v
id
eo
s
eq
u
en
ce
s
.
Siar
o
h
in
et
a
l.
[
2
2
]
ad
v
a
n
ce
d
p
o
s
e
-
b
ased
h
u
m
a
n
im
ag
e
g
e
n
er
atio
n
th
r
o
u
g
h
d
ef
o
r
m
ab
le
GANs,
a
ch
iev
in
g
n
at
u
r
al
m
o
tio
n
s
y
n
t
h
esis
.
I
n
f
o
GAN
b
y
C
h
en
et
a
l.
[
2
3
]
en
h
a
n
ce
d
in
ter
p
r
etab
ilit
y
in
r
ep
r
esen
tati
o
n
lear
n
in
g
b
y
m
ax
im
izin
g
m
u
tu
al
in
f
o
r
m
atio
n
,
b
en
e
f
itin
g
s
tr
u
ctu
r
ed
v
id
eo
g
en
er
atio
n
.
Z
h
a
n
g
et
a
l
.
[
2
4
]
p
r
o
p
o
s
ed
Stack
GAN,
a
two
-
s
tag
e
ar
ch
itectu
r
e
t
h
at
g
en
e
r
ates
p
h
o
to
r
ea
lis
tic
im
ag
es
f
r
o
m
tex
tu
al
d
escr
ip
ti
o
n
s
,
in
f
lu
e
n
cin
g
tex
t
-
to
-
s
ig
n
s
y
n
th
esis
.
T
h
ese
s
tu
d
ies
d
em
o
n
s
tr
ate
th
e
p
o
ten
tial
o
f
d
ee
p
lear
n
in
g
tech
n
iq
u
es
in
im
p
r
o
v
in
g
SL
T
,
v
id
eo
s
y
n
t
h
esis
,
an
d
m
o
tio
n
r
ep
r
esen
tat
io
n
.
B
y
c
o
m
b
in
i
n
g
n
eu
r
al
m
ac
h
in
e
tr
an
s
latio
n
,
p
o
s
e
esti
m
atio
n
,
an
d
GAN
-
b
as
ed
m
o
d
els,
r
esear
ch
e
r
s
co
n
ti
n
u
e
to
ad
v
an
ce
SL
ac
ce
s
s
ib
ilit
y
an
d
r
ea
lis
tic
m
o
tio
n
s
y
n
th
esis
.
T
h
e
f
o
llo
win
g
ar
e
th
e
m
aj
o
r
co
n
tr
ib
u
tio
n
s
f
o
r
th
is
f
r
am
ew
o
r
k
.
First,
t
h
e
atte
n
tio
n
-
en
h
an
ce
d
GAN
f
r
am
ewo
r
k
f
o
r
I
SL
v
id
e
o
g
en
er
atio
n
.
Seco
n
d
,
i
n
co
r
p
o
r
atin
g
s
q
u
ee
ze
an
d
e
x
citatio
n
(
SE)
atten
tio
n
b
lo
ck
s
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
7
2
-
3
6
8
2
3674
en
h
an
ce
s
ig
n
er
r
e
p
r
esen
tatio
n
an
d
p
o
s
e
-
awa
r
e
f
ea
tu
r
e
lear
n
in
g
an
d
f
in
e
-
t
u
n
in
g
o
f
s
ig
n
p
o
s
e
f
ea
tu
r
es.
T
h
ir
d
,
SL
v
id
eo
g
en
er
ated
with
en
h
an
ce
d
o
u
tp
u
t
q
u
ality
b
ased
o
n
SE
-
GAN
f
r
am
ewo
r
k
.
T
h
e
c
r
ea
ted
/g
en
er
ated
SL
v
id
eo
s
'
v
is
u
al
r
ea
li
s
m
,
m
o
tio
n
co
n
tin
u
ity
,
a
n
d
tem
p
o
r
al
co
n
s
is
ten
cy
clo
s
ely
m
atch
es
w
ith
co
m
p
u
tatio
n
all
y
d
em
an
d
in
g
tr
an
s
f
o
r
m
er
/d
if
f
u
s
io
n
m
o
d
els f
o
r
r
e
al
-
tim
e
I
SL
p
r
o
d
u
ctio
n
with
lig
h
tweig
h
t G
A
N
f
r
am
ewo
r
k
s
.
3.
M
E
T
H
O
D
SL
is
ess
en
tial
f
o
r
co
m
m
u
n
ic
atio
n
am
o
n
g
Dea
f
s
ig
n
er
s
,
y
e
t
its
in
ter
p
r
etatio
n
r
em
ain
s
ch
allen
g
in
g
.
SLT
f
ac
es
d
if
f
icu
lties
d
u
e
t
o
alig
n
m
e
n
t
lear
n
in
g
,
n
o
n
-
m
an
u
al
f
ea
tu
r
es,
an
d
a
lack
o
f
ex
p
e
r
tis
e
in
I
SL.
T
r
ad
itio
n
al
GANs
s
tr
u
g
g
le
w
ith
u
n
d
er
s
tan
d
in
g
lin
g
u
is
tic
co
n
tex
t,
lea
d
in
g
to
in
ef
f
ec
tiv
e
v
id
eo
g
e
n
er
atio
n
.
T
h
is
wo
r
k
aim
s
to
g
en
er
ate
r
ea
lis
tic
I
SL
v
id
eo
s
b
y
in
co
r
p
o
r
atin
g
b
o
th
m
an
u
al
an
d
n
o
n
-
m
an
u
al
ex
p
r
ess
io
n
s
.
T
h
e
p
r
ep
r
o
ce
s
s
in
g
p
h
ase
in
v
o
l
v
es
ex
tr
ac
tin
g
s
k
eleta
l
jo
in
t
co
o
r
d
in
ates
u
s
in
g
Op
e
n
Po
s
e
an
d
r
ef
in
in
g
p
o
s
e
d
ata
v
ia
d
y
n
am
ic
tim
e
war
p
i
n
g
(
D
T
W
)
.
T
h
ese
s
k
eleta
l f
ea
tu
r
es s
er
v
e
as in
p
u
t f
o
r
an
atten
tio
n
-
d
r
iv
en
GAN.
Un
lik
e
tr
ad
itio
n
al
GANs,
th
is
m
o
d
el
in
teg
r
ates
an
SE
atten
tio
n
m
ec
h
an
is
m
f
o
r
2
D
co
n
v
o
lu
tio
n
al
lay
er
s
,
en
h
an
cin
g
v
id
eo
r
ea
lis
m
.
T
h
e
s
y
s
tem
e
m
p
lo
y
s
a
m
o
tio
n
d
is
cr
im
in
at
o
r
to
d
if
f
er
en
tiate
b
etwe
en
r
e
al
an
d
f
ak
e
m
o
tio
n
s
am
p
les,
wh
ile
a
m
o
tio
n
g
en
e
r
ato
r
cr
ea
tes
life
lik
e
m
o
v
em
e
n
ts
.
T
h
e
atten
tio
n
m
ec
h
an
is
m
g
u
id
es
th
e
n
etwo
r
k
,
en
s
u
r
in
g
ac
cu
r
ate
s
ig
n
r
ep
r
ese
n
tatio
n
an
d
im
p
r
o
v
e
d
SL
v
id
e
o
tr
an
s
latio
n
,
as
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
is
s
ec
tio
n
d
escr
ib
e
th
e
s
tep
s
o
f
th
e
a
p
p
r
o
ac
h
f
o
llo
wed
t
o
co
n
v
er
t I
SL
v
id
eo
s
in
to
f
r
a
m
es.
Fig
u
r
e
1
.
Pro
p
o
s
ed
s
y
s
tem
f
o
r
I
SL
v
id
eo
g
en
er
atio
n
3
.
1
.
Video
t
o
f
ra
m
e
c
o
nv
er
s
io
n
R
GB
f
r
am
es
ar
e
cr
ea
ted
f
r
o
m
ea
ch
I
SL
in
p
u
t
v
id
eo
.
T
h
e
v
id
eo
s
wer
e
o
r
ig
in
ally
r
ec
o
r
d
ed
at
2
5
f
r
am
es
p
e
r
s
ec
o
n
d
,
s
o
ea
ch
v
id
eo
y
ield
s
a
d
if
f
er
en
t
n
u
m
b
er
o
f
f
r
a
m
es
d
ep
e
n
d
in
g
o
n
its
d
u
r
atio
n
.
Af
ter
th
e
ca
p
tu
r
ed
v
i
d
eo
h
as
b
ee
n
c
o
n
v
er
ted
in
to
a
s
eq
u
en
ce
o
f
f
r
am
es,
an
im
ag
e
p
r
ep
ar
atio
n
s
tep
(
cr
o
p
p
i
n
g
,
s
ca
lin
g
,
an
d
n
o
r
m
alizin
g
)
to
r
em
o
v
e
n
o
is
e
d
u
r
in
g
d
ata
co
llectio
n
o
n
ce
th
e
ca
p
t
u
r
ed
v
id
e
o
h
as
b
ee
n
co
n
v
er
ted
in
to
a
s
er
ies o
f
f
r
am
ed
p
ictu
r
es f
o
r
v
id
eo
p
r
o
ce
s
s
in
g
.
3
.
1
.
1
.
F
ra
m
e
t
o
s
k
elet
a
l p
o
s
e
A
p
o
p
u
lar
co
m
p
u
ter
v
is
io
n
li
b
r
ar
y
ca
lled
Op
en
Po
s
e
o
f
f
er
s
s
k
eleta
l
p
o
s
e
esti
m
at
io
n
f
r
o
m
im
ag
es
o
r
v
id
eo
s
as
well
a
s
r
ea
l
-
tim
e
m
u
lti
-
p
er
s
o
n
k
ey
p
o
in
t
r
ec
o
g
n
iti
o
n
.
T
h
e
m
o
d
el
ca
n
esti
m
ate
t
h
e
2
D
p
o
s
itio
n
s
o
f
im
p
o
r
tan
t
b
o
d
y
jo
in
ts
,
o
f
te
n
ca
lled
k
ey
p
o
in
ts
,
lik
e
th
e
h
ea
d
,
s
h
o
u
ld
e
r
s
,
elb
o
ws,
wr
is
ts
,
h
ip
s
,
k
n
ee
s
,
an
d
an
k
les.
Fu
r
th
er
m
o
r
e
,
wh
en
g
i
v
en
a
d
ep
th
m
ap
,
Op
e
n
Po
s
e
ca
n
g
en
er
ate
3
D
p
o
s
e
p
r
ed
ict
io
n
s
.
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
ar
e
u
s
ed
in
co
m
b
in
atio
n
with
a
d
ee
p
lear
n
in
g
tech
n
i
q
u
es
-
b
ased
m
o
d
el
to
s
im
u
ltan
eo
u
s
ly
d
etec
t
b
o
d
y
p
ar
ts
an
d
in
f
e
r
th
eir
s
p
atial
co
r
r
elatio
n
s
.
Op
en
Po
s
e
is
f
le
x
ib
le
an
d
h
as
b
ee
n
im
p
lem
en
ted
i
n
d
if
f
er
en
t
s
ec
to
r
s
,
in
clu
d
i
n
g
h
u
m
a
n
-
co
m
p
u
ter
in
ter
ac
tio
n
,
a
u
g
m
en
te
d
r
ea
lity
,
an
d
h
u
m
an
b
eh
av
io
r
an
al
y
s
is
.
I
ts
av
ailab
il
ity
an
d
o
p
en
-
s
o
u
r
ce
s
tatu
s
h
a
v
e
aid
ed
in
its
b
r
o
a
d
ac
ce
p
tan
ce
an
d
in
co
r
p
o
r
atio
n
in
to
a
m
u
ltit
u
d
e
o
f
k
e
y
p
o
i
n
t d
etec
tio
n
an
d
s
k
eleta
l p
o
s
e
esti
m
atio
n
ap
p
licatio
n
s
.
I
n
t
h
i
s
w
o
r
k
,
O
p
en
P
o
s
e
i
s
u
s
e
d
t
o
ex
t
r
ac
t
s
k
e
l
e
t
a
l
j
o
in
t
c
o
o
r
d
in
a
t
e
s
f
r
o
m
s
i
g
n
v
id
e
o
s
.
T
h
e
l
o
w
e
r
-
b
o
d
y
jo
i
n
t
s
ar
e
n
o
t
t
a
k
e
n
i
n
to
c
o
n
s
i
d
er
a
t
i
o
n
a
s
t
h
ey
p
l
a
y
n
o
r
o
l
e
in
SL
,
a
n
d
u
s
e
1
0
j
o
i
n
t
s
d
e
s
c
r
ib
i
n
g
t
h
e
u
p
p
er
b
o
d
y
(
h
e
ad
,
n
e
c
k
,
s
h
o
u
ld
e
r
s
,
e
l
b
o
w
s
,
w
r
i
s
t
s
,
an
d
h
ip
s
)
.
E
ac
h
jo
i
n
t
i
s
d
e
f
in
ed
a
s
a
p
ix
e
l
i
n
th
e
i
m
a
g
e
w
i
th
c
o
o
r
d
i
n
a
t
e
s
x
an
d
y
.
T
o
u
t
i
l
i
ze
m
u
l
t
ip
l
e
d
at
a
s
e
t
s
f
r
o
m
d
i
f
f
e
r
e
n
t
d
o
m
a
i
n
s
,
w
e
p
e
r
f
o
r
m
th
e
f
o
l
l
o
w
in
g
n
o
r
m
a
l
i
z
a
t
io
n
.
E
v
er
y
s
k
eleto
n
is
p
o
s
itio
n
ed
at
th
e
n
ec
k
j
o
in
t in
r
elatio
n
to
a
s
elec
ted
r
ef
er
en
ce
s
k
eleto
n
as in
(
1
)
.
S
k
e
l
e
tal
S
e
q
u
e
n
c
e
G
e
n
e
r
ati
on
u
s
i
n
g
O
p
e
n
p
os
e
F
ak
e
I
n
p
u
t V
i
d
e
os
Te
mp
or
al
A
l
i
gn
me
n
t u
s
i
n
g
D
TW
F
r
ame
s
P
r
e
-
p
r
oc
e
s
s
i
n
g
G
e
n
e
r
ator
S
E A
tte
n
ti
on
M
e
c
h
an
i
s
m
R
e
al
D
i
s
c
r
i
mi
n
ator
O
u
tp
u
t S
i
gn
V
i
d
e
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
I
n
d
ia
n
s
ig
n
la
n
g
u
a
g
e
vid
eo
g
e
n
era
tio
n
u
s
in
g
a
tten
tio
n
-
e
n
h
a
n
ce
d
g
e
n
era
tive
…
(
P
r
a
ch
i P
r
a
mo
d
Wa
g
h
ma
r
e)
3675
=
(
,
)
+
(
(
,
)
−
(
,
)
)
(
1
)
W
h
er
e
th
e
r
ef
er
en
ce
an
d
i
n
p
u
t
s
k
eleto
n
s
'
n
ec
k
j
o
in
ts
ar
e
d
en
o
ted
b
y
an
d
,
r
esp
ec
tiv
ely
.
Af
ter
alig
n
m
en
t,
is
th
e
tr
an
s
lated
i
n
p
u
t
s
k
eleto
n
.
Nex
t,
th
e
s
h
o
u
ld
er
-
to
-
s
h
o
u
ld
er
d
is
tan
ce
s
o
f
t
h
e
in
p
u
t
an
d
r
ef
er
en
ce
s
k
eleto
n
s
ar
e
u
s
ed
to
co
m
p
u
te
a
s
ca
lin
g
f
ac
to
r
as in
(
2
)
.
=
(
(
,
)
−
(
,
)
)
(
(
,
)
−
(
,
)
)
(
2
)
W
h
er
e
an
d
r
ep
r
esen
t
th
e
le
f
t
an
d
r
ig
h
t
s
h
o
u
l
d
er
j
o
in
ts
o
f
th
e
in
p
u
t
s
k
eleto
n
,
r
esp
ec
tiv
ely
,
an
d
an
d
r
ep
r
esen
t
th
e
r
ef
er
en
ce
s
k
eleto
n
s
.
Usi
n
g
th
e
tr
an
s
lated
s
k
eleto
n
,
,
th
at
was
p
r
ev
io
u
s
l
y
ac
q
u
ir
ed
a
n
d
th
e
s
ca
lin
g
f
ac
to
r
,
,
f
in
ally
co
m
p
u
te
th
e
n
o
r
m
alize
d
s
k
eleto
n
,
as in
(
3
)
.
=
(
,
)
+
(
(
,
)
−
(
,
)
)
∗
(
3
)
Fu
r
th
er
,
th
e
f
r
a
m
e
an
n
o
tatio
n
s
ar
e
u
s
ed
to
ca
teg
o
r
ize
all
s
k
eleta
l
s
eq
u
en
ce
s
b
y
f
r
am
es
to
cr
ea
te
th
e
f
r
am
e
to
p
o
s
e
s
eq
u
en
ce
.
Nex
t,
u
s
e
DT
W
is
u
s
ed
to
alig
n
e
v
er
y
s
k
el
etal
s
eq
u
en
ce
f
o
r
ev
er
y
f
r
am
e
,
an
d
th
en
co
m
b
in
e
th
em
in
to
a
s
in
g
le
r
ep
r
esen
tati
v
e
m
ea
n
s
k
eleto
n
s
eq
u
en
ce
as
in
(
4
)
.
=
1
∑
(
,
=
1
0
)
(
4
)
W
h
er
e
is
th
e
n
u
m
b
er
o
f
f
r
am
e
s
am
p
le
s
eq
u
en
ce
s
,
an
d
is
f
r
a
m
e'
s
r
ep
r
esen
tativ
e
m
ea
n
s
eq
u
en
ce
.
3
.
2
.
Dy
na
m
ic
t
i
m
e
wra
pp
ing
Dif
f
er
en
tiatin
g
th
e
lo
ca
l
tem
p
o
r
al
s
eq
u
en
ce
s
o
f
d
if
f
er
en
t
ac
tio
n
ca
teg
o
r
ies
is
a
ch
allen
g
in
g
p
r
o
ce
s
s
d
u
e
to
s
ev
er
al
f
ac
to
r
s
,
s
u
ch
as
f
r
am
e
r
ate
f
lu
ctu
atio
n
s
an
d
te
m
p
o
r
al
in
d
e
p
en
d
e
n
ce
in
ea
ch
s
u
b
lev
el,
r
eg
ar
d
less
o
f
wh
eth
er
s
ev
er
al
lev
els
o
f
s
h
ap
e
-
b
ased
alig
n
m
en
t/m
atch
i
n
g
(
SHs
)
ar
e
u
s
ed
.
T
o
s
o
lv
e
th
ese
p
r
o
b
lem
s
,
th
e
d
is
tan
ce
b
etwe
en
th
e
v
ar
io
u
s
m
u
ltip
le/m
u
ltid
im
en
s
io
n
al
DT
W
(
MSH/
m
u
ltip
le
s
eq
u
en
ce
alig
n
m
en
t
)
lev
els
f
o
r
ev
er
y
ac
tio
n
ca
teg
o
r
y
is
ca
lc
u
lated
u
s
in
g
DT
W
.
T
h
e
tim
e
s
er
ies
a
lig
n
m
en
t
alg
o
r
ith
m
i
s
ca
lled
DT
W
.
B
y
r
ep
ea
ted
ly
war
p
in
g
th
e
tim
e
a
x
is
,
it c
alcu
lates a
p
ath
b
etwe
en
two
s
eq
u
en
ce
s
u
n
til id
ea
l m
atch
is
id
en
tifie
d
.
Ass
u
m
e
th
at
we
h
av
e
two
tim
e
s
er
ies,
A
an
d
B
,
with
r
esp
e
ctiv
e
len
g
th
s
o
f
n
a
n
d
m
.
T
h
e
two
tim
e
s
er
ies ar
e
d
ef
in
ed
as
(
5
)
.
=
1
,
2
,
…
,
,
…
,
=
1
,
2
,
…
,
,
…
,
(
5
)
U
s
i
n
g
D
T
W
,
n
-
by
-
m
m
a
t
r
i
x
w
a
s
c
r
e
a
t
e
d
,
a
n
d
i
t
s
(
ℎ
,
ℎ
)
e
l
e
m
e
n
t
h
o
l
d
s
d
i
s
t
a
n
c
e
(
,
)
=
(
,
)
=
‖
−
‖
2
b
etwe
en
th
e
two
p
o
in
ts
,
an
d
.
ev
er
y
m
at
r
ix
elem
en
t
(
,
)
r
ep
r
esen
ts
h
o
w
th
e
two
p
o
in
ts
ar
e
ali
g
n
ed
.
A
co
n
tig
u
o
u
s
s
et
o
f
m
atr
ix
co
m
p
o
n
en
ts
th
at
d
ef
in
es
a
m
ap
p
i
n
g
f
r
o
m
to
is
ca
lled
a
war
p
i
n
g
p
ath
.
W
ith
(
,
)
≤
<
+
−
1
,
we
o
b
tain
=
1
,
2
,
.
.
.
,
,
.
.
.
,
as
th
e
ℎ
elem
en
t
o
f
,
d
ef
in
ed
as
=
(
,
)
.
T
h
e
DT
W
d
is
tan
ce
is
ca
lcu
lated
as
(
6
)
.
(
,
)
=
{
√
∑
=
1
}
(
6
)
L
ik
ewise,
n
o
m
in
al
d
is
tan
ce
s
a
m
o
n
g
MSHs
o
f
s
u
cc
ess
iv
e
le
v
els
f
o
r
ev
er
y
lo
ca
l
jo
in
t
ar
e
d
eter
m
in
ed
b
y
DT
W
.
Nex
t,
ea
ch
l
o
ca
l
j
o
in
t
d
is
p
lace
m
en
t'
s
d
is
tan
ce
v
ec
to
r
is
cr
ea
te
d
.
Fo
r
ev
er
y
ac
tio
n
ca
teg
o
r
y
,
th
e
tem
p
o
r
al
m
o
d
el
o
f
th
e
s
k
eleto
n
jo
in
ts
is
en
co
d
ed
as
a
co
n
ca
ten
atio
n
o
f
th
e
d
is
tan
ce
v
ec
to
r
=
[
1
,
.
.
.
,
]
.
A
s
k
eleto
n
r
ep
r
esen
tatio
n
f
ea
tu
r
e
v
ec
t
o
r
is
cr
ea
ted
b
y
co
m
b
in
in
g
th
e
s
k
eleto
n
jo
in
ts
'
p
o
s
tu
r
e
a
n
d
m
o
tio
n
f
ea
tu
r
e
v
ec
to
r
s
,
wh
ich
ar
e
co
m
p
u
ted
f
o
r
ea
ch
ac
tio
n
s
eq
u
e
n
ce
.
T
h
e
s
k
eleto
n
r
ep
r
esen
tati
o
n
f
ea
tu
r
e
v
ec
to
r
is
d
ef
in
ed
as
(
7
)
.
=
+
(
7
)
W
h
er
e
an
d
ar
e
th
e
weig
h
tin
g
p
ar
am
eter
s
.
I
n
f
o
r
m
atio
n
r
eg
ar
d
in
g
t
h
e
s
p
a
tio
tem
p
o
r
al
f
u
n
ctio
n
o
v
er
a
n
ac
tio
n
'
s
tim
e
s
eq
u
en
ce
ca
n
b
e
f
o
u
n
d
i
n
th
e
s
tatic
p
o
s
e
an
d
in
th
e
tem
p
o
r
al
d
y
n
am
ics
o
f
th
e
h
ar
m
o
n
ic
s
.
T
h
is
h
ar
m
o
n
ic
in
f
o
r
m
atio
n
ca
n
b
e
v
iewe
d
as
a
co
m
p
ac
t
r
ep
r
esen
tatio
n
o
f
th
e
s
k
eleta
l
jo
in
ts
in
th
e
b
o
d
y
.
As
a
r
esu
lt,
s
u
ch
h
ar
m
o
n
ic
in
f
o
r
m
atio
n
en
ab
les
ac
cu
r
ate
class
if
icatio
n
o
f
ac
tiv
ities
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
7
2
-
3
6
8
2
3676
3
.
3
.
Sq
ueez
e
a
nd
ex
cit
a
t
io
n
-
g
ener
a
t
iv
e
a
dv
er
s
a
ria
l net
w
o
rk
f
r
a
m
ewo
r
k
GANs
ar
e
ex
ten
s
iv
ely
u
tili
z
ed
in
n
u
m
er
o
u
s
d
o
m
ain
s
f
o
r
g
en
e
r
atio
n
task
s
.
T
h
e
co
n
v
en
tio
n
al
ar
ch
itectu
r
e
o
f
GANs,
wh
ich
co
n
s
is
ts
o
f
a
g
en
er
at
o
r
(
G
)
a
n
d
a
d
is
cr
im
in
ato
r
(
D
)
,
h
as
r
ec
eiv
ed
ex
ce
p
tio
n
al
co
n
s
id
er
atio
n
in
r
ec
en
t
y
ea
r
s
.
T
h
e
atten
tio
n
-
d
r
i
v
en
GAN
is
in
tr
o
d
u
ce
d
in
o
u
r
wo
r
k
f
o
r
SL
v
id
eo
g
en
er
atio
n
.
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
s
h
o
ws
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t
i
n
th
e
g
en
er
atio
n
o
f
SL
v
id
eo
q
u
ality
b
y
u
tili
zin
g
SE
atten
tio
n
m
o
d
u
le
an
d
ca
r
ef
u
lly
em
p
h
asizin
g
s
k
eleta
l
p
o
s
es
.
A
g
r
a
p
h
ical
d
e
p
ictio
n
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
'
s
g
en
er
al
s
tr
u
ctu
r
e
is
s
h
o
wn
in
Fig
u
r
e
1
.
I
ts
th
o
r
o
u
g
h
d
escr
ip
ti
o
n
is
g
iv
en
in
th
i
s
s
ec
tio
n
.
3
.
3
.
1
.
Sq
ueez
e
a
nd
ex
cit
a
t
io
n
-
g
ener
a
t
iv
e
a
dv
er
s
a
ria
l net
wo
rk
a
r
chit
ec
t
ure
SE
-
GAN
is
m
ad
e
u
p
o
f
two
k
ey
co
m
p
o
n
en
ts
:
a
g
en
er
ato
r
a
n
d
a
d
is
cr
im
in
ato
r
.
T
h
e
d
is
cr
im
in
ato
r
is
u
s
ed
to
ch
ar
ac
ter
ize
th
e
d
if
f
e
r
en
ce
s
b
etwe
en
th
e
cr
ea
ted
d
etec
tio
n
im
ag
e
an
d
th
e
r
ef
e
r
e
n
ce
im
ag
e,
wh
e
r
ea
s
th
e
g
en
er
at
o
r
is
in
ten
d
ed
to
p
r
o
v
id
e
a
d
etec
tio
n
im
ag
e
th
at
is
as
ac
cu
r
ate
as
th
e
r
ef
er
en
ce
im
ag
e.
T
h
e
f
r
am
ewo
r
k
m
i
n
im
izes
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
g
en
er
ate
d
d
etec
tio
n
im
ag
e
an
d
th
e
r
ef
er
en
ce
im
ag
e
b
y
u
s
e
o
f
ad
v
e
r
s
ar
ial
tr
ain
in
g
to
m
in
i
m
ize
f
-
d
iv
er
g
en
ce
.
T
h
e
g
en
er
ato
r
an
d
d
is
cr
im
in
ato
r
ar
e
co
m
p
r
is
ed
o
f
b
o
t
h
en
c
o
d
er
s
an
d
d
ec
o
d
er
s
.
T
h
e
e
n
co
d
er
in
th
e
g
en
er
ato
r
is
co
n
s
tr
u
cte
d
u
s
in
g
a
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
ac
tiv
atio
n
la
y
er
,
a
n
o
r
m
al
izatio
n
lay
er
,
an
d
a
co
n
v
o
l
u
tio
n
al
lay
e
r
.
A
d
ec
o
n
v
o
lu
tio
n
la
y
er
,
a
n
o
r
m
aliza
tio
n
lay
er
,
an
d
a
R
eL
U
ac
tiv
atio
n
lay
er
m
ak
e
u
p
th
e
d
ec
o
d
er
.
T
h
e
g
en
e
r
ato
r
G
h
as
a
U
-
s
h
ap
ed
s
tr
u
ctu
r
e
a
n
d
is
m
ad
e
u
p
o
f
n
en
co
d
er
s
an
d
d
ec
o
d
er
s
in
ad
d
itio
n
to
th
e
SE
atten
tio
n
m
o
d
u
le
th
at
is
co
v
er
ed
in
th
e
s
u
b
s
ec
tio
n
th
at
f
o
llo
ws.
I
t
is
th
e
r
esp
o
n
s
ib
ilit
y
o
f
th
e
SE
m
o
d
u
le
to
en
h
a
n
ce
th
e
s
ig
n
if
ican
t
f
ea
tu
r
es
tr
an
s
m
itted
b
y
th
e
co
n
v
o
lu
tio
n
p
r
o
ce
s
s
in
th
e
en
co
d
er
an
d
d
ec
o
d
e
r
wh
ile
d
is
r
eg
ar
d
in
g
t
h
e
less
s
ig
n
if
ican
t
o
n
es.
T
h
e
n
ex
t
la
y
er
s
r
ec
eiv
e
th
e
r
ewe
ig
h
ted
f
ea
tu
r
es
th
at
wer
e
ac
q
u
ir
ed
f
r
o
m
t
h
e
SE
m
o
d
u
le.
A
co
n
v
o
lu
tio
n
lay
er
,
a
R
eL
U
ac
tiv
atio
n
lay
e
r
,
th
r
ee
en
c
o
d
er
s
,
an
d
t
h
en
a
n
o
th
e
r
co
n
v
o
lu
tio
n
lay
e
r
an
d
a
tan
h
lay
er
m
a
k
e
u
p
th
e
d
is
cr
im
in
ato
r
.
T
h
e
g
en
er
at
o
r
g
en
er
ates
an
I
SL
im
ag
e
S
f
r
o
m
a
s
k
eleta
l
p
o
s
e
.
T
h
e
d
is
cr
im
in
ato
r
'
s
in
p
u
ts
ar
e
th
o
u
g
h
t
t
o
b
e
a
s
k
eleta
l
p
o
s
e
,
th
e
g
en
er
ated
d
etec
tio
n
im
ag
e
,
an
d
th
e
r
ef
er
en
ce
im
ag
e
.
Fo
r
ea
ch
im
ag
e,
th
e
d
is
cr
im
in
ato
r
cr
ea
t
es
a
r
ep
r
esen
tatio
n
o
f
th
e
v
a
r
iatio
n
al
v
ar
iab
le,
wh
ich
is
s
p
ec
if
ied
as
.
T
o
g
et
th
e
g
en
e
r
ated
d
etec
tio
n
im
a
g
es
to
r
esem
b
le
th
e
r
e
f
er
en
ce
im
ag
es,
ad
v
er
s
ar
ial
tr
ain
in
g
o
f
th
e
d
is
cr
im
in
ato
r
a
n
d
g
en
er
ato
r
aim
s
to
m
in
im
ize
f
-
d
iv
er
g
en
ce
.
3
.
3
.
2
.
Sq
ueez
e
a
nd
ex
cit
a
t
io
n m
o
du
le
T
h
e
r
esu
lt
o
f
th
e
co
n
v
o
lu
tio
n
o
p
er
atio
n
is
f
o
r
m
ed
b
y
s
u
m
m
ar
izin
g
all
ch
an
n
els,
an
d
th
e
r
elatio
n
s
h
ip
am
o
n
g
th
e
c
h
an
n
els
is
in
v
is
ib
ly
in
ter
twin
ed
in
th
e
s
p
atial
d
a
ta.
T
h
er
ef
o
r
e,
th
e
SE
m
o
d
u
le'
s
in
tr
o
d
u
ctio
n
aim
s
to
in
cr
ea
s
e
n
etwo
r
k
s
en
s
itiv
ity
to
k
e
y
asp
ec
ts
b
y
s
im
u
latin
g
ch
an
n
el
in
ter
d
e
p
en
d
e
n
ce
.
L
ess
b
en
ef
icial
tr
aits
ar
e
th
er
ef
o
r
e
s
u
p
p
r
ess
ed
.
T
h
e
SE
m
o
d
u
le
is
a
co
m
p
o
n
en
t
o
f
th
e
U
-
s
h
ap
ed
g
en
er
ato
r
co
n
s
tr
u
ctio
n
th
at
co
m
es
ju
s
t
af
ter
th
e
en
co
d
er
s
an
d
d
ec
o
d
er
s
.
T
h
e
m
o
d
u
le
c
o
n
s
is
t
s
o
f
two
d
is
tin
ct
o
p
er
ati
o
n
s
:
ex
citatio
n
a
n
d
s
q
u
ee
zin
g
.
T
h
e
r
esu
lt
o
f
th
e
d
ec
o
d
er
an
d
e
n
co
d
e
r
U
co
n
v
o
l
u
tio
n
p
r
o
ce
s
s
p
r
o
v
id
es
s
p
atial
an
d
ch
an
n
el
d
ata.
T
o
r
ec
o
n
s
tr
u
ct
f
ea
tu
r
es,
U
is
co
n
s
id
er
ed
t
h
e
in
p
u
t
to
th
e
S
E
m
o
d
u
le.
Her
e,
co
n
s
id
er
th
at
th
e
co
m
b
in
atio
n
o
f
ch
an
n
els
∈
×
s
er
v
es
as
th
e
in
p
u
t
f
o
r
th
e
SE
m
o
d
u
le
=
[
1
,
2
,
…
,
]
.
A
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
o
p
er
atio
n
is
th
en
u
s
ed
to
co
m
p
r
ess
th
e
f
ea
tu
r
e
m
ap
s
f
r
o
m
th
e
co
n
v
o
l
u
tio
n
o
p
er
ati
o
n
in
to
a
ch
an
n
el
d
escr
ip
to
r
.
Sq
u
ee
zin
g
with
s
p
atial
d
im
en
s
io
n
s
×
y
ield
s
∈
1
×
1
×
.
T
h
e
ex
p
r
ess
io
n
f
o
r
th
e
th
elem
en
t o
f
∈
1
×
1
×
is
as in
(
8
)
.
=
(
)
=
1
×
∑
∑
(
,
)
=
1
=
1
(
8
)
As
a
r
esu
lt,
th
e
v
ec
to
r
,
wh
ich
en
co
d
es
all
o
f
th
e
in
f
o
r
m
atio
n
in
th
e
im
ag
e,
in
co
r
p
o
r
ates
t
h
e
g
lo
b
al
s
p
atial
in
f
o
r
m
atio
n
th
r
o
u
g
h
th
e
s
q
u
ee
ze
p
r
o
ce
s
s
.
Fig
u
r
e
2
d
ep
icts
th
e
SE
m
o
d
u
le'
s
o
p
er
atio
n
.
Fig
u
r
e
2
.
T
h
e
SE
m
o
d
u
le
ar
c
h
itectu
r
e
G
l
o
b
a
l
A
v
e
r
a
g
e
P
o
o
l
i
n
g
F
u
l
l
y
C
o
n
n
e
c
t
e
d
R
e
L
U
F
u
l
l
y
C
o
n
n
e
c
t
e
d
S
i
g
m
o
i
d
x
1
×
1
×C
1
×
1
×C
1
×
1
×C
/
r
W
H
U
C
W
H
C
U
~
sc
al
e
F
sq
F
ex
F
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
I
n
d
ia
n
s
ig
n
la
n
g
u
a
g
e
vid
eo
g
e
n
era
tio
n
u
s
in
g
a
tten
tio
n
-
e
n
h
a
n
ce
d
g
e
n
era
tive
…
(
P
r
a
ch
i P
r
a
mo
d
Wa
g
h
ma
r
e)
3677
T
h
e
ex
citatio
n
o
p
er
atio
n
p
r
o
v
i
d
es
a
co
m
p
r
e
h
en
s
iv
e
d
er
iv
ed
s
et
o
f
ch
an
n
el
-
wis
e
d
ep
e
n
d
en
ci
es,
wh
ich
ca
n
b
e
b
etter
u
tili
ze
d
in
co
n
ju
n
ctio
n
with
th
e
in
teg
r
ated
g
lo
b
al
in
f
o
r
m
atio
n
g
ain
ed
b
y
th
e
s
q
u
ee
ze
o
p
er
atio
n
.
T
h
e
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
o
p
er
atio
n
'
s
o
u
tp
u
t
is
co
n
v
er
ted
to
2
(
(
1
)
)
.
Her
e,
th
e
weig
h
ts
o
f
two
co
m
p
letely
lin
k
e
d
lay
e
r
s
with
s
et
to
2
ar
e
d
en
o
te
d
b
y
1
∈
/
×
an
d
2
∈
×
/
.
T
h
e
n
o
n
-
lin
ea
r
ity
in
th
e
ex
citatio
n
b
lo
ck
th
en
f
o
r
m
s
a
b
o
ttlen
ec
k
am
o
n
g
th
e
t
wo
f
u
lly
co
n
n
ec
te
d
lay
er
s
.
As
s
h
o
wn
in
Fig
u
r
e
2
,
th
e
f
r
am
ewo
r
k
o
f
th
e
e
x
citatio
n
o
p
er
atio
n
c
o
n
s
is
ts
o
f
a
d
i
m
en
s
io
n
ality
-
in
cr
ea
s
in
g
la
y
er
with
2
,
a
R
eL
U
lay
er
,
an
d
a
d
im
en
s
io
n
ality
-
r
e
d
u
ctio
n
la
y
er
with
1
.
T
h
e
in
ter
v
al
[
0
,
1
]
is
th
e
co
n
s
tr
ain
t
o
f
2
(
(
1
)
)
by
a
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
(
·
)
.
T
h
e
in
itial
f
u
lly
co
n
n
ec
ted
lay
er
an
d
th
e
R
eL
U
lay
er
co
m
b
in
e
to
p
r
o
v
id
e
a
f
ea
tu
r
e
v
ec
to
r
with
a
ch
a
n
n
el
d
im
en
s
io
n
o
f
/
.
T
h
is
f
ea
tu
r
e
v
ec
to
r
is
th
e
n
tr
a
n
s
f
o
r
m
ed
b
y
th
e
s
ig
m
o
i
d
lay
er
s
an
d
th
e
s
ec
o
n
d
f
u
lly
c
o
n
n
ec
ted
lay
e
r
in
to
a
v
ec
to
r
with
a
ch
an
n
el
d
im
e
n
s
io
n
o
f
.
T
h
u
s
,
th
e
f
in
a
l
s
ig
m
o
id
lay
er
'
s
r
esu
lt a
s
in
(
9
)
.
=
(
,
)
=
(
2
(
1
)
)
(
9
)
W
h
er
e
d
en
o
tes
th
ch
an
n
el'
s
s
ig
n
if
ican
ce
.
T
h
e
f
in
al
r
esu
lt is
p
r
o
d
u
ce
d
b
y
s
ca
lin
g
u
s
in
g
as
in
(
1
0
)
.
̅
=
(
,
)
=
.
(
1
0
)
I
n
th
e
ex
citatio
n
o
p
er
atio
n
,
th
e
f
ir
s
t
f
u
lly
co
n
n
ec
te
d
lay
er
'
s
d
im
en
s
io
n
ality
r
ed
u
ctio
n
is
in
ten
d
ed
to
lo
wer
th
e
n
etwo
r
k
'
s
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
wh
ile
th
e
s
ec
o
n
d
f
u
lly
co
n
n
ec
ted
lay
er
'
s
d
im
en
s
io
n
ality
in
cr
ea
s
e
g
u
ar
an
tees
th
at
th
e
d
im
en
s
io
n
s
r
em
ain
th
e
s
am
e
f
o
r
d
eter
m
in
in
g
th
e
weig
h
t
o
f
ea
ch
lay
er
f
ea
tu
r
e.
T
h
e
s
h
ap
e
o
f
is
r
ec
alcu
lated
to
b
e
=
[
1
1
,
2
2
,
.
.
.
,
]
.
T
h
e
SE
atten
tio
n
m
o
d
u
le
ad
a
p
tiv
ely
h
ig
h
lig
h
ts
th
e
k
ey
c
h
an
n
els
w
h
ile
ig
n
o
r
in
g
th
e
less
s
ig
n
if
ican
t
o
n
es.
T
h
e
d
etailed
lay
e
r
-
wi
s
e
co
n
f
ig
u
r
atio
n
is
s
h
o
wn
in
T
a
b
le
1
.
Fig
u
r
e
3
s
h
o
ws
s
tep
wis
e
p
r
o
ce
s
s
o
f
t
h
e
o
v
er
all
SE
-
GAN
ar
ch
itectu
r
e,
wh
er
e
GAN
-
b
ased
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
r
ec
o
n
s
t
r
u
ctio
n
alo
n
g
with
SE
atten
tio
n
m
ec
h
an
is
m
an
d
d
is
cr
im
in
at
o
r
b
lo
ck
s
ar
e
u
s
ed
.
I
n
th
is
f
i
g
u
r
e,
T
is
n
u
m
b
e
r
o
f
f
r
am
es
,
H
is
h
eig
h
t,
W
is
w
id
th
,
C
in
is
in
p
u
t
c
h
an
n
els
,
C
o
n
v
is
co
n
v
o
l
u
tio
n
al
lay
er
,
C
o
n
v
tr
a
n
s
p
o
s
e
is
tr
an
s
p
o
s
ed
co
n
v
o
lu
tio
n
,
s
is
s
tr
id
e
,
p
is
p
ad
d
in
g
,
an
d
B
N
is
b
atch
n
o
r
m
aliza
tio
n
.
T
ab
le
1
.
L
a
y
er
d
etails f
o
r
th
e
p
r
o
p
o
s
ed
atten
tio
n
-
d
r
i
v
en
GAN
La
y
e
r
n
o.
N
e
t
w
o
r
k
c
o
n
f
i
g
u
r
a
t
i
o
n
a
n
d
h
y
p
e
r
p
a
r
a
met
e
r
s
G
e
n
e
r
a
t
o
r
(
e
n
c
o
d
e
r
)
La
y
e
r
1
C
o
n
v
6
4
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
R
e
LU
La
y
e
r
2
C
o
n
v
1
2
8
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
La
y
e
r
3
C
o
n
v
1
2
8
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
SE
La
y
e
r
1
G
l
o
b
a
l
a
v
e
r
a
g
e
p
o
o
l
i
n
g
(
6
4
,
1
,
1
,
3
)
La
y
e
r
2
D
e
n
se
,
R
e
LU
La
y
e
r
3
D
e
n
se
,
s
i
g
m
o
i
d
G
e
n
e
r
a
t
o
r
(
d
e
c
o
d
e
r
)
La
y
e
r
1
C
o
n
v
1
2
8
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
La
y
e
r
2
C
o
n
v
1
2
8
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
La
y
e
r
3
C
o
n
v
6
4
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
R
e
L
U
D
i
scri
m
i
n
a
t
o
r
La
y
e
r
1
C
o
n
v
3
2
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
R
e
L
U
La
y
e
r
2
C
o
n
v
3
2
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
La
y
e
r
3
C
o
n
v
3
2
(
4
×
4
)
,
s
=
2
,
p
=
2
,
s
=
2
,
B
N
,
R
e
LU
T
h
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
h
a
s
g
en
er
atio
n
an
d
d
is
cr
im
in
ato
r
b
lo
ck
s
,
wh
er
e
th
e
g
en
er
ato
r
b
lo
ck
h
as
en
co
d
er
a
n
d
d
ec
o
d
er
u
n
its
.
E
a
ch
en
co
d
e
r
h
as
a
th
r
ee
-
la
y
er
n
etwo
r
k
.
T
h
e
f
i
r
s
t
lay
er
ex
tr
ac
t
lo
w
f
ea
tu
r
es
s
u
ch
as
ed
g
es
an
d
co
n
to
u
r
s
.
T
h
e
s
ec
o
n
d
la
y
er
e
x
tr
ac
ts
m
id
-
lev
e
l
f
ea
tu
r
es
s
u
ch
as
u
p
p
e
r
b
o
d
y
p
ar
ts
an
d
jo
i
n
ts
,
wh
ile
th
e
th
ir
d
lay
er
e
x
tr
ac
ts
h
ig
h
-
lev
el
f
ea
t
u
r
es
s
u
ch
as
p
o
s
e
s
tr
u
ctu
r
e.
All
th
e
lear
n
ed
f
e
atu
r
es
ar
e
ex
ten
d
e
d
to
th
e
SE
b
lo
ck
,
wh
e
r
e
th
e
s
q
u
ee
ze
b
lo
ck
c
o
m
p
u
tes
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
t
o
ca
p
tu
r
e
g
l
o
b
al
in
f
o
r
m
atio
n
o
f
ea
ch
ch
a
n
n
el.
Her
e
,
c
h
an
n
els
ca
n
b
e
k
ey
jo
in
ts
f
o
r
h
an
d
an
d
f
ac
e.
T
h
e
ex
citatio
n
b
l
o
ck
lear
n
s
ch
a
n
n
el
im
p
o
r
tan
ce
an
d
g
en
er
ates
ap
p
r
o
p
r
iate
weig
h
ts
an
d
f
in
e
d
eta
ils
.
T
h
e
g
en
er
ato
r
d
ec
o
d
e
r
d
o
es
u
p
s
am
p
lin
g
an
d
r
ec
o
n
s
tr
u
cts
h
ig
h
-
lev
el
f
ea
tu
r
e
s
,
r
ec
o
v
er
s
m
id
-
lev
el
an
d
f
in
e
d
etails
,
an
d
en
h
an
ce
s
th
e
s
p
atial
r
eso
lu
tio
n
.
T
h
e
d
is
cr
im
in
ato
r
ex
tr
ac
ts
f
ea
tu
r
es
an
d
m
ak
e
t
h
e
d
ec
is
io
n
wh
eth
er
th
e
g
en
e
r
ated
s
am
p
le
is
r
ea
l
an
d
f
ak
e
.
W
ith
th
e
in
clu
s
io
n
o
f
SE
-
GAN
atten
tio
n
m
ec
h
an
is
m
,
th
e
d
is
cr
im
in
ato
r
r
ec
eiv
es
i
n
f
o
r
m
atio
n
o
f
im
p
o
r
tan
t
c
h
an
n
e
l
g
estu
r
es
,
wh
ich
ar
e
en
h
an
ce
d
with
th
e
atten
tio
n
m
ec
h
an
is
m
.
T
h
er
e
is
im
p
r
o
v
ed
h
a
n
d
m
o
ti
o
n
an
d
p
o
s
e
d
etails
an
d
s
m
o
o
th
tem
p
o
r
al
co
n
s
is
ten
cy
,
wh
ich
is
ev
id
e
n
t
with
th
e
im
p
r
o
v
em
en
t
in
th
e
s
tr
u
ctu
r
al
s
im
ilar
ity
in
d
ex
m
ea
s
u
r
e
(
SS
I
M)
,
p
ea
k
s
ig
n
al
-
to
-
n
o
is
e
r
atio
(
PS
NR
)
,
an
d
tem
p
o
r
al
co
n
s
is
ten
cy
m
etr
ic
(
T
C
M)
s
co
r
e
ca
lcu
lated
f
o
r
b
asic G
AN,
d
y
n
am
i
c
GAN
o
n
two
d
i
f
f
er
en
t
d
atasets
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
7
2
-
3
6
8
2
3678
Fig
u
r
e
3
.
Pro
ce
s
s
an
d
lay
er
wi
s
e
r
esu
lts
o
f
th
e
GAN
-
b
ased
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
r
ec
o
n
s
tr
u
ctio
n
with
SE
atten
tio
n
m
ec
h
an
is
m
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
d
ataset
d
escr
ip
tio
n
.
I
t
also
r
esu
lts
an
d
co
m
p
ar
es
th
e
p
er
f
o
r
m
an
ce
o
f
b
aselin
e
m
o
d
els
an
d
th
e
p
r
o
p
o
s
ed
m
o
d
el.
Op
tim
al
h
y
p
e
r
p
ar
am
eter
s
s
elec
ted
th
r
o
u
g
h
ex
p
er
im
en
tatio
n
ar
e
also
in
clu
d
ed
.
4
.
1
.
Da
t
a
s
et
des
cr
iptio
n
An
SL
v
i
d
eo
d
ataset,
I
n
d
ian
s
ig
n
lan
g
u
ag
e
v
i
d
eo
an
d
tex
t
d
at
aset
f
o
r
s
en
ten
ce
s
(
I
SLVT
)
[
2
5
]
f
o
r
I
SL
is
cr
ea
ted
f
o
r
th
is
r
esear
ch
.
T
h
e
p
u
r
p
o
s
e
o
f
t
h
is
co
r
p
u
s
is
to
aid
th
e
I
n
d
ian
Dea
f
c
o
m
m
u
n
ity
.
T
h
e
7
0
0
v
id
eo
s
in
th
is
u
n
iq
u
e
c
o
r
p
u
s
wer
e
g
at
h
er
ed
f
r
o
m
a
s
in
g
le
s
ig
n
er
an
d
co
n
tain
o
n
ly
a
s
in
g
le
b
ac
k
g
r
o
u
n
d
with
well
-
s
et
b
r
ig
h
tn
ess
co
n
d
itio
n
s
wh
ile
r
e
co
r
d
in
g
.
T
h
e
o
th
er
d
ataset
u
s
e
d
is
th
e
d
ataset
f
o
r
I
SL
-
c
o
n
tin
u
o
u
s
s
ig
n
lan
g
u
ag
e
tr
an
s
latio
n
an
d
r
ec
o
g
n
itio
n
(
I
S
L
-
C
SLT
R
)
[
2
6
]
.
4
.
2
.
E
x
perim
ent
a
l r
esu
lt
s
T
h
e
p
er
f
o
r
m
a
n
ce
c
o
m
p
ar
is
o
n
o
f
th
e
b
aselin
e
GAN,
d
y
n
am
ic
GAN
,
an
d
p
r
o
p
o
s
ed
SE
-
GAN
m
o
d
el
is
p
r
esen
ted
in
T
a
b
le
2
.
T
h
e
co
m
p
ar
is
o
n
is
b
ased
o
n
th
r
ee
k
ey
ev
alu
atio
n
m
etr
ics:
SS
I
M,
PS
NR
,
an
d
T
C
M.
T
h
ese
m
etr
ics ass
es
s
th
e
v
is
u
al
q
u
ality
,
clar
ity
,
a
n
d
tem
p
o
r
al
co
n
s
is
ten
cy
o
f
th
e
g
en
e
r
ated
f
r
am
es.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
g
en
e
r
ativ
e
m
o
d
els
M
o
d
e
l
s
I
S
L
-
C
S
LTR
d
a
t
a
se
t
[
2
6
]
I
S
LV
T
(
o
u
r
)
si
g
n
l
a
n
g
u
a
g
e
d
a
t
a
s
e
t
[
2
5
]
S
S
I
M
(
%)
P
S
N
R
(
d
B
)
TC
M
S
S
I
M
(
%)
P
S
N
R
(
d
B
)
TC
M
GAN
[
2
7
]
81
7
.
6
0
.
4
8
8
0
.
1
5
1
6
.
8
4
0
.
5
D
y
n
a
mi
c
G
A
N
[
1
8
]
9
3
.
7
8
.
5
0
.
7
1
2
8
5
.
8
1
1
8
.
7
9
0
.
6
5
SE
-
G
A
N
(
a
t
t
e
n
t
i
o
n
-
d
r
i
v
e
n
)
(
p
r
o
p
o
s
e
d
)
99
9
.
6
0
.
9
1
1
9
9
.
6
0
3
1
.
1
0
0
.
9
1
1
Fro
m
T
ab
le
2
,
it
is
ev
id
en
t
th
at
th
e
atten
tio
n
-
d
r
iv
en
GAN
o
u
tp
er
f
o
r
m
s
b
o
th
th
e
s
tan
d
a
r
d
GAN
an
d
d
y
n
am
ic
GAN
ac
r
o
s
s
all
th
r
ee
m
etr
ics.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ac
h
iev
es
9
9
.
6
0
%
SS
I
M,
d
e
m
o
n
s
tr
atin
g
s
u
p
e
r
io
r
s
tr
u
ctu
r
al
s
im
ilar
ity
to
th
e
o
r
ig
in
al
f
r
am
es
co
m
p
ar
ed
to
d
y
n
am
ic
GAN
(
9
8
.
8
1
%)
an
d
GAN
(
9
8
.
1
5
%).
A
h
ig
h
er
SS
I
M
s
u
g
g
ests
th
at
t
h
e
g
en
e
r
ated
f
r
am
es
r
etain
m
o
r
e
s
tr
u
ctu
r
al
d
etails,
m
ak
in
g
th
e
m
v
is
u
ally
cl
o
s
er
to
r
ea
l
SL
v
id
eo
s
.
Similar
ly
,
th
e
PS
N
R
v
alu
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
s
ig
n
if
ican
tly
h
ig
h
er
at
3
1
.
1
0
d
B
,
co
m
p
ar
ed
to
1
8
.
7
9
d
B
f
o
r
d
y
n
am
ic
GAN
an
d
1
6
.
8
4
d
B
f
o
r
t
h
e
s
tan
d
ar
d
GAN.
A
h
ig
h
e
r
P
SNR
v
alu
e
s
ig
n
if
ies
r
ed
u
ce
d
n
o
is
e
a
n
d
im
p
r
o
v
ed
im
ag
e
clar
ity
in
th
e
g
e
n
er
a
ted
f
r
am
es,
co
n
f
i
r
m
in
g
th
e
m
o
d
el’
s
ab
ilit
y
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
I
n
d
ia
n
s
ig
n
la
n
g
u
a
g
e
vid
eo
g
e
n
era
tio
n
u
s
in
g
a
tten
tio
n
-
e
n
h
a
n
ce
d
g
e
n
era
tive
…
(
P
r
a
ch
i P
r
a
mo
d
Wa
g
h
ma
r
e)
3679
r
ec
o
n
s
tr
u
ct
h
ig
h
-
f
id
elity
SL
v
id
eo
s
.
T
em
p
o
r
al
co
n
s
is
ten
cy
,
m
ea
s
u
r
ed
b
y
T
C
M,
is
cr
u
cial
f
o
r
s
m
o
o
th
m
o
ti
o
n
r
ep
r
esen
tatio
n
in
v
id
eo
g
en
e
r
atio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ac
h
iev
es
a
T
C
M
s
co
r
e
o
f
0
.
9
1
1
,
s
u
b
s
tan
tially
o
u
tp
er
f
o
r
m
in
g
th
e
d
y
n
a
m
ic
GAN
(
0
.
4
8
)
an
d
GAN
(
0
.
5
0
)
.
T
h
is
in
d
icate
s
th
at
t
h
e
att
en
tio
n
-
d
r
iv
en
GAN
g
en
er
ates
tem
p
o
r
ally
s
tab
le
f
r
am
es
with
co
n
s
is
ten
t
m
o
tio
n
tr
an
s
itio
n
s
,
r
ed
u
cin
g
f
lick
e
r
in
g
an
d
u
n
n
at
u
r
al
m
o
v
em
en
ts
.
T
h
is
p
er
f
o
r
m
an
ce
s
h
o
ws co
n
s
is
ten
t r
esu
lts
f
o
r
b
o
th
d
atasets
.
T
h
e
q
u
an
titativ
e
co
m
p
a
r
is
o
n
in
ter
m
s
o
f
PS
N
R
,
S
SIM
,
an
d
T
C
M
f
o
r
th
e
im
p
lem
en
ted
GAN,
d
y
n
am
ic
GAN
m
o
d
els
,
a
n
d
t
h
e
p
r
o
p
o
s
ed
SE
-
GAN
m
o
d
el
v
er
s
u
s
th
e
tr
ain
in
g
ep
o
ch
s
is
s
h
o
wn
in
Fig
u
r
e
4
.
I
t
s
h
o
ws
th
at
SE
-
GAN
co
n
s
is
t
en
tly
o
u
tp
er
f
o
r
m
s
b
aselin
e
m
o
d
els
in
ter
m
s
o
f
p
ictu
r
e
q
u
ality
,
tem
p
o
r
al
co
n
s
is
ten
cy
,
an
d
s
tr
u
ctu
r
al
s
im
ilar
ity
.
I
t p
r
o
v
es th
e
ef
f
icac
y
o
f
th
is
m
eth
o
d
f
o
r
p
r
o
d
u
cin
g
clo
s
er
to
r
ea
lis
tic
SL
v
id
eo
s
.
Fig
u
r
e
5
s
h
o
ws
q
u
alitativ
e
r
esu
lts
co
m
p
ar
is
o
n
f
o
r
GAN,
d
y
n
am
ic
GAN,
an
d
th
e
p
r
o
p
o
s
ed
SE
-
GAN
m
eth
o
d
s
f
o
r
t
h
e
g
e
n
er
ated
SL
f
r
am
es
ac
r
o
s
s
v
a
r
io
u
s
tr
ain
i
n
g
ep
o
c
h
s
(
2
0
–
1
0
0
)
.
T
h
ese
r
esu
l
ts
also
in
d
icate
th
at
in
co
m
p
a
r
is
o
n
to
th
e
b
asic
GAN
an
d
d
y
n
am
ic
GAN
m
o
d
els,
th
e
SE
-
GAN
m
o
d
el
p
r
o
d
u
ce
s
m
o
r
e
v
is
u
ally
co
n
s
is
ten
t,
an
d
en
h
an
ce
d
-
q
u
al
ity
SL
f
r
am
es a
s
th
e
n
u
m
b
er
o
f
ep
o
ch
s
in
c
r
ea
s
es
.
T
h
e
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
ts
in
ev
alu
atio
n
m
etr
ics
h
ig
h
lig
h
t
th
e
m
o
d
el
’
s
p
o
ten
tial
f
o
r
r
ea
l
-
wo
r
ld
SL
T
ap
p
licatio
n
s
.
Fu
r
t
h
er
,
c
o
m
p
ar
ativ
e
r
esu
lts
f
o
r
th
e
v
id
e
o
f
r
am
es
g
en
er
ate
d
b
y
th
e
b
as
elin
e
m
o
d
els
an
d
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
th
e
o
r
ig
i
n
al
s
am
p
le
v
id
e
o
,
f
o
llo
wed
b
y
th
e
p
o
s
e
esti
m
atio
n
s
tep
b
y
th
e
Op
en
Po
s
e
ar
e
s
h
o
wn
in
Fig
u
r
e
6
.
Fig
u
r
e
6
(
a)
s
h
o
ws
th
e
v
id
eo
f
r
am
es
g
en
er
atio
n
r
esu
lts
f
o
r
th
e
s
ig
n
s
m
ad
e
f
o
r
th
e
tex
t
“M
o
th
er
Fo
o
d
C
o
o
k
”
f
r
o
m
I
S
L
VT
[
2
5
]
d
ataset
an
d
Fig
u
r
e
6
(
b
)
d
ep
icts
th
e
r
esu
lts
f
o
r
th
e
s
ig
n
s
m
ad
e
f
o
r
th
e
tex
t
“R
u
f
r
ee
to
d
a
y
”
f
r
o
m
th
e
I
SL
-
C
SLT
R
d
ataset
[
2
6
]
.
Ov
er
all,
th
e
r
esu
lts
v
alid
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
Atten
tio
n
-
Dr
iv
en
GAN
in
en
h
an
cin
g
th
e
q
u
ality
,
clar
ity
,
an
d
s
m
o
o
th
n
ess
o
f
SL
v
i
d
eo
g
e
n
e
r
atio
n
.
Fig
u
r
e
4
.
Qu
a
n
titativ
e
co
m
p
a
r
is
o
n
o
f
GAN,
d
y
n
am
ic
GAN,
an
d
p
r
o
p
o
s
ed
SE
-
GAN
m
o
d
el
s
Fig
u
r
e
5
.
Qu
alitativ
e
co
m
p
ar
is
o
n
f
o
r
g
en
e
r
ated
SL
f
r
am
es a
n
d
p
r
o
g
r
ess
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
6
7
2
-
3
6
8
2
3680
(
a)
(
b
)
Fig
u
r
e
6
.
C
o
m
p
a
r
ativ
e
v
id
e
o
f
r
am
e
g
en
e
r
atio
n
r
esu
lts
f
o
r
(
a)
s
am
p
le
o
u
tp
u
t
f
o
r
d
ataset
[
2
5
]
an
d
(
b
)
s
am
p
le
o
u
tp
u
t
f
o
r
d
ataset
[
2
6
]
5.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
ad
d
r
ess
es
th
e
c
h
allen
g
es
in
SLT
f
o
r
I
SL
v
id
eo
s
y
n
th
esis
with
im
p
r
o
v
ed
v
id
eo
q
u
ality
b
y
in
tr
o
d
u
cin
g
SE
-
GAN
f
r
a
m
ewo
r
k
.
I
t
co
n
n
ec
ts
co
m
p
u
t
er
v
is
io
n
f
o
r
g
estu
r
e
an
d
p
o
s
e
u
n
d
er
s
tan
d
i
n
g
,
NL
P
-
d
r
iv
en
tex
t
-
to
-
s
ig
n
tr
an
s
l
atio
n
,
an
d
m
u
ltimo
d
al
lear
n
in
g
f
r
am
ewo
r
k
s
th
at
in
teg
r
ate
li
n
g
u
is
tic
an
d
v
is
u
al
cu
es
f
o
r
m
o
r
e
e
f
f
ec
tiv
e
SL
s
y
n
th
esis
.
T
h
r
o
u
g
h
a
c
o
m
p
r
eh
e
n
s
iv
e
ap
p
r
o
ac
h
in
v
o
lv
in
g
v
id
eo
f
r
a
m
e
e
x
tr
ac
tio
n
,
s
k
eleta
l
jo
in
t
co
o
r
d
in
ate
ex
tr
a
ctio
n
,
an
d
DT
W
f
o
r
d
ata
in
te
r
p
o
latio
n
/ex
tr
a
p
o
latio
n
in
th
e
p
r
ep
r
o
ce
s
s
in
g
s
tep
,
th
e
atten
tio
n
-
d
r
i
v
en
GAN
with
an
SE
atten
tio
n
m
ec
h
an
is
m
d
em
o
n
s
tr
ates
its
ef
f
ec
tiv
en
ess
in
g
en
er
atin
g
clo
s
e
to
r
ea
lis
tic
SL
v
id
eo
s
.
T
h
e
atten
tio
n
m
ec
h
a
n
is
m
d
y
n
am
icall
y
m
o
d
u
lates
th
e
g
en
er
atio
n
p
r
o
ce
s
s
,
allo
win
g
th
e
m
o
d
el
to
atten
d
to
r
ele
v
an
t
s
k
eleta
l
p
o
s
e
s
eq
u
en
ce
s
an
d
f
r
am
es,
r
esu
ltin
g
in
im
p
r
o
v
e
d
r
ec
o
g
n
itio
n
m
etr
ics.
Per
f
o
r
m
an
ce
e
v
alu
atio
n
u
s
in
g
I
SL
-
C
SLT
R
b
en
ch
m
ar
k
s
ig
n
co
r
p
o
r
a
an
d
I
SLVT
s
ig
n
c
o
r
p
o
r
a
cr
ea
ted
f
o
r
th
is
r
esear
ch
in
d
icate
s
s
u
b
s
tan
tial
en
h
a
n
ce
m
en
ts
in
v
id
e
o
f
r
am
e
q
u
ality
an
d
tem
p
o
r
al
c
o
h
er
en
ce
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
with
im
p
lem
en
ted
m
o
d
els
o
f
GAN
an
d
d
y
n
am
ic
GAN
s
h
o
wca
s
es
th
e
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
ter
m
s
o
f
SS
I
M,
PS
NR
,
an
d
T
C
M.
T
h
ese
r
esu
lts
v
alid
ate
th
e
ef
f
ica
cy
o
f
th
e
p
r
o
p
o
s
ed
s
y
s
tem
in
ad
v
an
cin
g
SLT
,
p
a
v
in
g
th
e
way
f
o
r
im
p
r
o
v
ed
c
o
m
m
u
n
icatio
n
f
o
r
I
SL
s
ig
n
er
s
an
d
n
o
n
-
s
ig
n
in
g
in
d
iv
id
u
als.
Fu
tu
r
e
s
tu
d
ies
wi
ll
lo
o
k
at
d
if
f
u
s
io
n
-
b
ased
g
en
er
ativ
e
m
o
d
els
an
d
r
ein
f
o
r
ce
m
en
t
lear
n
in
g
as
a
lo
n
g
-
ter
m
A
I
r
o
a
d
m
ap
f
o
r
c
o
n
tin
u
o
u
s
SL
p
r
o
d
u
ctio
n
.
Di
f
f
u
s
io
n
m
o
d
els
m
ay
o
f
f
er
s
m
o
o
th
er
a
n
d
m
o
r
e
ac
cu
r
ate
v
id
eo
s
y
n
th
esis
f
o
r
c
o
n
tin
u
o
u
s
s
en
ten
ce
-
lev
el
s
ig
n
i
n
g
,
wh
ile
r
ein
f
o
r
ce
m
e
n
t
lear
n
in
g
ag
en
ts
ca
n
u
s
e
g
estu
r
e
co
n
s
is
ten
cy
f
ee
d
b
ac
k
to
d
y
n
am
ically
im
p
r
o
v
e
tem
p
o
r
al
m
o
tio
n
q
u
ality
.
T
h
ese
d
e
v
elo
p
m
en
ts
m
i
g
h
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
I
n
d
ia
n
s
ig
n
la
n
g
u
a
g
e
vid
eo
g
e
n
era
tio
n
u
s
in
g
a
tten
tio
n
-
e
n
h
a
n
ce
d
g
e
n
era
tive
…
(
P
r
a
ch
i P
r
a
mo
d
Wa
g
h
ma
r
e)
3681
lead
to
m
u
ltil
in
g
u
al,
ad
ap
tiv
e,
s
ig
n
er
-
in
d
ep
e
n
d
en
t
SL
co
m
m
u
n
icatio
n
s
y
s
tem
s
th
at
ca
n
s
u
p
p
o
r
t
p
r
ac
tical
ac
ce
s
s
ib
ilit
y
ap
p
licatio
n
s
in
a
v
ar
iety
o
f
lin
g
u
is
tic
co
n
tex
ts
.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
wo
u
ld
lik
e
to
t
h
an
k
f
o
r
Mr
s
.
J
an
h
av
i
an
d
h
e
r
team
f
r
o
m
B
leeT
ec
h
I
n
n
o
v
at
io
n
,
Pu
n
e
,
wh
o
h
elp
ed
u
s
in
cu
r
atin
g
th
e
d
ataset.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
was
p
ar
tially
s
u
p
p
o
r
ted
b
y
MK
SS
S’s
C
u
m
m
in
s
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
f
o
r
W
o
m
en
,
Pu
n
e,
u
n
d
er
th
e
r
esear
ch
in
itia
tio
n
f
u
n
d
o
f
th
e
in
s
titu
te.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Pra
ch
i Pr
am
o
d
W
ag
h
m
ar
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ash
win
i M
an
g
esh
Desh
p
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
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est r
eg
ar
d
in
g
th
e
p
u
b
licatio
n
o
f
th
is
p
ap
e
r
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
p
ar
tial
d
ata
th
at
s
u
p
p
o
r
t
th
e
ex
p
er
im
en
ts
an
d
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
is
o
p
en
ly
av
ailab
le
in
Me
n
d
eley
Data
at
h
ttp
s
://d
ata.
m
en
d
eley
.
c
o
m
/d
atasets
/9
8
m
zk
8
2
wb
b
/
1
.
Oth
er
d
ata
u
s
ed
is
p
r
o
p
er
l
y
cited
.
RE
F
E
R
E
NC
E
S
[
1
]
A
.
D
h
a
n
j
a
l
a
n
d
W
.
S
i
n
g
h
,
“
A
n
a
u
t
o
mat
i
c
c
o
n
v
e
r
s
i
o
n
o
f
P
u
n
j
a
b
i
t
e
x
t
t
o
I
n
d
i
a
n
s
i
g
n
l
a
n
g
u
a
g
e
,
”
EAI
En
d
o
rse
d
T
ra
n
s
a
c
t
i
o
n
s
o
n
S
c
a
l
a
b
l
e
I
n
f
o
rm
a
t
i
o
n
S
y
st
e
m
s
,
v
o
l
.
7
,
n
o
.
2
8
,
J
u
l
.
2
0
1
8
,
d
o
i
:
1
0
.
4
1
0
8
/
e
a
i
.
1
3
-
7
-
2
0
1
8
.
1
6
5
2
7
9
.
[
2
]
J.
P
u
,
W
.
Z
h
o
u
,
H
.
H
u
,
a
n
d
H
.
Li
,
“
B
o
o
s
t
i
n
g
c
o
n
t
i
n
u
o
u
s
s
i
g
n
l
a
n
g
u
a
g
e
r
e
c
o
g
n
i
t
i
o
n
v
i
a
c
r
o
ss
mo
d
a
l
i
t
y
a
u
g
m
e
n
t
a
t
i
o
n
,
”
i
n
2
8
t
h
AC
M
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
M
u
l
t
i
m
e
d
i
a
,
O
c
t
.
2
0
2
0
,
p
p
.
1
4
9
7
–
1
5
0
5
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
9
4
1
7
1
.
3
4
1
3
9
3
1
.
[
3
]
H
.
H
u
,
W
.
Zh
o
u
,
J
.
P
u
,
a
n
d
H
.
Li
,
“
G
l
o
b
a
l
-
l
o
c
a
l
e
n
h
a
n
c
e
me
n
t
n
e
t
w
o
r
k
f
o
r
N
M
F
-
a
w
a
r
e
si
g
n
l
a
n
g
u
a
g
e
r
e
c
o
g
n
i
t
i
o
n
,
”
A
C
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
M
u
l
t
i
m
e
d
i
a
C
o
m
p
u
t
i
n
g
,
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
7
,
n
o
.
3
,
p
p
.
1
–
1
9
,
A
u
g
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
3
6
7
5
4
.
[
4
]
B
.
S
a
u
n
d
e
r
s,
N
.
C
.
C
a
m
g
o
z
,
a
n
d
R
.
B
o
w
d
e
n
,
“
C
o
n
t
i
n
u
o
u
s
3
D
m
u
l
t
i
-
c
h
a
n
n
e
l
si
g
n
l
a
n
g
u
a
g
e
p
r
o
d
u
c
t
i
o
n
v
i
a
p
r
o
g
r
e
ssi
v
e
t
r
a
n
sf
o
r
mers
a
n
d
mi
x
t
u
r
e
d
e
n
s
i
t
y
n
e
t
w
o
r
k
s,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
v
o
l
.
1
2
9
,
n
o
.
7
,
p
p
.
2
1
1
3
–
2
1
3
5
,
J
u
l
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
3
-
021
-
0
1
4
5
7
-
9.
[
5
]
D
.
B
r
a
g
g
e
t
a
l
.
,
“
S
i
g
n
l
a
n
g
u
a
g
e
r
e
c
o
g
n
i
t
i
o
n
,
g
e
n
e
r
a
t
i
o
n
,
a
n
d
t
r
a
n
s
l
a
t
i
o
n
,
”
i
n
2
1
s
t
I
n
t
e
r
n
a
t
i
o
n
a
l
A
C
M
S
I
G
AC
C
E
S
S
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
rs
a
n
d
A
c
c
e
ssi
b
i
l
i
t
y
,
O
c
t
.
2
0
1
9
,
p
p
.
1
6
–
3
1
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
0
8
5
6
1
.
3
3
5
3
7
7
4
.
[
6
]
S
.
S
t
o
l
l
,
S
.
H
a
d
f
i
e
l
d
,
a
n
d
R
.
B
o
w
d
e
n
,
“
S
i
g
n
S
y
n
t
h
:
d
a
t
a
-
d
r
i
v
e
n
si
g
n
l
a
n
g
u
a
g
e
v
i
d
e
o
g
e
n
e
r
a
t
i
o
n
,
”
i
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
–
E
C
C
V
2
0
2
0
Wo
r
k
sh
o
p
s
,
2
0
2
0
,
p
p
.
3
5
3
–
3
7
0
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
6
6
8
2
3
-
5
_
2
1
.
[
7
]
W
.
X
u
,
S
.
K
e
sh
m
i
r
i
,
a
n
d
G
.
W
a
n
g
,
“
A
d
v
e
r
sar
i
a
l
l
y
a
p
p
r
o
x
i
ma
t
e
d
a
u
t
o
e
n
c
o
d
e
r
f
o
r
i
m
a
g
e
g
e
n
e
r
a
t
i
o
n
a
n
d
ma
n
i
p
u
l
a
t
i
o
n
,
”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
2
1
,
n
o
.
9
,
p
p
.
2
3
8
7
–
2
3
9
6
,
S
e
p
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
T
M
M
.
2
0
1
9
.
2
8
9
8
7
7
7
.
[
8
]
S
.
C
h
e
n
a
n
d
W
.
G
u
o
,
“
A
u
t
o
-
e
n
c
o
d
e
r
s
i
n
d
e
e
p
l
e
a
r
n
i
n
g
—
a
r
e
v
i
e
w
w
i
t
h
n
e
w
p
e
r
sp
e
c
t
i
v
e
s,
”
Ma
t
h
e
m
a
t
i
c
s
,
v
o
l
.
1
1
,
n
o
.
8
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
mat
h
1
1
0
8
1
7
7
7
.
[
9
]
S
.
Ju
n
g
,
W
.
S
.
C
h
o
i
,
S
.
C
h
o
i
,
a
n
d
B
.
-
T.
Z
h
a
n
g
,
“
L
a
n
g
u
a
g
e
-
a
g
n
o
s
t
i
c
sem
a
n
t
i
c
c
o
n
si
s
t
e
n
t
t
e
x
t
-
to
-
i
ma
g
e
g
e
n
e
r
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
Wo
rks
h
o
p
o
n
M
u
l
t
i
l
i
n
g
u
a
l
M
u
l
t
i
m
o
d
a
l
L
e
a
r
n
i
n
g
,
2
0
2
2
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
2
.
m
ml
-
1
.
1
.
[
1
0
]
H
.
W
u
e
t
a
l
.
,
“
se
c
G
A
N
:
a
c
y
c
l
e
-
c
o
n
s
i
st
e
n
t
G
A
N
f
o
r
se
c
u
r
e
l
y
-
r
e
c
o
v
e
r
a
b
l
e
v
i
d
e
o
t
r
a
n
sf
o
r
mat
i
o
n
,
”
i
n
2
0
1
9
W
o
rks
h
o
p
o
n
H
o
t
T
o
p
i
c
s
i
n
Vi
d
e
o
An
a
l
y
t
i
c
s
a
n
d
I
n
t
e
l
l
i
g
e
n
t
E
d
g
e
s
,
O
c
t
.
2
0
1
9
,
p
p
.
3
3
–
3
8
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
4
9
6
1
4
.
3
3
5
6
0
2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.