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.
9
,
No
.
2
,
J
u
n
e
2020
,
p
p
.
282
~
2
8
9
I
SS
N:
2
2
5
2
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
9
.i
2
.
p
p
282
-
2
8
9
282
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Tra
ining
config
ur
a
tion a
na
ly
sis
of a co
nv
o
lutiona
l ne
ura
l
netw
o
rk
object
t
r
a
ck
er
f
o
r n
ig
ht
su
rv
eilla
nce applica
tion
Z
u
la
i
kh
a
K
a
di
m
1
,
M
o
hd
Asy
ra
f
Z
ul
kifley
2
,
No
r
Azw
a
n
M
o
ha
m
ed
K
a
m
a
ri
3
1
,
2,
3
De
p
a
rtm
e
n
t
o
f
El
e
c
tri
c
a
l,
El
e
c
tro
n
ic a
n
d
S
y
ste
m
s E
n
g
in
e
e
rin
g
,
F
a
c
u
lt
y
o
f
En
g
in
e
e
rin
g
a
n
d
Bu
il
t
En
v
iro
n
m
e
n
t,
Un
iv
e
rsiti
Ke
b
a
n
g
sa
a
n
M
a
la
y
sia
,
Ba
n
g
i
4
3
6
5
0
,
M
a
lay
sia
1
M
IM
OS
Be
rh
a
d
,
T
e
c
h
n
o
l
o
g
y
P
a
rk
M
a
la
y
sia
,
5
7
0
0
0
B
u
k
it
Ja
li
l,
K
u
a
la L
u
m
p
u
r
,
M
a
lay
sia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
8
,
2
0
20
R
ev
i
s
ed
A
p
r
3
,
2
0
20
A
cc
ep
ted
A
p
r
23
,
2
0
20
A
u
to
m
a
ted
su
rv
e
il
lan
c
e
d
u
rin
g
th
e
n
ig
h
t
is
im
p
o
rtan
t
a
s
it
is
th
e
p
e
rio
d
w
h
e
n
c
rime
s
u
su
a
ll
y
h
a
p
p
e
n
e
d
.
By
p
ro
v
id
in
g
c
o
n
ti
n
u
o
u
s
m
o
n
it
o
ri
n
g
,
c
o
u
p
led
w
it
h
a
re
a
l
-
ti
m
e
a
lert
s
y
ste
m
,
a
p
p
ro
p
riate
a
c
ti
o
n
c
a
n
b
e
tak
e
n
imm
e
d
iate
l
y
i
f
a
c
rime
is
d
e
tec
ted
.
Ho
w
e
v
e
r,
lo
w
li
g
h
ti
n
g
c
o
n
d
it
i
o
n
s
d
u
rin
g
th
e
n
ig
h
t
c
a
n
d
e
g
ra
d
e
th
e
q
u
a
li
ty
o
f
su
rv
e
il
lan
c
e
v
id
e
o
s,
w
h
e
re
th
e
c
a
p
tu
re
d
ima
g
e
s
w
il
l
h
a
v
e
lo
w
c
o
n
tras
t
a
n
d
les
s
d
is
c
rim
in
a
ti
v
e
f
e
a
tu
re
s.
Co
n
se
q
u
e
n
tl
y
,
th
e
se
f
a
c
to
rs
c
o
n
tri
b
u
te
to
th
e
p
ro
b
le
m
o
f
b
a
d
a
p
p
e
a
ra
n
c
e
re
p
re
se
n
tatio
n
o
f
th
e
o
b
jec
t
o
f
in
tere
st
in
t
h
e
trac
k
in
g
a
lg
o
rit
h
m
.
T
h
u
s,
a
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
-
b
a
se
d
o
b
jec
t
trac
k
e
r
f
o
r
n
ig
h
t
su
rv
e
il
lan
c
e
is
p
ro
p
o
se
d
b
y
e
x
p
lo
it
in
g
th
e
d
e
e
p
f
e
a
tu
re
stre
n
g
th
i
n
r
e
p
re
se
n
ti
n
g
o
b
jec
t
f
e
a
tu
re
s
sp
a
ti
a
ll
y
a
n
d
se
m
a
n
ti
c
a
ll
y
.
T
h
e
p
ro
p
o
se
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
tw
o
rk
c
o
n
sists
o
f
six
la
y
e
rs
th
a
t
c
o
n
sist
o
f
th
re
e
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
(CNN
)
a
n
d
t
h
re
e
f
u
ll
y
c
o
n
n
e
c
ted
(
F
C)
lay
e
rs.
T
h
e
n
e
tw
o
rk
w
il
l
b
e
train
e
d
b
y
u
sin
g
a
b
i
n
a
ry
c
las
si
f
ier
a
p
p
ro
a
c
h
o
f
o
b
jec
ts
a
n
d
it
s
b
a
c
k
g
ro
u
n
d
c
las
se
s,
w
h
ich
is
u
p
d
a
te
d
o
n
a
f
ix
e
d
in
terv
a
l
so
th
a
t
i
t
f
u
ll
y
e
n
c
a
p
su
late
s
th
e
c
h
a
n
g
e
s
in
o
b
jec
t
a
p
p
e
a
ra
n
c
e
a
s
it
m
o
v
e
s
in
th
e
s
c
e
n
e
.
T
h
e
a
lg
o
rit
h
m
h
a
s
b
e
e
n
t
e
ste
d
w
it
h
d
if
fe
re
n
t
se
ts
o
f
train
in
g
d
a
ta
c
o
n
f
ig
u
ra
ti
o
n
s
to
f
in
d
th
e
b
e
st
o
p
t
i
m
u
m
o
n
e
s
w
it
h
re
g
a
rd
s
to
V
OT
2
0
1
5
e
v
a
lu
a
ti
o
n
p
ro
t
o
c
o
ls,
tes
ted
o
n
1
4
-
n
ig
h
t
su
rv
e
il
lan
c
e
v
id
e
o
s.
T
h
e
re
su
lt
s
s
h
o
w
th
a
t
t
h
e
c
o
n
f
ig
u
ra
ti
o
n
o
f
a
to
tal
o
f
2
5
0
train
in
g
sa
m
p
les
w
it
h
a
sa
m
p
le
ra
ti
o
o
f
4
:1
b
e
tw
e
e
n
p
o
siti
v
e
a
n
d
n
e
g
a
ti
v
e
d
a
ta
d
e
li
v
e
rs
th
e
b
e
st
p
e
rf
o
rm
a
n
c
e
f
o
r
th
e
se
q
u
e
n
c
e
len
g
th
o
f
[1
,
5
5
0
]
.
I
t
c
a
n
b
e
in
f
e
rre
d
th
a
t
m
o
re
in
f
o
rm
a
ti
o
n
o
n
t
h
e
o
b
jec
t
is
re
q
u
ire
d
c
o
m
p
a
re
d
to
th
e
b
a
c
k
g
ro
u
n
d
,
w
h
e
re
th
e
b
a
c
k
g
ro
u
n
d
m
ig
h
t
b
e
h
o
m
o
g
e
n
e
o
u
s
d
u
e
to
lo
w
li
g
h
ti
n
g
c
o
n
d
it
i
o
n
s.
I
n
c
o
n
c
lu
sio
n
,
t
h
is
a
lg
o
rit
h
m
is
su
it
a
b
le
t
o
b
e
im
p
le
m
e
n
ted
f
o
r
n
ig
h
t
su
rv
e
il
lan
c
e
a
p
p
li
c
a
ti
o
n
.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
o
b
j
ec
t tr
ac
k
in
g
Nig
h
t s
u
r
v
eil
lan
ce
T
r
ac
k
in
g
alg
o
r
it
h
m
T
r
ain
in
g
d
ata
an
al
y
s
is
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
:
Mo
h
d
As
y
r
a
f
Z
u
l
k
i
f
le
y
,
Dep
ar
t
m
en
t o
f
E
lectr
ical,
E
lectr
o
n
ic
an
d
S
y
s
te
m
s
E
n
g
i
n
ee
r
i
n
g
,
Facu
lt
y
o
f
E
n
g
i
n
ee
r
i
n
g
a
n
d
B
u
ilt E
n
v
ir
o
n
m
e
n
t,
Un
i
v
er
s
iti Ke
b
a
n
g
s
aa
n
Ma
la
y
s
ia,
B
an
g
i 4
3
6
5
0
,
Ma
lay
s
ia
.
E
m
ail: a
s
y
r
a
f
.
zu
l
k
i
f
le
y
@
u
k
m
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
Vis
u
a
l
o
b
j
ec
t
tr
ac
k
in
g
h
as
b
ee
n
an
ac
tiv
e
r
esear
ch
ar
ea
in
co
m
p
u
ter
v
is
io
n
an
d
i
m
a
g
e
p
r
o
ce
s
s
in
g
f
ield
s
b
ec
au
s
e
o
f
its
w
id
e
ap
p
licatio
n
i
n
n
u
m
er
o
u
s
r
ea
l
-
wo
r
ld
p
r
o
b
lem
s
i
n
cl
u
d
in
g
v
i
s
u
al
s
u
r
v
eil
lan
ce
[
1
]
,
r
o
b
o
tics
[
2
]
,
h
u
m
a
n
-
c
o
m
p
u
te
r
in
ter
ac
tio
n
s
[
3
]
,
tr
af
f
ic
an
al
y
s
i
s
[
4
]
,
p
h
y
s
io
th
er
ap
y
[
5
]
,
an
d
au
to
n
o
m
o
u
s
v
eh
ic
les
[
6
]
.
I
n
t
h
is
p
ap
er
,
w
e
p
r
o
p
o
s
e
an
o
b
j
ec
t
tr
ac
k
in
g
al
g
o
r
ith
m
b
ased
o
n
a
co
n
v
o
l
u
ti
o
n
al
n
e
u
r
al
n
et
w
o
r
k
(
C
NN)
ap
p
r
o
ac
h
f
o
r
th
e
n
ig
h
t
s
u
r
v
eilla
n
ce
ap
p
licatio
n
.
T
h
e
r
o
le
o
f
th
e
o
b
j
ec
t
tr
ac
k
er
is
to
e
s
ti
m
ate
m
o
v
e
m
e
n
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
Tr
a
in
in
g
co
n
fig
u
r
a
tio
n
a
n
a
lysi
s
o
f a
co
n
vo
l
u
tio
n
a
l
n
e
u
r
a
l n
e
tw
o
r
k
o
b
ject
tr
a
ck
er fo
r
... (
Zu
l
a
ikh
a
K
a
d
im
)
283
tr
aj
ec
to
r
ies
th
r
o
u
g
h
o
u
t
i
m
a
g
e
s
eq
u
en
ce
s
b
y
ass
i
g
n
in
g
a
co
n
s
is
ten
t
lab
el
to
th
e
tr
ac
k
ed
o
b
j
e
ct.
T
h
e
tr
ac
k
er
task
is
also
to
p
r
o
v
id
e
o
b
j
ec
t
-
ce
n
t
r
ic
in
f
o
r
m
atio
n
s
u
c
h
as
o
r
ien
tatio
n
,
ar
ea
o
r
s
h
ap
e
o
f
th
e
tr
ac
k
ed
o
b
j
ec
t.
T
h
is
tr
ac
k
in
g
i
n
f
o
r
m
atio
n
s
er
v
es
a
s
lo
w
-
le
v
el
i
n
p
u
t
f
o
r
h
ig
h
er
-
l
ev
el
ap
p
licatio
n
s
s
u
c
h
as
b
eh
av
io
r
an
al
y
s
i
s
o
f
th
e
o
b
j
ec
t
o
f
i
n
ter
est
an
d
co
n
d
i
tio
n
m
o
n
i
to
r
in
g
o
f
p
h
y
s
io
t
h
er
ap
y
p
atien
t
s
.
T
h
er
e
ar
e
n
u
m
b
er
s
o
f
tr
ac
k
i
n
g
alg
o
r
ith
m
s
th
at
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
in
th
e
l
iter
atu
r
e,
h
o
w
e
v
er
,
m
o
s
t
o
f
t
h
e
m
ar
e
d
ev
elo
p
ed
f
o
r
th
e
ap
p
licatio
n
in
a
b
r
ig
h
t e
n
v
ir
o
n
m
en
t,
w
ith
l
ittl
e
e
m
p
h
asi
s
o
n
d
ar
k
s
u
r
r
o
u
n
d
in
g
s
.
Vid
eo
s
f
r
o
m
n
i
g
h
t
s
ec
u
r
it
y
ca
m
er
as
p
o
s
e
a
m
o
r
e
d
i
f
f
ic
u
lt
c
h
alle
n
g
e
to
th
e
s
ta
n
d
ar
d
o
b
j
ec
t
tr
ac
k
i
n
g
alg
o
r
ith
m
s
.
D
u
e
to
lo
w
li
g
h
ti
n
g
co
n
d
itio
n
s
d
u
r
in
g
th
e
n
ig
h
t
ti
m
e,
t
h
e
q
u
alit
y
o
f
ca
p
t
u
r
ed
i
m
a
g
es
is
n
o
r
m
all
y
n
o
t
th
at
g
o
o
d
w
ith
c
h
ar
ac
ter
is
tics
o
f
lo
w
b
r
ig
h
t
n
ess
,
lo
w
co
n
tr
ast,
an
d
al
m
o
s
t
n
o
d
is
tin
g
u
i
s
h
ab
le
co
lo
r
in
f
o
r
m
atio
n
.
T
h
is
co
n
d
it
io
n
is
ev
en
w
o
r
s
e
i
f
t
h
e
tar
g
et
o
b
j
ec
t is s
m
all
in
s
ize
[
7
]
.
On
e
o
f
th
e
ap
p
r
o
ac
h
es
in
h
a
n
d
lin
g
tr
ac
k
in
g
p
er
f
o
r
m
an
ce
f
o
r
th
e
n
ig
h
t
s
eq
u
e
n
ce
i
s
b
y
e
n
h
an
cin
g
th
e
i
m
a
g
e
q
u
alit
y
f
ir
s
t
b
ef
o
r
e
d
etec
tio
n
an
d
tr
ac
k
in
g
ar
e
d
o
n
e.
T
h
is
en
h
an
ce
m
e
n
t
ca
n
b
e
ac
h
iev
ed
b
y
ap
p
l
y
i
n
g
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
s
u
c
h
as
h
i
s
to
g
r
a
m
eq
u
aliza
t
io
n
,
h
is
to
g
r
a
m
s
p
ec
i
f
icatio
n
,
a
n
d
in
te
n
s
it
y
m
ap
p
in
g
.
H
u
an
g
et
al.
[
8
]
an
al
y
ze
th
e
o
b
j
ec
t'
s
lo
ca
l
co
n
tr
ast
c
h
an
g
es
to
i
m
p
r
o
v
e
o
b
j
ec
t
d
etec
tio
n
ac
cu
r
ac
y
in
t
h
e
n
i
g
h
t
v
id
eo
ap
p
licatio
n
.
L
o
ca
l
co
n
tr
ast
i
s
co
m
p
u
ted
b
y
f
in
d
i
n
g
t
h
e
r
ati
o
b
et
w
ee
n
t
h
e
lo
ca
l
s
ta
n
d
ar
d
d
ev
iatio
n
o
f
i
m
a
g
e
in
te
n
s
it
y
w
it
h
lo
ca
l
m
ea
n
i
n
te
n
s
it
y
,
w
h
ic
h
is
t
h
e
b
asis
f
o
r
Hu
an
g
'
s
C
o
n
tr
ast
C
h
a
n
g
e
(
C
C
)
m
o
d
el.
Ob
j
ec
ts
ar
e
th
en
d
etec
ted
b
y
th
r
es
h
o
ld
in
g
th
e
co
n
tr
ast
ch
a
n
g
e
v
al
u
es
i
n
th
e
s
u
cc
ess
i
v
e
f
r
a
m
es.
T
h
e
co
m
p
u
tat
io
n
s
p
ee
d
is
r
elativ
el
y
f
a
s
t;
h
o
w
e
v
er
,
it
i
s
p
r
o
n
e
to
th
e
p
r
o
b
lem
o
f
s
i
m
ilar
ap
p
ea
r
an
ce
b
et
w
ee
n
th
e
o
b
j
ec
t
an
d
its
s
u
r
r
o
u
n
d
in
g
b
ac
k
g
r
o
u
n
d
.
Hu
an
g
et
al.
[
9
]
f
u
r
th
er
i
m
p
r
o
v
e
th
e
d
etec
tio
n
ac
cu
r
ac
y
b
y
u
til
izin
g
m
o
tio
n
p
r
ed
ictio
n
an
d
s
p
atial
n
ea
r
est
n
eig
h
b
o
r
d
ata
ass
o
ci
atio
n
.
W
an
g
et
al.
[
1
0
]
f
u
r
th
er
i
m
p
r
o
v
e
Hu
a
n
g
’
s
C
C
m
o
d
el
b
y
in
tr
o
d
u
cin
g
a
s
alie
n
t
co
n
tr
ast
c
h
an
g
e
(
S
C
C
)
m
o
d
el,
w
h
ich
r
eq
u
ir
e
s
o
n
li
n
e
lear
n
i
n
g
a
n
d
an
al
y
s
is
o
f
t
h
e
d
etec
ted
o
b
j
ec
t
tr
aj
ec
t
o
r
ies.
T
h
e
w
o
r
k
i
n
[
1
1
]
in
tr
o
d
u
ce
s
ill
u
m
in
at
io
n
i
n
v
ar
ian
t
r
ep
r
esen
t
atio
n
b
y
m
u
ltip
l
y
i
n
g
Sh
a
h
n
o
n
’
s
en
tr
o
p
y
e
s
ti
m
atio
n
w
it
h
t
h
eir
o
w
n
co
n
tr
a
s
t e
s
ti
m
a
tio
n
.
I
n
g
e
n
er
al,
tr
ac
k
in
g
r
eq
u
ir
es
t
h
e
tar
g
et
to
b
e
r
ep
r
esen
ted
b
y
a
m
o
d
el
t
h
at
m
i
g
h
t
i
n
cl
u
d
e
i
n
f
o
r
m
atio
n
ab
o
u
t
th
e
s
h
ap
e
o
r
ap
p
ea
r
an
ce
o
f
th
e
o
b
j
ec
t.
T
h
e
m
o
d
el
w
il
l
b
e
u
s
ed
as
a
r
ef
er
en
ce
i
n
f
i
n
d
i
n
g
t
h
e
m
o
s
t
p
r
o
b
a
b
le
lo
ca
tio
n
o
f
th
e
o
b
j
ec
t
in
t
h
e
n
e
x
t
f
r
a
m
e.
Ob
j
ec
t
ap
p
ea
r
an
ce
ca
n
b
e
r
ep
r
esen
ted
u
s
in
g
g
lo
b
al
o
r
lo
ca
l
f
ea
t
u
r
es,
in
w
h
ich
s
o
m
e
ar
e
m
o
r
e
s
u
itab
le
f
o
r
ce
r
tai
n
tr
a
ck
in
g
c
h
alle
n
g
es
s
u
c
h
as
ill
u
m
i
n
atio
n
v
ar
iatio
n
,
b
ac
k
g
r
o
u
n
d
cl
u
tter
,
an
d
o
cc
l
u
s
io
n
.
Ya
n
g
et
a
l.
[
1
2
]
an
d
L
i
et
al.
[
1
3
]
p
r
o
v
id
e
a
g
o
o
d
o
v
e
r
v
ie
w
o
f
lo
ca
l
a
n
d
g
lo
b
al
f
ea
tu
r
e
r
ep
r
ese
n
tatio
n
s
f
o
r
tr
ac
k
in
g
p
u
r
p
o
s
es
a
n
d
a
s
u
m
m
ar
y
o
f
tar
g
et
a
p
p
ea
r
an
ce
m
o
d
els,
r
esp
ec
tiv
el
y
.
A
g
o
o
d
o
b
j
ec
t
r
e
p
r
esen
tatio
n
m
e
th
o
d
s
h
o
u
ld
b
e
ab
le
to
clea
r
l
y
d
i
s
ti
n
g
u
i
s
h
th
e
tar
g
et
f
r
o
m
o
th
er
b
ac
k
g
r
o
u
n
d
o
b
j
ec
ts
.
T
h
e
au
th
o
r
s
o
f
[
1
4
]
ca
teg
o
r
ize
tr
ac
k
in
g
alg
o
r
ith
m
s
b
ased
o
n
f
ea
tu
r
e
r
ep
r
esen
tatio
n
m
eth
o
d
s
,
in
w
h
ic
h
t
h
e
y
ar
e
d
iv
id
ed
i
n
t
o
t
w
o
g
r
o
u
p
s
n
a
m
el
y
h
a
n
d
cr
af
ted
an
d
d
ee
p
f
ea
t
u
r
es.
E
f
f
ec
ti
v
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
s
h
o
u
ld
b
e
d
i
s
cr
i
m
i
n
ati
v
e
w
h
ile
m
ain
tain
i
n
g
th
e
g
eo
m
etr
ic,
s
tr
u
ct
u
r
al
an
d
s
p
atial
tar
g
e
t
in
f
o
r
m
atio
n
.
S
tr
u
ct
u
r
al,
g
eo
m
etr
ics
an
d
s
p
atial
tar
g
e
t
in
f
o
r
m
atio
n
e
n
co
d
e
th
e
ap
p
ea
r
an
ce
v
ar
iatio
n
,
s
h
ap
es,
an
d
lo
ca
tio
n
o
f
d
if
f
er
en
t
o
b
j
e
ct
p
ar
ts
,
r
e
s
p
ec
tiv
el
y
.
So
m
e
o
f
t
h
e
h
an
d
cr
af
ted
f
ea
t
u
r
es
ca
p
tu
r
e
t
h
is
lo
w
-
le
v
el
in
f
o
r
m
atio
n
b
u
t
e
n
co
d
e
o
n
l
y
a
s
m
a
ll
f
r
ac
tio
n
o
f
s
e
m
a
n
tic
i
n
f
o
r
m
atio
n
.
Dee
p
f
ea
t
u
r
es,
o
n
th
e
o
t
h
er
h
a
n
d
,
ar
e
ab
le
to
en
co
d
e
lo
w
-
lev
e
l
s
p
a
tial
an
d
h
ig
h
-
le
v
el
s
e
m
a
n
tic
in
f
o
r
m
atio
n
w
h
ich
ar
e
e
s
s
e
n
t
ial
co
m
p
o
n
e
n
ts
i
n
lo
ca
tin
g
t
h
e
o
b
j
ec
ts
p
r
ec
is
ely
.
T
h
ese
ab
ilit
ies
m
a
k
e
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
p
o
p
u
lar
in
m
a
n
y
i
m
a
g
e
p
r
o
ce
s
s
in
g
ta
s
k
s
; i
n
cl
u
d
in
g
o
b
j
ec
t d
etec
tio
n
[
1
5
-
1
7
]
,
class
if
i
ca
tio
n
[
1
8
]
,
an
d
tr
ac
k
in
g
.
I
n
[
1
9
]
,
J
o
n
g
et
al.
p
r
o
p
o
s
e
a
C
NN
b
ased
h
u
m
a
n
d
etec
tio
n
,
w
h
ich
s
er
v
es
a
s
an
i
n
p
u
t
to
an
o
b
j
ec
t
tr
ac
k
er
f
o
r
th
e
n
i
g
h
t
ti
m
e
co
n
d
itio
n
s
.
T
h
e
i
n
p
u
t
i
m
a
g
es
ar
e
r
esized
to
1
8
3
x
1
1
9
an
d
its
h
is
to
g
r
a
m
r
ep
r
esen
tatio
n
i
s
eq
u
alize
d
f
ir
s
t
b
ef
o
r
e
p
as
s
in
g
t
h
e
i
m
a
g
es
to
th
e
C
NN
m
o
d
el
to
i
m
p
r
o
v
e
h
u
m
a
n
d
etec
tio
n
ac
cu
r
ac
y
.
I
n
[
2
0
]
,
Ha
m
a
n
d
Han
p
r
o
p
o
s
e
an
o
n
li
n
e
tr
ac
k
in
g
f
r
a
m
e
w
o
r
k
b
ased
o
n
m
u
lti
-
d
o
m
ai
n
r
ep
r
esen
tatio
n
s
.
T
h
eir
n
et
w
o
r
k
ar
ch
itect
u
r
e
co
n
s
is
ts
o
f
m
u
lt
ip
le
s
h
ar
ed
lay
er
s
k
n
o
w
n
a
s
d
o
m
a
in
-
i
n
d
ep
en
d
en
t
la
y
er
s
a
n
d
cla
s
s
i
f
ica
tio
n
la
y
e
r
s
w
h
ich
is
k
n
o
w
n
a
s
d
o
m
a
i
n
-
s
p
ec
if
ic
o
n
e
s
.
Do
m
ai
n
i
n
d
e
p
en
d
en
t
la
y
er
s
ar
e
tr
ain
ed
o
f
f
li
n
e
u
s
i
n
g
m
u
ltip
le
an
n
o
tated
v
id
eo
s
eq
u
en
ce
s
,
w
h
ile
c
lass
if
icatio
n
la
y
er
s
ar
e
tr
ain
ed
s
ep
ar
atel
y
b
ased
o
n
t
h
e
s
p
ec
if
ic
n
e
w
i
m
ag
e
s
eq
u
en
ce
s
.
I
n
[
2
1
]
,
m
u
ltip
le
C
N
Ns
is
m
a
in
ta
in
ed
i
n
a
tr
ee
s
tr
u
ct
u
r
e
to
r
ep
r
e
s
en
t
m
u
lti
-
m
o
d
al
tar
g
et
a
p
p
ea
r
an
ce
.
A
g
e
n
er
al
tr
ac
k
i
n
g
f
r
a
m
e
w
o
r
k
f
o
r
th
er
m
al
i
n
f
r
ar
ed
v
id
eo
s
h
as
al
s
o
b
ee
n
p
r
o
p
o
s
ed
in
[
2
2
]
.
T
h
er
m
al
i
m
a
g
es
h
a
v
e
t
h
e
m
o
s
t
s
i
m
ilar
f
ea
t
u
r
es
to
t
h
e
n
ig
h
t
s
u
r
v
eil
lan
ce
i
m
a
g
es,
s
p
ef
icall
y
i
n
ter
m
s
o
f
lo
w
co
n
tr
ast
in
f
o
r
m
at
io
n
a
n
d
n
eg
li
g
ab
le
tex
t
u
r
es.
M
u
ltip
le
C
NN
s
ap
p
r
o
ac
h
to
m
o
d
el
th
e
tar
g
e
t
ap
p
ea
r
an
ce
in
d
i
f
f
er
en
t
ca
s
es
is
al
s
o
p
r
o
p
o
s
ed
.
Du
r
i
n
g
n
et
w
o
r
k
u
p
d
ates,
p
a
r
en
t
n
o
d
es
w
i
ll
b
e
r
ep
lace
d
b
y
t
h
e
n
e
w
n
o
d
e
s
o
th
at
th
er
e
is
n
o
r
ed
u
n
d
an
c
y
in
th
e
p
o
o
l
o
f
tar
g
et
o
b
j
ec
t
a
p
p
ea
r
an
ce
m
o
d
els.
I
n
[
2
3
]
,
a
Sia
m
ese
ap
p
r
o
ac
h
is
u
tili
ze
d
i
n
w
h
ic
h
p
air
o
f
p
atc
h
es
ar
e
co
m
p
ar
ed
to
f
i
n
d
t
h
e
m
o
s
t
li
k
el
y
lo
ca
tio
n
o
f
th
e
tar
g
e
t o
b
j
ec
t.
I
n
v
ie
w
o
f
t
h
at,
th
is
p
ap
er
p
r
o
p
o
s
es
a
co
n
v
o
lu
tio
n
a
l
n
e
u
r
al
n
et
w
o
r
k
-
b
ased
o
b
j
ec
t
tr
ac
k
er
alg
o
r
ith
m
f
o
r
n
ig
h
t
s
u
r
v
ei
l
lan
ce
ap
p
li
ca
tio
n
b
y
ex
p
lo
iti
n
g
t
h
e
s
tr
en
g
t
h
o
f
d
ee
p
f
ea
tu
r
e
s
in
r
ep
r
esen
tin
g
o
b
j
ec
t
ap
p
ea
r
an
ce
.
Var
io
u
s
tr
ain
in
g
d
ata
co
n
f
ig
u
r
atio
n
s
ar
e
also
ex
a
m
in
ed
to
f
i
n
d
th
e
b
est
s
et
ti
n
g
f
o
r
th
e
p
r
o
p
o
s
ed
tr
ac
k
er
.
T
h
e
p
ap
e
r
is
o
r
g
an
iz
ed
as
f
o
llo
w
s
:
Sectio
n
2
s
u
m
m
ar
iz
es
t
h
e
p
r
o
p
o
s
ed
tr
ac
k
er
alg
o
r
ith
m
f
o
r
n
ig
h
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
2
,
J
u
n
e
20
20
:
282
–
2
8
9
284
s
u
r
v
eilla
n
ce
,
w
h
o
s
e
p
er
f
o
r
m
an
ce
i
s
ev
al
u
ated
an
d
d
is
cu
s
s
ed
in
Sectio
n
3
.
Fin
all
y
,
co
n
cl
u
s
io
n
i
s
g
iv
e
n
i
n
Sectio
n
4
.
2.
CO
NVO
L
U
T
I
O
NA
L
NE
UR
AL
NE
T
WO
RK
O
B
J
E
CT
T
RACK
E
R
T
h
e
p
r
o
p
o
s
ed
co
n
v
o
lu
tio
n
al
o
b
j
ec
t
tr
ac
k
er
alg
o
r
it
h
m
is
b
ased
o
n
th
e
co
n
ce
p
t
o
f
tr
ac
k
in
g
b
y
d
etec
tio
n
,
w
h
er
eb
y
th
e
tr
ac
k
e
d
o
b
j
ec
t
lo
ca
tio
n
is
i
n
f
er
r
ed
b
ased
o
n
th
e
b
est
m
atc
h
ed
s
a
m
p
les
in
ea
c
h
f
r
a
m
e.
I
n
t
h
is
w
o
r
k
,
th
e
d
etec
tio
n
is
ac
h
iev
ed
b
y
co
n
s
tr
u
cti
n
g
a
n
o
b
j
ec
t
m
o
d
el
to
cla
s
s
i
f
y
ei
th
e
r
th
e
in
p
u
t
s
a
m
p
l
e
s
b
elo
n
g
to
th
e
b
ac
k
g
r
o
u
n
d
o
r
o
b
j
ec
t
o
f
in
ter
est.
T
h
e
to
p
f
iv
e
s
a
m
p
les
th
at
ar
e
b
est
cla
s
s
i
f
ie
d
to
b
e
th
e
o
b
j
ec
t
o
f
in
ter
est
ar
e
u
s
ed
to
co
n
s
tr
u
e
th
e
f
i
n
al
o
b
j
ec
t
lo
ca
tio
n
.
T
h
e
o
v
er
all
f
lo
w
o
f
t
h
e
t
r
ac
k
er
alg
o
r
it
h
m
i
s
illu
s
tr
ated
in
Fi
g
u
r
e
1
.
Du
r
in
g
t
h
e
f
ir
s
t
f
r
a
m
e,
t
h
e
m
o
d
el
w
ill
b
e
tr
ai
n
ed
u
s
in
g
s
a
m
p
led
ca
n
d
id
ates
th
at
ar
e
g
en
er
a
ted
b
ased
o
n
th
e
i
n
itia
l
u
s
er
-
s
p
ec
if
ied
o
b
j
ec
t
lo
ca
tio
n
.
N
n
u
m
b
er
o
f
p
o
s
itiv
e
a
n
d
m
n
u
m
b
er
o
f
n
e
g
ati
v
e
s
a
m
p
les
ar
e
g
en
er
ated
r
an
d
o
m
l
y
ar
o
u
n
d
t
h
e
in
itial
o
b
j
ec
t
lo
ca
tio
n
,
g
o
v
er
n
e
d
b
y
a
ce
r
tain
p
er
ce
n
ta
g
e
o
f
o
v
er
lap
p
in
g
ar
ea
b
et
w
ee
n
th
e
s
a
m
p
le
an
d
th
e
i
n
itial
o
b
j
ec
t
lo
ca
tio
n
b
o
x
.
T
h
ese
an
n
o
tated
s
a
m
p
les
ar
e
th
e
n
u
s
ed
to
tr
ain
th
e
f
ir
s
t
o
b
j
ec
t
ap
p
ea
r
an
ce
m
o
d
el.
T
h
e
n
et
w
o
r
k
ar
ch
i
tectu
r
e
co
n
s
is
ts
o
f
t
h
r
ee
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
(
C
NN)
an
d
th
r
ee
f
u
l
l
y
co
n
n
ec
t
ed
(
FC
)
la
y
er
s
.
T
h
is
m
o
d
el
ar
ch
itect
u
r
e
w
ill b
e
f
u
r
th
er
d
is
c
u
s
s
ed
in
s
ec
tio
n
3
.
1
.
Fo
r
th
e
s
u
b
s
eq
u
e
n
t
f
r
a
m
es,
p
r
ev
io
u
s
l
y
k
n
o
w
n
o
b
j
ec
t
lo
ca
t
io
n
w
i
ll
b
e
u
s
ed
as
th
e
p
iv
o
t
p
o
in
t
f
o
r
g
en
er
ati
n
g
s
a
m
p
le
ca
n
d
id
ates.
T
h
is
ap
p
r
o
ac
h
ass
u
m
e
s
th
at
t
h
e
o
b
j
ec
t
in
th
e
cu
r
r
en
t
f
r
a
m
e
d
o
es
n
o
t
m
o
v
e
to
o
f
ar
f
r
o
m
t
h
e
p
r
ev
io
u
s
k
n
o
w
n
l
o
ca
tio
n
.
T
h
ese
s
a
m
p
les
ar
e
t
h
en
m
atc
h
ed
w
it
h
t
h
e
tr
ai
n
ed
m
o
d
el
b
y
cla
s
s
i
f
y
i
n
g
ea
ch
s
a
m
p
le
a
s
b
ac
k
g
r
o
u
n
d
o
r
o
b
j
ec
t
o
f
in
ter
est.
T
h
e
w
ei
g
h
ted
av
er
ag
e
o
f
t
h
e
b
o
u
n
d
ar
y
p
o
in
t
s
o
f
th
e
to
p
-
5
s
a
m
p
les,
w
h
ich
h
a
v
e
b
ee
d
cla
s
s
i
f
ied
as
th
e
o
b
j
ec
t
is
th
e
n
u
s
ed
to
g
en
er
ate
th
e
s
m
o
o
th
ed
o
b
j
ec
t
lo
ca
tio
n
in
t
h
e
cu
r
r
en
t
i
n
p
u
t
f
r
a
m
e.
Fi
n
all
y
,
th
e
m
o
d
el
w
i
ll
b
e
u
p
d
ated
p
er
io
d
ically
to
ca
p
tu
r
e
th
e
o
b
j
ec
t'
s
ap
p
ea
r
an
ce
ch
an
g
es a
s
it
m
o
v
es a
r
o
u
n
d
.
Fig
u
r
e
1
.
A
l
g
o
r
ith
m
f
lo
w
o
f
t
h
e
p
r
o
p
o
s
ed
co
n
v
o
lu
tio
n
al
o
b
j
ec
t tr
ac
k
er
2
.
1
.
Net
wo
rk
a
rc
hite
ct
ure
Fig
u
r
e
2
ill
u
s
tr
ates
th
e
n
et
w
o
r
k
ar
ch
i
tectu
r
e
u
s
ed
in
t
h
e
p
r
o
p
o
s
ed
tr
ac
k
er
.
T
h
e
n
et
w
o
r
k
co
n
s
is
ts
o
f
th
r
ee
C
NN
a
n
d
th
r
ee
F
C
la
y
er
s
.
T
ab
le
1
d
etails
o
u
t
th
e
n
u
m
b
er
o
f
f
ilter
s
,
f
i
lter
s
ize,
s
tr
id
e
an
d
p
ad
d
in
g
o
f
th
e
r
ec
ep
tiv
e
f
ield
in
ea
ch
C
NN
la
y
er
.
T
h
e
s
ize
o
f
th
e
in
p
u
t
i
m
a
g
e
is
s
et
to
7
5
x
7
5
,
th
u
s
th
e
o
u
t
p
u
t
f
r
o
m
t
h
e
co
n
v
3
la
y
er
is
a
f
lat
la
y
er
o
f
5
1
2
x
1
x
1
f
ea
t
u
r
e
m
ap
.
T
h
e
f
i
n
al
o
u
tp
u
t
o
f
th
e
n
et
w
o
r
k
i
s
a
b
in
ar
y
cla
s
s
i
f
icat
io
n
p
r
o
b
a
b
ilit
y
t
h
at
t
h
e
in
p
u
t
i
m
a
g
e
b
elo
n
g
s
to
th
e
o
b
j
ec
t
-
of
-
i
n
ter
est
o
r
b
ac
k
g
r
o
u
n
d
.
T
h
e
w
ei
g
h
ts
o
f
t
h
e
th
r
ee
C
NN
la
y
er
s
ar
e
p
o
r
ted
d
ir
ec
tly
f
r
o
m
tr
ai
n
ed
co
n
v
o
l
u
tio
n
a
l
w
ei
g
h
ts
o
f
t
h
e
f
ir
s
t
3
C
NN
la
y
er
s
o
f
th
e
Vg
g
-
m
m
o
d
el
[
2
4
]
th
at
w
er
e
tr
ain
ed
o
n
t
h
e
I
m
a
g
eNe
t
I
L
SV
R
C
-
2
0
1
2
d
ataset
[
2
5
]
.
Vg
g
-
m
m
o
d
el
w
a
s
f
ir
s
t
i
n
tr
o
d
u
ce
d
b
y
[
2
4
]
as
C
NN
-
m
t
h
at
co
n
s
is
ts
o
f
f
i
v
e
C
NN
la
y
er
s
an
d
th
r
e
e
FC
la
y
er
s
a
s
s
h
o
w
n
i
n
T
ab
le
2
.
T
h
u
s
,
th
er
e
is
a
clea
r
d
if
f
er
en
ce
b
et
w
ee
n
o
u
r
s
an
d
t
h
e
V
g
g
-
m
n
et
w
o
r
k
i
n
ter
m
s
o
f
th
e
n
u
m
b
er
o
f
C
N
N
la
y
er
s
(
3
v
s
.
5
)
.
A
p
ar
t
f
r
o
m
t
h
at,
th
e
co
n
v
o
lu
tio
n
al
s
t
r
id
e
an
d
s
p
atial
p
ad
d
in
g
in
co
n
v
2
an
d
co
n
v
3
la
y
er
s
al
s
o
d
if
f
er
en
t
co
m
p
ar
ed
to
Vg
g
-
m
,
as
w
el
l a
s
les
s
d
i
m
e
n
s
io
n
alit
y
f
o
r
f
c1
to
f
c3
(
5
1
2
v
s
.
4
0
9
6
)
.
Fig
u
r
e
2
.
Net
w
o
r
k
ar
c
h
itect
u
r
e
o
f
th
e
co
n
v
o
lu
t
io
n
al
o
b
j
ec
t t
r
ac
k
er
w
it
h
t
h
r
ee
C
NN
la
y
er
s
an
d
th
r
ee
FC
la
y
er
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
Tr
a
in
in
g
co
n
fig
u
r
a
tio
n
a
n
a
lysi
s
o
f a
co
n
vo
l
u
tio
n
a
l
n
e
u
r
a
l n
e
tw
o
r
k
o
b
ject
tr
a
ck
er fo
r
... (
Zu
l
a
ikh
a
K
a
d
im
)
285
T
ab
le
1
.
Nu
m
b
er
o
f
f
ilter
s
,
f
ilt
er
s
ize,
s
tr
id
e
an
d
p
ad
d
in
g
o
f
r
ec
ep
tiv
e
f
ield
s
in
ea
c
h
la
y
er
o
f
p
r
o
p
o
s
ed
m
o
d
el
L
a
y
e
r
C
o
n
v
1
P
o
o
l
1
C
o
n
v
2
P
o
o
l
2
C
o
n
v
3
N
u
m
b
e
r
o
f
f
i
l
t
e
rs
96
96
2
5
6
2
5
6
5
1
2
Fi
l
t
e
r
si
ze
7
x
7
3
x
3
5
x
5
3
x
3
3
x
3
S
t
r
i
d
e
2
2
2
2
2
Pa
d
d
i
n
g
0
0
0
0
0
T
ab
le
2
.
VGG
-
M
ar
ch
itect
u
r
e
as p
r
o
p
o
s
ed
in
[
2
2
]
Arc
h
C
o
n
v
1
C
o
n
v
2
C
o
n
v
3
C
o
n
v
4
C
o
n
v
5
F
u
l
l
6
F
u
l
l
7
F
u
l
l
8
C
N
N
-
M
9
6
x
7
x
7
st
.
2
,
p
a
d
0
L
R
N
,
x
2
p
o
o
l
2
5
6
x
5
x
5
st
.
2
,
p
a
d
1
L
R
N
,
x
2
p
o
o
l
5
1
2
x
3
x
3
st
.
1
,
p
a
d
1
-
5
1
2
x
3
x
3
st
.
1
,
p
a
d
1
-
5
1
2
x
3
x
3
st
.
1
,
p
a
d
1
,
x
2
p
o
o
l
4
0
9
6
d
r
o
p
o
u
t
4
0
9
6
d
r
o
p
o
u
t
1
0
0
0
S
o
f
t
max
2
.
2
.
Sa
m
p
les g
ener
a
t
io
n
T
h
e
p
r
o
p
o
s
ed
tr
ac
k
er
alg
o
r
ith
m
tr
ies
to
f
i
n
d
th
e
b
est
ca
n
d
id
ate
r
ep
r
esen
tatio
n
s
o
f
t
h
e
tr
ac
k
ed
o
b
j
ec
t
f
r
o
m
a
s
et
o
f
s
a
m
p
les
u
s
i
n
g
a
d
ee
p
-
lear
n
i
n
g
m
o
d
el
t
h
at
a
cts
as
a
b
in
ar
y
clas
s
i
f
ier
.
P
o
s
iti
v
e
an
d
n
e
g
ati
v
e
ca
n
d
id
ates r
ef
er
to
t
h
e
s
u
b
-
i
m
ag
e
t
h
at
co
r
r
esp
o
n
d
s
to
a
tr
ac
k
ed
o
b
j
ec
t o
r
b
ac
k
g
r
o
u
n
d
ar
ea
,
r
esp
ec
tiv
el
y
.
T
h
ese
s
a
m
p
les
w
i
ll
th
e
n
s
er
v
e
as
th
e
tr
ain
in
g
d
ata
u
s
ed
to
tr
ain
t
h
e
n
et
w
o
r
k
in
a
s
u
p
er
v
i
s
ed
m
a
n
n
er
.
Fo
r
th
e
i
n
itial
tr
ain
i
n
g
p
h
a
s
e
in
t
h
e
f
ir
s
t
f
r
a
m
e,
m
o
r
e
s
a
m
p
le
s
ar
e
g
en
er
at
ed
s
in
ce
th
e
ac
tu
a
l
o
b
j
ec
t
l
o
ca
tio
n
is
k
n
o
w
n
to
b
e
tr
u
e.
Ho
w
e
v
er
,
d
u
r
i
n
g
o
n
li
n
e
n
et
w
o
r
k
u
p
d
ates,
a
les
s
er
n
u
m
b
er
o
f
s
a
m
p
le
s
ar
e
cr
ea
ted
to
r
ed
u
ce
th
e
co
m
p
u
tatio
n
b
u
r
d
en
,
as
w
ell
a
s
to
m
i
n
i
m
ize
th
e
p
o
s
s
ib
ilit
y
o
f
b
ac
k
g
r
o
u
n
d
i
n
clu
s
io
n
i
n
t
h
e
m
o
d
el.
T
o
g
en
er
ate
th
e
s
a
m
p
les,
o
b
j
e
ct
lo
ca
tio
n
in
th
e
p
r
ev
io
u
s
f
r
a
m
e
is
u
s
ed
a
s
th
e
p
i
v
o
t
p
o
in
t
to
s
p
a
w
n
p
o
s
itiv
e
an
d
n
e
g
ati
v
e
ca
n
d
id
ate
lo
ca
tio
n
s
.
T
o
en
s
u
r
e
th
e
v
ar
iab
ilit
y
o
f
t
h
e
ca
n
d
id
ate
s
,
th
e
s
a
m
p
les
ar
e
s
p
a
w
n
ed
r
an
d
o
m
l
y
ac
co
r
d
in
g
to
Gau
s
s
ian
d
is
tr
ib
u
tio
n
,
w
h
er
eb
y
th
e
s
a
m
p
les
ar
e
co
n
ce
n
tr
ated
clo
s
er
to
th
e
p
iv
o
t
p
o
in
t.
T
h
e
g
en
er
ated
p
o
s
iti
v
e
s
a
m
p
le
s
h
o
u
ld
h
a
v
e
at
least
8
0
%
o
v
er
lap
p
in
g
ar
ea
s
w
it
h
t
h
e
p
r
ev
io
u
s
l
y
k
n
o
w
n
o
b
j
ec
t
ar
ea
,
w
h
i
le
n
e
g
ativ
e
s
a
m
p
les
s
h
o
u
ld
also
h
av
e
an
o
v
er
lap
p
in
g
ar
ea
w
it
h
at
m
o
s
t
2
0
%.
Sa
m
p
le
s
ize
i
s
o
n
e
o
f
th
e
m
o
s
t
i
m
p
o
r
tan
t
f
ac
to
r
s
i
n
tr
ain
i
n
g
a
d
ee
p
-
lear
n
i
n
g
m
o
d
el.
A
h
ig
h
er
n
u
m
b
er
o
f
tr
ai
n
i
n
g
s
a
m
p
les
i
s
b
etter
f
o
r
g
en
er
aliz
in
g
t
h
e
m
o
d
el.
T
h
u
s
,
in
t
h
is
w
o
r
k
,
t
h
e
s
elec
tio
n
o
f
t
h
e
n
u
m
b
er
o
f
s
a
m
p
les
a
n
d
th
e
r
atio
b
et
w
ee
n
p
o
s
iti
v
e
a
n
d
n
eg
a
tiv
e
s
a
m
p
les
ar
e
a
n
al
y
ze
d
to
d
eter
m
i
n
e
t
h
e
b
es
t
co
m
b
in
at
io
n
f
o
r
t
h
e
p
r
o
p
o
s
ed
o
b
j
ec
t tr
ac
k
er
alg
o
r
ith
m
.
2
.
3
.
Net
wo
rk
lea
rning
a
nd
u
pd
a
t
e
T
r
ain
in
g
o
f
t
h
e
n
et
w
o
r
k
i
n
t
h
e
f
ir
s
t
f
r
a
m
e
w
il
l
d
ictate
h
o
w
w
ell
t
h
e
tr
ac
k
er
w
ill
p
er
f
o
r
m
i
n
t
h
e
s
u
b
s
eq
u
en
t
f
r
a
m
es
.
T
h
e
t
h
r
ee
C
NN
la
y
er
’
s
w
ei
g
h
ts
ar
e
tr
an
s
f
er
r
ed
f
r
o
m
th
e
p
r
e
-
tr
ain
ed
w
e
ig
h
ts
o
f
t
h
e
f
ir
s
t
3
C
NN
la
y
er
s
i
n
t
h
e
Vg
g
-
m
n
et
w
o
r
k
,
w
h
ile
t
h
e
w
ei
g
h
ts
a
n
d
b
ias
es i
n
t
h
e
F
C
la
y
er
s
ar
e
in
it
i
alize
d
r
an
d
o
m
l
y
a
n
d
f
i
x
ed
to
0
.
0
5
,
r
esp
ec
tiv
el
y
.
O
n
l
y
th
e
s
e
F
C
la
y
er
s
w
il
l
g
o
t
h
r
o
u
g
h
lear
n
i
n
g
s
ta
g
e
u
s
i
n
g
p
o
s
iti
v
e
an
d
n
eg
a
tiv
e
s
a
m
p
les
g
e
n
er
ated
f
r
o
m
o
u
r
s
p
ec
if
ic
tes
tin
g
d
ata.
C
h
o
o
s
i
n
g
a
lear
n
in
g
r
ate
f
o
r
o
p
ti
m
al
lear
n
i
n
g
is
a
ch
allen
g
i
n
g
task
as
it
m
a
y
ca
u
s
e
s
lo
w
co
n
v
er
g
en
ce
o
r
f
r
eq
u
en
t
f
lu
ct
u
atio
n
o
f
t
h
e
lo
s
s
f
u
n
ctio
n
,
w
h
ic
h
w
il
l
lead
to
th
e
m
o
d
el
d
i
v
er
g
e
n
t.
T
h
u
s
,
in
t
h
i
s
w
o
r
k
,
w
e
h
av
e
s
elec
ted
A
d
a
m
o
p
ti
m
iz
er
[
2
6
]
as
th
e
b
ac
k
p
r
o
p
ag
ati
o
n
o
p
ti
m
iza
tio
n
alg
o
r
it
h
m
.
Ad
am
i
s
a
n
ad
ap
ti
v
e
m
o
m
e
n
t
est
i
m
a
tio
n
t
h
at
co
m
p
u
tes
d
if
f
er
en
t
lear
n
i
n
g
r
ates
f
o
r
d
if
f
er
e
n
t
p
ar
a
m
eter
s
(
i.e
.
m
o
d
el
w
ei
g
h
ts
a
n
d
b
iases
)
b
y
u
s
in
g
th
e
esti
m
ate
s
o
f
th
e
ir
f
ir
s
t
a
n
d
s
ec
o
n
d
-
o
r
d
er
m
o
m
e
n
ts
o
f
th
e
g
r
ad
ie
n
t.
T
h
e
f
i
n
al
m
o
d
el
p
ar
am
eter
u
p
d
ates
w
ill
b
e
b
a
s
ed
o
n
t
h
e
n
e
w
l
y
ad
j
u
s
ted
le
ar
n
in
g
r
ate,
w
h
ic
h
eq
u
als
to
in
itial
lear
n
i
n
g
r
ate
m
u
ltip
lies
w
it
h
th
e
r
at
io
o
f
its
f
ir
s
t
a
n
d
s
ec
o
n
d
-
o
r
d
er
m
o
m
e
n
ts
o
f
t
h
e
g
r
ad
ien
ts
.
T
h
e
i
n
itial
lear
n
in
g
r
ate
i
s
s
et
as
0
.
0
0
0
7
5
b
ased
o
n
th
e
f
i
n
d
in
g
i
n
[
2
7
]
,
an
d
A
d
a
m
’
s
h
y
p
e
r
p
ar
am
eter
s
e
x
p
o
n
en
t
ial
d
ec
ay
r
ate
f
o
r
th
e
f
ir
s
t
-
o
r
d
er
(
β1
)
,
ex
p
o
n
en
tial
d
ec
a
y
r
ate
f
o
r
th
e
s
ec
o
n
d
-
o
r
d
er
(
β2
)
an
d
s
m
all
v
al
u
es
co
n
s
ta
n
ts
to
p
r
ev
en
t
d
iv
is
io
n
b
y
ze
r
o
(
ε)
,
w
h
ic
h
ar
e
s
et
to
0
.
9
,
0
.
9
9
9
an
d
1
e
-
0
8
r
esp
ec
tiv
el
y
.
A
d
a
m
h
as
b
ee
n
g
ai
n
in
g
p
o
p
u
lar
it
y
s
i
n
ce
it
s
in
tr
o
d
u
c
tio
n
i
n
2
0
1
5
,
w
h
er
e
it
h
as b
ee
n
w
id
el
y
u
s
ed
in
o
th
er
s
n
et
w
o
r
k
lear
n
i
n
g
[
2
8
]
.
I
t
is
i
m
p
o
r
tan
t
to
u
p
d
ate
th
e
n
et
w
o
r
k
to
en
s
u
r
e
th
at
t
h
e
m
o
d
el
ca
p
t
u
r
es
th
e
c
h
an
g
es
in
o
b
j
ec
t
ap
p
ea
r
an
ce
as
it
m
o
v
e
s
ar
o
u
n
d
u
n
d
er
d
i
f
f
er
e
n
t
li
g
h
ti
n
g
co
n
d
itio
n
s
a
n
d
b
ac
k
g
r
o
u
n
d
s
ce
n
e
s
.
I
n
t
h
i
s
w
o
r
k
,
o
u
r
m
o
d
el
w
ill
b
e
u
p
d
ated
at
a
f
i
x
ed
in
ter
v
al
o
f
e
v
er
y
1
0
f
r
a
m
es
.
P
o
s
itiv
e
a
n
d
n
e
g
ati
v
e
s
a
m
p
le
s
t
h
at
ar
e
g
en
er
ated
in
th
e
p
er
io
d
o
f
1
0
f
r
a
m
es
w
il
l
b
e
ac
cu
m
u
lated
an
d
u
s
e
d
to
r
etr
ain
th
e
m
o
d
el.
Ho
w
ev
er
,
th
e
tr
ain
in
g
is
co
n
s
id
er
ed
as
w
ea
k
s
u
p
er
v
is
i
o
n
as
th
e
tr
u
e
o
b
j
ec
t
lo
ca
tio
n
i
s
n
o
t
k
n
o
w
n
a
n
d
it
m
a
y
l
ea
d
to
b
ac
k
g
r
o
u
n
d
in
cl
u
s
io
n
in
t
h
e
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
2
,
J
u
n
e
20
20
:
282
–
2
8
9
286
3
R
E
SU
L
T
S
A
ND
D
I
SCU
SS
I
O
N
3
.
1
.
E
x
peri
m
ent
a
l set
up
14
-
n
i
g
h
t
s
u
r
v
ei
llan
ce
v
id
eo
s
ar
e
u
s
ed
to
e
v
al
u
ate
t
h
e
p
r
o
p
o
s
ed
o
b
j
ec
t
tr
ac
k
er
alg
o
r
ith
m
,
w
h
ic
h
h
as
b
ee
n
r
ec
o
r
d
ed
f
r
o
m
th
r
ee
d
if
f
er
en
t
s
u
r
v
e
illa
n
ce
ca
m
er
as
th
at
co
v
er
ch
al
len
g
i
n
g
n
i
g
h
t
s
ce
n
ar
io
s
o
f
lo
w
lig
h
ti
n
g
,
lo
w
co
n
tr
ast
b
et
w
ee
n
o
b
j
ec
t
an
d
b
ac
k
g
r
o
u
n
d
,
s
m
all
o
b
j
ec
t
s
ize,
o
b
j
ec
t
o
cc
lu
s
io
n
,
an
d
m
o
v
e
-
s
to
p
-
m
o
v
e
o
b
j
ec
t
m
o
v
e
m
e
n
t.
So
m
e
s
a
m
p
le
i
m
ag
es
o
f
t
h
e
s
e
s
ce
n
ar
io
s
ar
e
s
h
o
w
n
in
Fi
g
u
r
e
3
.
T
h
e
r
eso
lu
tio
n
o
f
th
e
v
id
eo
is
o
n
l
y
3
5
2
x
2
8
8
w
it
h
t
h
e
o
b
j
ec
t
o
f
in
ter
est
s
ize
o
f
ap
p
r
o
x
i
m
atel
y
3
0
x
7
0
p
ix
els.
Vid
eo
len
g
t
h
v
ar
ies
f
r
o
m
3
3
t
o
5
5
0
f
r
am
e
s
ea
ch
.
Gr
o
u
n
d
tr
u
th
s
ar
e
g
en
er
ated
b
y
a
n
e
x
p
er
t
in
co
m
p
u
ter
v
is
io
n
u
s
i
n
g
an
a
n
n
o
tatio
n
to
o
l
to
d
r
a
w
t
h
e
b
o
u
n
d
in
g
b
o
x
t
h
at
s
u
r
r
o
u
n
d
s
t
h
e
o
b
j
ec
t
in
ea
ch
f
r
a
m
e
f
o
r
ea
ch
v
id
eo
.
T
h
e
tr
ac
k
er
co
d
e
is
d
e
v
elo
p
ed
in
P
y
th
o
n
w
i
th
T
en
s
o
r
f
lo
w
a
s
t
h
e
m
ai
n
lib
r
ar
y
to
p
er
f
o
r
m
m
o
d
el
tr
ain
in
g
a
n
d
test
i
n
g
.
Sev
e
n
d
i
f
f
er
en
t
co
m
b
i
n
atio
n
o
f
tr
ain
i
n
g
s
a
m
p
le
s
s
iz
e
an
d
r
atio
s
ar
e
co
n
f
i
g
u
r
ed
f
o
r
tr
ac
k
er
ev
a
lu
at
io
n
as lis
ted
in
T
ab
le
3
.
Fig
u
r
e
3
.
Sa
m
p
le
i
m
ag
e
s
w
it
h
ch
alle
n
g
i
n
g
n
i
g
h
t sce
n
ar
io
in
t
esti
n
g
d
ataset
T
ab
le
3
.
T
r
ain
in
g
s
a
m
p
le
co
n
f
ig
u
r
at
io
n
f
o
r
tr
ac
k
er
e
v
alu
at
io
n
C
o
n
f
i
g
u
r
a
t
i
o
n
1
:
1
1
:
2
1
:3
1
:
4
2
:
1
3
:
1
4
:
1
P
o
si
t
i
v
e
sam
p
l
e
si
z
e
1
0
0
50
50
50
1
0
0
1
5
0
2
0
0
N
e
g
a
t
i
v
e
samp
l
e
si
z
e
1
0
0
1
0
0
1
5
0
2
0
0
50
50
50
T
o
t
a
l
samp
l
e
si
z
e
2
0
0
1
5
0
2
0
0
2
5
0
1
5
0
2
0
0
2
5
0
3
.
2
.
P
er
f
o
rm
a
nce
ev
a
lua
t
io
n
m
ea
s
ure
T
h
r
ee
VOT
2
0
1
5
[
29
]
(
Vis
u
al
Ob
j
ec
t
T
r
ac
k
in
g
)
ev
a
lu
at
io
n
m
etr
ics
ar
e
u
s
ed
to
q
u
an
ti
f
y
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
o
u
r
tr
ac
k
er
.
T
h
ese
m
etr
ics
ar
e
ac
cu
r
ac
y
(
A
cc
)
,
r
o
b
u
s
tn
e
s
s
(
Ro
)
an
d
ex
p
ec
ted
ar
ea
o
v
er
lap
(
EAO
)
.
A
cc
u
r
ac
y
an
d
r
o
b
u
s
tn
ess
r
eq
u
ir
e
t
h
e
tr
ac
k
er
to
b
e
r
e
-
i
n
itialized
o
n
ce
it
d
r
if
t
s
o
f
f
t
h
e
tar
g
et.
Acc
u
r
ac
y
m
ea
s
u
r
es
h
o
w
w
e
ll
th
e
tr
ac
k
e
d
b
o
u
n
d
in
g
b
o
x
r
elati
v
e
to
th
e
g
r
o
u
n
d
tr
u
t
h
b
o
x
b
y
co
m
p
u
t
in
g
t
h
e
i
n
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
OU
)
.
T
h
e
h
ig
h
e
r
th
e
I
OU
,
th
e
b
etter
th
e
tr
ac
k
in
g
ac
cu
r
ac
y
is
.
On
t
h
e
o
th
er
h
an
d
,
r
o
b
u
s
tn
es
s
m
ea
s
u
r
es
t
h
e
n
u
m
b
er
o
f
tr
ac
k
i
n
g
f
ail
u
r
es
in
a
v
id
eo
,
w
h
ic
h
i
s
tr
ig
g
er
ed
w
h
e
n
ze
r
o
I
OU
o
c
cu
r
s
.
T
o
r
ed
u
ce
th
e
b
ias
in
r
o
b
u
s
tn
e
s
s
m
ea
s
u
r
e
m
e
n
t,
th
e
tr
ac
k
er
is
r
e
-
i
n
itialized
f
i
v
e
f
r
a
m
e
s
af
ter
th
e
f
ail
u
r
e,
w
h
ile
to
r
ed
u
ce
b
ias
in
ac
cu
r
ac
y
ca
lc
u
lat
io
n
,
th
e
a
cc
u
r
ac
y
v
a
lu
e
s
f
r
o
m
t
h
e
f
ir
s
t
1
0
f
r
am
e
s
af
ter
th
e
r
e
-
in
it
ial
izatio
n
p
r
o
ce
s
s
ar
e
ig
n
o
r
ed
f
r
o
m
o
v
er
all
p
er
f
o
r
m
a
n
ce
co
m
p
u
ta
tio
n
[
30
]
.
EAO
d
o
es
n
o
t
r
eq
u
ir
e
r
e
-
i
n
itia
lizatio
n
o
f
t
h
e
tr
ac
k
er
a
n
d
it
i
s
u
s
ed
to
r
an
k
t
h
e
tr
ac
k
er
.
I
t
a
v
er
ag
es
t
h
e
I
OU
o
v
er
a
r
an
g
e
o
f
f
r
a
m
es
b
et
w
ee
n
lo
w
er
(
N
lo
)
an
d
u
p
p
er
li
m
i
t
(
N
h
i
)
o
f
th
e
test
i
n
g
s
eq
u
en
ce
.
I
n
t
h
is
w
o
r
k
,
th
e
lo
w
er
an
d
u
p
p
er
li
m
it i
s
b
ased
o
n
th
e
r
an
g
e
o
f
[
1
,
5
5
0
]
.
T
h
e
A
cc
,
Ro
an
d
EAO
ca
lcu
lati
o
n
ar
e
as f
o
llo
w
s
:
A
cc
u
r
ac
y
,
=
1
∑
,
∩
,
,
∪
,
=
1
(
1
)
w
h
er
e
d
en
o
tes
th
e
n
u
m
b
er
o
f
f
r
a
m
es
in
t
h
e
test
v
id
eo
,
w
h
il
e
,
an
d
,
ar
e
th
e
b
o
u
n
d
in
g
b
o
x
e
s
o
f
o
b
j
ec
t in
f
r
a
m
e
i
f
r
o
m
t
h
e
tr
ac
k
er
o
u
tp
u
t a
n
d
g
r
o
u
n
d
tr
u
t
h
,
r
esp
ec
tiv
el
y
.
R
o
b
u
s
t
n
es
s
,
=
∑
=
1
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
Tr
a
in
in
g
co
n
fig
u
r
a
tio
n
a
n
a
lysi
s
o
f a
co
n
vo
l
u
tio
n
a
l
n
e
u
r
a
l n
e
tw
o
r
k
o
b
ject
tr
a
ck
er fo
r
... (
Zu
l
a
ikh
a
K
a
d
im
)
287
w
h
er
e
=
{
0
>
0
1
≤
0
E
x
p
ec
ted
av
er
ag
e
o
v
er
lap
,
=
1
ℎ
−
∑
=
:
ℎ
(
3
)
3
.
3
.
Resul
t
s
A
cc
u
r
ac
y
a
n
d
r
o
b
u
s
tn
es
s
v
al
u
es
f
o
r
d
if
f
er
e
n
t
tr
ain
i
n
g
co
n
f
ig
u
r
atio
n
s
ar
e
s
u
m
m
ar
ized
in
T
ab
le
4
an
d
T
ab
le
5
,
r
esp
ec
tiv
el
y
.
Fro
m
T
ab
le
4
,
th
er
e
is
n
o
s
in
g
le
co
n
f
ig
u
r
atio
n
w
o
r
k
s
th
e
b
est
f
o
r
all
test
s
eq
u
en
ce
s
.
A
ll
d
if
f
er
en
t
co
n
f
ig
u
r
atio
n
s
p
er
f
o
r
m
al
m
o
s
t
r
elati
v
el
y
t
h
e
s
a
m
e
i
n
ter
m
s
o
f
ac
cu
r
ac
y
,
h
o
w
e
v
er
in
ter
m
s
o
f
r
o
b
u
s
tn
es
s
,
co
n
f
ig
u
r
atio
n
1
:4
s
h
o
w
s
s
i
g
n
i
f
ica
n
tl
y
b
e
tter
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
to
th
e
o
th
er
s
.
Fo
r
n
o
-
r
ei
n
itial
izatio
n
p
r
o
to
co
l
ev
al
u
atio
n
,
th
e
EAO
c
u
r
v
e
is
s
h
o
w
n
in
Fi
g
u
r
e
4
.
T
h
e
c
u
r
v
e
s
h
o
w
s
t
h
at
t
h
e
co
n
f
i
g
u
r
atio
n
o
f
4
:1
w
o
r
k
s
c
o
n
s
is
ten
tl
y
b
etter
t
h
an
o
t
h
er
co
n
f
i
g
u
r
atio
n
s
ap
p
r
o
x
i
m
atel
y
af
ter
th
e
f
ir
s
t
1
5
0
f
r
a
m
e
s
,
f
o
llo
w
ed
b
y
co
n
f
i
g
u
r
atio
n
o
f
1
:3
,
t
h
u
s
m
a
k
es
b
o
th
co
n
f
i
g
u
r
atio
n
s
r
a
n
k
ed
a
s
t
h
e
to
p
t
w
o
a
s
d
em
o
n
s
tr
ated
i
n
T
ab
le
6
.
T
h
is
e
x
p
lai
n
s
th
at
d
if
f
er
en
t
tr
ain
i
n
g
s
a
m
p
le
s
ize
s
a
n
d
r
ati
o
s
af
f
ec
t
tr
ac
k
i
n
g
p
er
f
o
r
m
a
n
ce
.
T
h
e
tr
ac
k
er
p
er
f
o
r
m
s
b
etter
w
h
e
n
th
er
e
a
r
e
m
o
r
e
p
o
s
itiv
e
s
a
m
p
le
s
u
s
ed
to
r
ep
r
esen
t
th
e
o
b
j
ec
t
co
m
p
ar
ed
to
th
e
n
e
g
ati
v
e
s
a
m
p
les.
T
h
is
in
d
icate
s
t
h
at
alt
h
o
u
g
h
t
h
e
o
b
j
ec
t
-
b
ac
k
g
r
o
u
n
d
co
n
tr
ast
i
s
lo
w
,
it
al
s
o
lo
o
k
s
h
o
m
o
g
e
n
eo
u
s
d
u
e
to
lo
w
li
g
h
ti
n
g
,
t
h
u
s
a
lo
w
er
n
u
m
b
er
o
f
s
a
m
p
les ar
e
s
u
f
f
ic
ien
t to
r
ep
r
esen
t th
e
m
.
T
ab
le
4
.
A
cc
u
r
ac
y
r
es
u
lt
s
o
f
d
if
f
er
en
t tr
ai
n
i
n
g
s
a
m
p
le
co
n
f
i
g
u
r
atio
n
p
er
ea
ch
test
s
eq
u
e
n
ce
No
M
e
t
h
o
d
N
u
mb
e
r
o
f
f
r
a
me
s
1
:
1
1
:
2
1
:
3
1
:
4
2
:
1
3
:
1
4
:
1
1
C
a
m
0
1
-
v
i
d
e
o
0
1
2
7
1
0
.
6
4
9
0
.
6
1
8
0
.
5
6
5
0
.
5
8
0
0
.
1
5
9
0
.
1
6
5
0
.
6
5
1
2
C
a
m
0
1
-
v
i
d
e
o
0
2
4
5
4
0
.
5
6
2
0
.
4
5
2
0
.
5
2
9
0
.
5
9
5
0
.
5
8
5
0
.
5
7
6
0
.
5
2
8
3
C
a
m
0
1
-
v
i
d
e
o
0
3
70
0
.
1
4
2
0
.
2
1
3
0
.
0
8
8
0
.
2
8
4
0
.
1
4
3
0
.
1
5
9
0
.
0
6
5
4
C
a
m
0
1
-
v
i
d
e
o
0
4
1
3
0
0
.
6
5
0
0
.
6
2
2
0
.
5
9
9
0
.
5
7
3
0
.
6
0
4
0
.
5
8
3
0
.
5
9
8
5
C
a
m
0
1
-
v
i
d
e
o
0
5
33
0
.
6
8
0
0
.
7
5
1
0
.
6
7
3
0
.
6
1
6
0
.
7
7
3
0
.
5
9
3
0
.
7
7
7
6
C
a
m
0
1
-
v
i
d
e
o
0
6
2
2
4
0
.
5
6
0
0
.
6
0
5
0
.
6
1
9
0
.
5
3
8
0
.
6
2
0
0
.
6
1
4
0
.
6
2
7
7
C
a
m
0
1
-
v
i
d
e
o
0
7
4
6
0
0
.
1
5
3
0
.
1
0
1
0
.
4
0
1
0
.
1
2
1
0
.
4
4
5
0
.
1
0
8
0
.
6
3
7
8
C
a
m
0
1
-
v
i
d
e
o
0
8
1
2
4
0
.
5
4
8
0
.
3
1
4
0
.
5
1
7
0
.
5
6
2
0
.
4
2
0
0
.
2
3
7
0
.
2
0
8
9
C
a
m
0
2
-
v
i
d
e
o
0
1
4
0
0
0
.
2
5
7
0
.
4
6
3
0
.
2
5
9
0
.
3
3
2
0
.
3
4
3
0
.
3
2
6
0
.
3
5
0
10
C
a
m
0
2
-
v
i
d
e
o
0
2
4
5
9
0
.
5
3
5
0
.
4
6
6
0
.
5
7
0
0.
5
4
0
0
.
3
5
5
0
.
5
0
2
0
.
5
6
3
11
C
a
m
0
3
-
v
i
d
e
o
0
1
3
4
3
0
.
2
6
2
0
.
5
5
2
0
.
5
4
7
0
.
5
6
4
0
.
4
3
7
0
.
4
9
8
0
.
1
8
2
12
C
a
m
0
3
-
v
i
d
e
o
0
2
1
5
0
0
.
4
1
0
0
.
1
6
9
0
.
2
1
7
0
.
5
2
9
0
.
3
6
8
0
.
5
4
7
0
.
4
1
3
13
C
a
m
0
3
-
v
i
d
e
o
0
3
1
5
0
0
.
6
4
4
0
.
6
1
1
0
.
6
2
2
0
.
6
8
2
0
.
5
8
5
0
.
5
8
7
0
.
5
9
5
14
C
a
m
0
3
-
v
i
d
e
o
0
4
5
5
0
0
.
5
2
1
0
.
4
9
2
0
.
4
9
7
0.
3
3
9
0
.
5
0
7
0
.
4
6
1
0
.
4
6
5
A
v
e
r
a
g
e
a
c
c
u
r
a
c
y
0
.
4
6
9
0
.
4
5
9
0
.
4
7
9
0
.
4
9
0
0
.
4
5
3
0
.
4
2
5
0
.
4
7
6
T
ab
le
5
.
R
a
w
r
o
b
u
s
t
n
ess
r
es
u
l
ts
o
f
d
if
f
er
en
t tr
ain
i
n
g
s
a
m
p
le
co
n
f
i
g
u
r
atio
n
p
er
ea
ch
test
s
eq
u
en
ce
C
o
n
f
i
g
u
r
a
t
i
o
n
1
:
1
1
:
2
1
:
3
1
:
4
2
:
1
3
:
1
4
:
1
T
o
t
a
l
samp
l
e
s
2
0
0
1
5
0
2
0
0
2
5
0
1
5
0
2
0
0
2
5
0
A
v
e
r
a
g
e
r
o
b
u
st
n
e
ss
6
.
6
4
3
6
.
2
1
4
6
.
0
7
1
3
.
9
2
9
7
.
9
2
9
8
.
8
5
7
6
.
0
7
1
Fig
u
r
e
4
.
E
x
p
ec
ted
av
er
ag
e
o
v
er
lap
cu
r
v
e
f
o
r
d
if
f
er
en
t tr
ai
n
i
n
g
s
a
m
p
le
co
n
f
i
g
u
r
atio
n
tes
te
d
o
n
all
test
s
eq
u
en
ce
s
0
0
,
5
1
1
19
37
55
73
91
1
0
9
1
2
7
1
4
5
1
6
3
1
8
1
1
9
9
2
1
7
2
3
5
2
5
3
2
7
1
2
8
9
3
0
7
3
2
5
3
4
3
3
6
1
3
7
9
3
9
7
4
1
5
4
3
3
4
5
1
4
6
9
4
8
7
5
0
5
5
2
3
5
4
1
Ex
p
e
c
t
e
d
O
v
e
r
l
a
p
S
e
q
u
e
n
c
e
L
e
n
g
t
h
E
x
p
e
c
t
e
d
Ov
e
r
l
a
p
C
u
r
v
e
1
0
0
:
1
0
0
5
0
:
1
0
0
5
0
:
1
5
0
5
0
:
2
0
0
1
0
0
:
5
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
938
I
n
t J
A
r
ti
f
I
n
tell
,
Vo
l.
9
,
No
.
2
,
J
u
n
e
20
20
:
282
–
2
8
9
288
T
ab
le
6
.
E
x
p
ec
ted
av
er
ag
e
o
v
er
lap
r
esu
lts
an
d
t
h
e
tr
ac
k
er
r
a
n
k
o
f
d
if
f
er
en
t tr
ai
n
i
n
g
s
a
m
p
le
co
n
f
i
g
u
r
atio
n
u
s
i
n
g
a
r
an
g
e
o
f
[
1
,
5
5
0
]
T
r
a
i
n
i
n
g
sam
p
l
e
c
o
n
f
i
g
u
r
a
t
i
o
n
1
:
1
1
:
2
1
:
3
1
:
4
2
:
1
3
:
1
4
:
1
EA
O
0
.
2
9
6
7
4
5
0
.
2
8
8
1
7
1
0
.
3
4
0
4
7
0
.
2
5
3
5
2
4
0
.
2
8
6
3
3
3
0
.
2
2
0
6
3
6
0
.
3
6
0
5
2
9
R
a
n
k
3
4
2
6
5
7
1
4
CO
NCLU
SI
O
N
I
n
t
h
is
p
ap
er
,
w
e
h
av
e
p
r
o
p
o
s
ed
a
co
n
v
o
lu
tio
n
al
o
b
j
ec
t
tr
ac
k
er
f
o
r
th
e
n
ig
h
t
v
id
eo
s
u
r
v
eil
lan
ce
.
T
h
e
tr
ac
k
er
r
eq
u
ir
es
th
e
m
o
d
el
to
l
ea
r
n
t
h
e
o
b
j
ec
t'
s
ap
p
ea
r
an
ce
a
n
d
it
s
b
ac
k
g
r
o
u
n
d
p
r
o
p
er
ties
as
t
h
e
o
b
j
ec
t
m
o
v
e
s
ar
o
u
n
d
i
n
t
h
e
s
ce
n
e.
T
h
e
al
g
o
r
ith
m
w
as
test
ed
w
it
h
d
i
f
f
er
en
t
tr
ai
n
i
n
g
d
ata
co
n
f
i
g
u
r
atio
n
s
to
f
in
d
th
e
m
o
s
t
s
u
itab
le
o
n
es
w
it
h
t
h
e
h
i
g
h
e
s
t
ac
c
u
r
ac
y
a
n
d
r
o
b
u
s
t
n
e
s
s
.
T
h
e
r
esu
lt
s
s
h
o
w
t
h
at
th
e
r
a
tio
o
f
4
:1
b
et
w
ee
n
p
o
s
itiv
e
a
n
d
n
eg
a
tiv
e
s
a
m
p
les
p
r
o
d
u
ce
s
t
h
e
b
est
p
er
f
o
r
m
a
n
c
e
f
o
r
t
h
e
s
eq
u
e
n
ce
r
a
n
g
e
o
f
[
1
,
5
5
0
]
,
f
o
llo
w
ed
b
y
th
e
1
:3
co
n
f
ig
u
r
atio
n
.
B
o
th
c
o
n
f
i
g
u
r
atio
n
s
a
ls
o
p
r
o
d
u
ce
th
e
s
ec
o
n
d
-
b
est
p
er
f
o
r
m
an
ce
in
ter
m
s
o
f
tr
ac
k
i
n
g
r
o
b
u
s
tn
es
s
.
T
h
is
ex
p
lai
n
s
t
h
at
d
if
f
er
en
t
tr
ai
n
in
g
s
a
m
p
le
s
izes
an
d
r
atio
s
a
f
f
ec
t
tr
a
ck
in
g
p
er
f
o
r
m
a
n
ce
d
if
f
er
e
n
tl
y
a
n
d
as
m
o
r
e
p
o
s
iti
v
e
s
a
m
p
les
ar
e
u
s
ed
to
r
ep
r
esen
t
t
h
e
o
b
j
ec
t
co
m
p
ar
ed
to
t
h
e
n
e
g
ati
v
e
s
a
m
p
les,
th
e
b
etter
th
e
tr
ac
k
er
p
er
f
o
r
m
an
ce
is
.
T
h
is
in
d
icate
s
t
h
at
alt
h
o
u
g
h
t
h
e
o
b
j
ec
t
-
b
ac
k
g
r
o
u
n
d
co
n
tr
ast
is
lo
w
,
th
e
b
ac
k
g
r
o
u
n
d
is
al
s
o
h
o
m
o
g
en
e
o
u
s
d
u
e
to
lo
w
li
g
h
ti
n
g
co
n
d
iti
o
n
,
th
u
s
a
lo
w
er
n
u
m
b
er
o
f
s
a
m
p
les ar
e
s
u
f
f
icien
t
to
r
ep
r
esen
t th
e
m
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
e
au
th
o
r
s
w
o
u
ld
li
k
e
t
o
ac
k
n
o
w
led
g
e
f
u
n
d
in
g
s
f
r
o
m
Mi
n
is
tr
y
o
f
E
d
u
ca
tio
n
Ma
la
y
s
ia
(
FR
GS/1
/2
0
1
9
/I
C
T
0
2
/UKM
/0
2
/1
)
an
d
Un
iv
er
s
i
ti
Keb
an
g
s
aa
n
Ma
la
y
s
ia
(
GUP
-
2019
-
0
0
8
)
.
T
h
e
g
r
ap
h
ic
p
r
o
ce
s
s
in
g
u
n
it u
s
ed
in
t
h
is
r
e
s
ea
r
ch
is
d
o
n
ated
b
y
NVI
DI
A
C
o
r
p
o
r
atio
n
(
KK
-
2019
-
0
0
5
)
.
RE
F
E
R
E
NC
E
S
[1
]
A
.
A
li
,
A
.
Ja
li
l,
J.
Niu
,
X
.
Zh
a
o
,
S
.
Ra
th
o
re
,
J.
A
h
m
e
d
,
a
n
d
M
.
A
.
I
k
h
a
r.
“
V
isu
a
l
o
b
jec
t
trac
k
in
g
:
c
las
sic
a
l
a
n
d
c
o
n
tem
p
o
ra
r
y
a
p
p
ro
a
c
h
e
s,”
in
F
ro
n
ti
e
rs
o
f
Co
m
p
u
ter
S
c
ien
c
e
:
S
e
l
e
c
ted
Pu
b
li
c
a
t
io
n
s
fro
m
Ch
i
n
e
se
Un
ive
rs
it
ies
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
6
7
-
1
8
8
,
2
0
1
6
.
[2
]
L
.
Zh
a
n
g
,
C.
L
i
m
,
Y.
Ch
e
n
,
a
n
d
H.
R.
Ka
ri
m
i,
"
T
ra
c
k
in
g
M
o
b
il
e
Ro
b
o
t
in
In
d
o
o
r
W
irele
ss
S
e
n
so
r
Ne
t
w
o
rk
s,"
in
M
a
th
e
ma
ti
c
a
l
Pro
b
lem
s in
E
n
g
in
e
e
rin
g
,
V
o
l
u
m
e
2
0
1
4
h
t
tp
s://
d
o
i
.
o
rg
/1
0
.
1
1
5
5
/
2
0
1
4
/
8
3
7
0
5
0
[3
]
J.
S
e
v
e
rso
n
,
“
Hu
m
a
n
-
d
ig
it
a
l
m
e
d
i
a
in
tera
c
ti
o
n
trac
k
in
g
”
,
US
Pa
te
n
t
9
,
7
1
3
,
4
4
4
,
2
0
1
7
[4
]
P
.
R
I
y
e
r,
S
.
R.
I
y
e
r,
R.
Ra
m
e
s
h
,
A
n
a
la,
K.
N.
S
u
b
ra
m
a
n
y
a
,
“
Ad
a
p
ti
v
e
re
a
l
ti
me
tra
ff
ic
p
re
d
ictio
n
u
sin
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s,”
in
IAE
S
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Arti
fi
c
i
a
l
In
tell
ig
e
n
c
e
(IJ
-
A
I),
v
o
l.
8
,
n
o
.
2
,
Ju
n
e
2
0
1
9
,
p
p
.
1
0
7
-
1
1
9
.
[5
]
M
.
A
.
Zu
lk
if
l
e
y
,
N.
A
.
M
o
h
a
m
e
d
,
a
n
d
N.
H.
Zu
lk
if
le
y
,
“
S
q
u
a
t
A
n
g
le
A
ss
e
ss
m
e
n
t
T
h
ro
u
g
h
T
ra
c
k
in
g
Bo
d
y
M
o
v
e
m
e
n
ts,” i
n
IEE
E
Acc
e
ss
,
Vo
l.
7
,
p
p
.
4
8
6
3
5
-
4
8
6
4
4
,
2
0
1
9
.
[6
]
V
.
A
L
a
u
re
n
se
,
J.
Y
G
o
h
,
a
n
d
J
C.
Ge
rd
e
s,
“
P
a
th
-
trac
k
in
g
f
o
r
a
u
to
n
o
m
o
u
s
v
e
h
icle
s
a
t
th
e
li
m
it
o
f
f
rictio
n
,
”
In
Ame
ric
a
n
Co
n
tro
l
C
o
n
fer
e
n
c
e
,
p
p
.
5
5
8
6
-
5
5
9
1
,
2
0
1
7
.
[7
]
M
.
A
.
Zu
lk
i
f
le
y
,
N.
S
.
S
a
m
a
n
u
,
N.
Zu
lk
e
p
e
li
,
Z.
Ka
d
im
,
a
n
d
H.
H.
W
o
o
n
,
“
Ka
l
m
a
n
f
il
ter
-
b
a
se
d
a
g
g
re
ss
i
v
e
b
e
h
a
v
io
u
r
d
e
tec
ti
o
n
f
o
r
in
d
o
o
r
e
n
v
iro
n
m
e
n
t,
”
in
L
e
c
tu
re
No
tes
in
El
e
c
trica
l
En
g
in
e
e
rin
g
,
V
o
l
.
3
7
6
,
p
p
.
8
2
9
-
8
3
7
,
2
0
1
6
.
[8
]
K.
Hu
a
n
g
,
L
.
W
a
n
g
,
a
n
d
T
.
T
a
n
,
“
De
tec
ti
n
g
a
n
d
trac
k
in
g
d
istan
t
o
b
jec
ts
a
t
n
ig
h
t
b
a
se
d
o
n
h
u
m
a
n
v
isu
a
l
s
y
ste
m
,
”
L
e
c
t.
No
tes
Co
m
p
u
t.
S
c
i.
,
v
o
l.
3
8
5
2
L
NC
S
,
p
p
.
8
2
2
-
8
3
1
,
2
0
0
6
.
[9
]
K.
Hu
a
n
g
,
L
.
W
a
n
g
,
T
.
T
a
n
,
S
.
M
a
y
b
a
n
k
,
“
A
r
e
a
l
-
ti
m
e
o
b
jec
t
d
e
tec
ti
n
g
a
n
d
trac
k
in
g
s
y
ste
m
f
o
r
o
u
td
o
o
r
n
ig
h
t
su
rv
e
il
lan
c
e
,
”
in
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
4
1
,
n
p
.
1
,
p
p
.
4
3
2
-
4
4
4
,
2
0
0
8
.
[1
0
]
L
.
W
a
n
g
,
K.
Hu
a
n
g
,
Y.
Hu
a
n
g
a
n
d
T
.
T
a
n
,
"
Ob
jec
t
d
e
tec
ti
o
n
a
n
d
trac
k
in
g
f
o
r
n
ig
h
t
su
rv
e
il
lan
c
e
b
a
se
d
o
n
sa
li
e
n
t
c
o
n
tras
t
a
n
a
ly
sis,"
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ima
g
e
Pro
c
e
ss
in
g
,
p
p
.
1
1
1
3
-
1
1
1
6
,
2
0
0
9
.
[1
1
]
A
.
Na
z
ib
,
C.
Oh
a
n
d
C
.
W
.
L
e
e
,
"
Ob
jec
t
d
e
tec
ti
o
n
a
n
d
trac
k
in
g
in
n
ig
h
t
ti
m
e
v
id
e
o
su
rv
e
il
lan
c
e
,
"
1
0
t
h
In
te
rn
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ub
i
q
u
i
to
u
s R
o
b
o
ts
a
n
d
Am
b
ien
t
I
n
telli
g
e
n
c
e
(
URAI
)
,
p
p
.
6
2
9
-
6
3
2
,
2
0
1
3
.
[1
2
]
H.
Ya
n
g
,
L
.
S
h
a
o
,
F
.
Zh
e
n
g
,
L
.
W
a
n
g
,
a
n
d
Z.
S
o
n
g
,
“
Re
c
e
n
t
a
d
v
a
n
c
e
s
a
n
d
tren
d
s
i
n
v
isu
a
l
trac
k
in
g
:
A
re
v
ie
w
,
”
in
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
7
4
,
n
o
.
1
8
,
p
p
.
3
8
2
3
-
3
8
3
1
,
2
0
1
1
.
[1
3
]
X
.
L
i,
W
.
Hu
,
C.
S
h
e
n
,
Z.
Zh
a
n
g
,
A
.
Dic
k
,
a
n
d
A
.
V
De
n
He
n
g
e
l,
“
A
su
rv
e
y
o
f
a
p
p
e
a
ra
n
c
e
m
o
d
e
ls
in
v
isu
a
l
o
b
jec
t
trac
k
in
g
,
”
in
ACM
T
ra
n
s
a
c
ti
o
n
s o
n
In
tell
ig
e
n
t
S
y
ste
ms
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
4
,
n
o
.
4
,
2
0
1
3
.
[1
4
]
M
.
F
iaz
,
A
.
M
a
h
m
o
o
d
,
S
.
Ja
v
e
d
,
a
n
d
S
.
K.
Ju
n
g
,
‘‘
Ha
n
d
c
ra
f
ted
a
n
d
d
e
e
p
trac
k
e
rs:
Re
c
e
n
t
v
isu
a
l
o
b
jec
t
trac
k
in
g
a
p
p
ro
a
c
h
e
s a
n
d
tren
d
s,’’
i
n
ACM
Co
mp
u
t
in
g
S
u
rv
e
y
,
v
o
l.
5
2
,
n
o
.
2
,
p
p
.
1
-
4
3
,
2
0
1
9
.
[1
5
]
C.
S
z
e
g
e
d
y
,
A
.
T
o
sh
e
v
,
a
n
d
D.
Erh
a
n
,
“
De
e
p
Ne
u
ra
l
Ne
tw
o
rk
f
o
r
O
b
jec
t
De
tec
ti
o
n
,
”
Ne
u
ra
l
I
n
f.
Pr
o
c
e
ss
.
S
y
st
.
,
v
o
l
.
7
,
n
o
.
5
,
p
p
.
2
0
0
-
2
0
5
,
2
0
1
3
.
[1
6
]
S
.
Re
n
,
K
.
He
,
R.
G
irsh
ick
,
a
n
d
J.
S
u
n
,
“
F
a
ste
r
R
-
CNN
:
T
o
wa
rd
s
Re
a
l
-
T
i
m
e
Ob
jec
t
De
te
c
ti
o
n
w
it
h
Re
g
io
n
P
r
o
p
o
sa
l
Ne
tw
o
rk
s.”
in
IEE
E
T
r
a
n
s.
Pa
tt
e
rn
A
n
a
l
.
M
a
c
h
.
In
tell
.
,
v
o
l.
3
9
,
n
o
.
6
,
p
p
.
1
1
3
7
-
1
1
4
9
,
2
0
1
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
A
r
ti
f
I
n
tell
I
SS
N:
2252
-
8938
Tr
a
in
in
g
co
n
fig
u
r
a
tio
n
a
n
a
lysi
s
o
f a
co
n
vo
l
u
tio
n
a
l
n
e
u
r
a
l n
e
tw
o
r
k
o
b
ject
tr
a
ck
er fo
r
... (
Zu
l
a
ikh
a
K
a
d
im
)
289
[1
7
]
M
.
A
.
Zu
lk
i
f
le
y
,
S
.
R.
A
b
d
a
n
i
a
n
d
N.
H.
Zu
lk
if
le
y
,
"
P
ter
y
g
iu
m
-
N
e
t:
a
d
e
e
p
lea
rn
in
g
a
p
p
ro
a
c
h
to
p
tery
g
iu
m
d
e
tec
ti
o
n
a
n
d
l
o
c
a
li
z
a
ti
o
n
,
"
in
M
u
lt
ime
d
ia
T
o
o
ls a
n
d
Ap
p
li
c
a
ti
o
n
s
,
v
o
l.
7
8
.
n
o
.
2
4
,
p
p
.
3
4
5
6
3
-
3
4
5
8
4
,
2
0
1
9
.
[1
8
]
S
.
B.
Ja
d
h
a
v
,
V
.
Ud
u
p
i,
S
.
P
a
ti
l,
“
Co
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
fo
r
lea
f
i
m
a
g
e
-
b
a
se
d
p
lan
t
d
ise
a
se
c
las
si
f
ica
ti
o
n
,
”
IAE
S
I
n
ter
n
a
t
io
n
a
l
J
o
u
r
n
a
l
o
f
Art
if
icia
l
I
n
telli
g
e
n
c
e
(IJ
-
A
I),
v
o
l.
8
,
n
o
.
4
,
p
p
.
3
2
8
-
3
4
1
,
2
0
1
9
.
[1
9
]
J.
H.
Ki
m
,
H.
G
.
Ho
n
g
,
K.
R
P
a
rk
,
“
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
r
k
-
Ba
se
d
Hu
m
a
n
D
e
tec
ti
o
n
in
Nig
h
tt
im
e
I
m
a
g
e
s
Usi
n
g
V
isi
b
le L
ig
h
t
Ca
m
e
ra
S
e
n
so
rs
”
,
in
S
e
n
so
rs
(
S
w
it
z
e
rlan
d
),
v
o
l.
1
7
.
n
o
.
5
,
p
p
.
1
-
2
6
,
2
0
1
7
.
[2
0
]
H.
Na
m
a
n
d
B.
Ha
n
,
‘‘L
e
a
rn
in
g
m
u
lt
i
-
d
o
m
a
in
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
v
isu
a
l
trac
k
in
g
,
’’
Pro
c
.
IEE
E
Co
n
f.
Co
mp
u
ter
Vi
si
o
n
Pa
tt
e
rn
R
e
c
o
g
n
it
i
o
n
,
p
p
.
4
2
9
3
-
4
3
0
2
,
2
0
1
6
.
[2
1
]
H.
Na
m
,
M
.
Ba
e
k
,
a
n
d
B.
Ha
n
.
‘‘
M
o
d
e
li
n
g
a
n
d
p
r
o
p
a
g
a
ti
n
g
CNN
s
in
a
tree
stru
c
tu
re
f
o
r
v
isu
a
l
trac
k
in
g
,
’’
in
ArXi
v
Co
RR
,
v
o
l
.
a
b
s/
1
6
0
8
.
0
7
2
4
2
,
p
p
.
1
-
1
0
,
2
0
1
6
[2
2
]
M
.
A
.
Zu
lk
if
le
y
a
n
d
N.
T
rig
o
n
i,
“
M
u
lt
i
p
le
-
M
o
d
e
l
F
u
ll
y
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
s
f
o
r
S
in
g
le
O
b
jec
t
T
ra
c
k
in
g
o
n
T
h
e
rm
a
l
In
f
ra
r
e
d
V
i
d
e
o
,
”
i
n
I
EE
E
Acc
e
ss
,
v
o
l.
6
,
p
p
.
4
2
7
9
0
-
4
2
7
9
9
,
2
0
1
8
.
[2
3
]
M
.
A
.
Zu
lk
ifl
e
y
,
"
Tw
o
S
trea
m
s
M
u
lt
i
p
le
-
M
o
d
e
l
Ob
jec
t
T
ra
c
k
e
r
fo
r
T
h
e
r
m
a
l
In
f
ra
re
d
V
id
e
o
"
,
in
IEE
E
Acc
e
ss
,
v
o
l.
7
,
p
p
.
3
2
3
8
3
-
3
2
3
9
2
,
2
0
1
9
.
[2
4
]
K.
Ch
a
tf
ield
,
K.
S
im
o
n
y
a
n
,
A
.
Ve
d
a
ld
i,
a
n
d
A
.
Zi
ss
e
rm
a
n
,
“
Re
tu
rn
o
f
t
h
e
De
v
il
in
th
e
D
e
tails:
De
l
v
in
g
De
e
p
in
to
Co
n
v
o
l
u
ti
o
n
a
l
Ne
tw
o
rk
s”
,
in
Brit
ish
M
a
c
h
i
n
e
Vi
sio
n
C
o
n
fer
e
n
c
e
,
p
p
.
1
-
12
,
2
0
1
4
.
[2
5
]
O.
Ru
ss
a
k
o
v
sk
y
e
t
a
l.
,
‘‘Im
a
g
e
N
e
t
larg
e
sc
a
le
v
isu
a
l
re
c
o
g
n
it
io
n
c
h
a
ll
e
n
g
e
’’,
in
I
n
ter
n
a
ti
o
n
a
l
J
.
C
o
mp
u
ter
Vi
sio
n
,
v
o
l.
1
1
5
,
n
o
.
3
,
p
p
.
2
1
1
-
2
5
2
,
2
0
1
5
.
[2
6
]
D.
P
.
Kin
g
m
a
a
n
d
J.
Ba
,
“
A
d
a
m
:
A
M
e
th
o
d
f
o
r
S
to
c
h
a
stic
Op
ti
m
iza
ti
o
n
,
”
in
ArXi
v
Co
RR
,
v
o
l.
a
rX
iv
:1
4
1
2
.
6
9
8
0
,
p
p
.
1
-
1
5
,
2
0
1
4
.
[2
7
]
Z.
Ka
d
im
,
M
.
A
.
Zu
lk
if
le
y
,
N.
H
a
m
z
a
h
,
“
D
e
e
p
-
lea
rn
in
g
b
a
se
d
sin
g
le
o
b
jec
t
tra
c
k
e
r
f
o
r
n
ig
h
t
su
rv
e
i
ll
a
n
c
e
,
”
in
IAE
S
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
E
lec
trica
l
a
n
d
C
o
mp
u
ter
En
g
in
e
e
rin
g
(
IJ
ECE
)
,
v
o
l.
1
0
,
n
o
.
4
,
2
0
2
0
.
[2
8
]
Y.
W
u
,
"
G
o
o
g
le
'
s
N
e
u
ra
l
M
a
c
h
in
e
T
ra
n
sla
ti
o
n
S
y
ste
m
:
B
rid
g
in
g
th
e
G
a
p
b
e
t
w
e
e
n
Hu
m
a
n
a
n
d
M
a
c
h
in
e
T
ra
n
sla
ti
o
n
,
"
in
ArX
iv Co
RR
,
v
o
l.
a
b
s/1
6
0
9
.
0
8
1
4
4
,
p
p
.
1
-
2
3
,
2
0
1
6
[2
9
]
M
.
Krista
n
a
n
d
o
th
e
rs.
“
T
h
e
V
i
su
a
l
Ob
jec
t
T
ra
c
k
in
g
V
OT
2
0
1
5
Ch
a
ll
e
n
g
e
Re
su
lt
s,”
in
Vi
su
a
l
Ob
jec
t
T
ra
c
k
in
g
W
o
rk
sh
o
p
ICCV
2
0
1
5
,
2
0
1
5
.
[3
0
]
M
.
Krista
n
,
R.
P
f
lu
g
f
e
ld
e
r,
A
.
L
e
o
n
a
rd
is,
J
.
M
a
tas
,
F
.
P
o
rik
li
,
L
.
Ce
h
o
v
in
,
G
.
Ne
b
e
h
a
y
,
G
.
F
e
rn
a
n
d
e
z
,
a
n
d
T
.
V
o
ji
r
.
“
T
h
e
V
OT
2
0
1
3
c
h
a
ll
e
n
g
e
:
o
v
e
rv
iew
a
n
d
a
d
d
it
io
n
a
l
re
su
lt
s
,
”
I
n
C
o
mp
u
ter
Vi
si
o
n
W
in
ter
W
o
rk
sh
o
p
,
2
0
1
4
.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
Zu
laik
h
a
Ka
d
im
re
c
e
iv
e
d
h
e
r
B.
En
g
d
e
g
re
e
in
e
lec
tro
n
ics
e
n
g
in
e
e
rin
g
m
a
jo
rin
g
in
tele
c
o
m
m
u
n
ica
ti
o
n
f
ro
m
M
u
lt
im
e
d
ia
Un
iv
e
rsity
,
M
a
la
y
sia
in
2
0
0
0
a
n
d
c
u
rre
n
tl
y
p
e
rsu
in
g
h
e
r
P
h
.
D
d
e
g
re
e
in
Un
iv
e
rsiti
Ke
b
a
n
g
sa
a
n
M
a
la
y
sia
.
S
h
e
is
a
lso
a
re
se
a
rc
h
e
r
in
n
a
ti
o
n
a
l
R&
D
in
stit
u
ti
o
n
.
He
r
c
u
rre
n
t
re
se
a
rc
h
in
tere
sts
a
re
in
v
id
e
o
a
n
a
l
y
ti
c
s,
b
e
h
a
v
io
ra
l
a
n
a
ly
sis
a
n
d
v
isu
a
l
o
b
jec
t
trac
k
in
g
.
M
o
h
d
A
s
y
r
a
f
Zu
lk
i
f
le
y
(M
’1
8
)
r
e
c
e
iv
e
d
th
e
B.
En
g
.
d
e
g
re
e
in
m
e
c
h
a
tro
n
ics
f
ro
m
In
tern
a
ti
o
n
a
l
Isla
m
ic
Un
iv
e
r
sit
y
M
a
la
y
sia
in
2
0
0
8
a
n
d
th
e
P
h
.
D.
d
e
g
re
e
in
e
lec
tri
c
a
l
a
n
d
e
lec
tro
n
ic
e
n
g
in
e
e
rin
g
f
ro
m
T
h
e
Un
iv
e
rsit
y
o
f
M
e
lb
o
u
r
n
e
in
2
0
1
2
.
He
w
a
s
a
sp
o
n
so
re
d
Re
se
a
rc
h
e
r
a
t
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Un
iv
e
rsit
y
o
f
O
x
f
o
rd
f
ro
m
2
0
1
6
to
2
0
1
8
.
He
is
c
u
rre
n
tl
y
a
n
A
s
so
c
iate
P
ro
f
e
ss
o
r
a
t
th
e
F
a
c
u
lt
y
o
f
En
g
in
e
e
rin
g
a
n
d
Bu
il
t
En
v
ir
o
n
m
e
n
t,
Un
iv
e
rsiti
Ke
b
a
n
g
sa
a
n
M
a
la
y
sia
.
His c
u
rre
n
t
re
se
a
rc
h
in
t
e
re
sts a
re
v
isu
a
l
o
b
jec
t
trac
k
in
g
a
n
d
m
e
d
ica
l
im
a
g
e
a
n
a
l
y
sis.
No
r
A
z
wa
n
M
o
h
a
m
m
e
d
Ka
m
a
ri
re
c
e
i
v
e
d
B.
En
g
.
f
ro
m
M
e
ij
i
Un
iv
e
rsit
y
,
J
a
p
a
n
,
M
.
S
c
.
d
e
g
re
e
s
f
ro
m
Eh
i
m
e
Un
iv
e
rsit
y
,
J
a
p
a
n
a
n
d
P
h
D
d
e
g
re
e
f
ro
m
Un
iv
e
r
siti
Te
k
n
o
lo
g
i
M
a
ra
,
M
a
la
y
sia
2
0
0
0
,
2
0
0
4
a
n
d
2
0
1
6
,
re
sp
e
c
ti
v
e
ly
.
He
h
a
s
a
u
t
h
o
re
d
a
n
d
c
o
-
a
u
t
h
o
re
d
m
o
re
th
a
n
3
0
tec
h
n
ica
l
p
a
p
e
rs
p
u
b
li
sh
e
d
i
n
t
h
e
in
tern
a
ti
o
n
a
l
j
o
u
r
n
a
ls,
n
a
ti
o
n
a
l
a
n
d
in
ter
n
a
ti
o
n
a
l
c
o
n
f
e
re
n
c
e
s.
His
re
se
a
rc
h
in
tere
sts in
c
lu
d
e
p
o
w
e
r
s
y
ste
m
st
a
b
il
it
y
a
n
d
a
rti
f
icia
l
in
telli
g
e
n
c
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.