I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
A
I
)
V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024, pp.
4158
~
4168
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v13.i4.pp
4158
-
4168
4158
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
C
on
c
i
se
C
ar
N
e
t
:
c
on
vol
u
t
i
on
al
n
e
u
r
al
n
e
t
w
or
k
f
or
p
ar
k
i
n
g sp
ac
e
c
l
ass
i
f
i
c
at
i
on
M
ar
w
an
R
am
li
1
, S
ayu
t
i
R
ah
m
an
2
, R
ah
m
ad
B
ayu
S
yah
2
1
D
e
pa
r
t
m
e
nt
of
M
a
t
he
m
a
t
i
c
s
,
F
a
c
ul
t
y of
M
a
t
he
m
a
t
i
c
s
a
nd N
a
t
ur
a
l
S
c
i
e
nc
e
s
,
U
n
i
ve
r
s
i
t
a
s
S
yi
a
h K
ua
l
a
, B
a
nd
a
A
c
e
h,
I
ndone
s
i
a
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
c
s
, E
ngi
ne
e
r
i
ng
F
a
c
ul
t
y, U
ni
ve
r
s
i
t
a
s
M
e
da
n A
r
e
a
,
M
e
da
n, I
ndone
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
J
ul
24, 2023
R
e
vi
s
e
d
A
pr
2, 2024
A
c
c
e
pt
e
d
A
pr
17, 2024
The
car
is
a
mode
of
transporta
tion
that
brings
numerous
benefits
to
the
communi
ty.
As
a
result,
the
growth
of
vehicles
is
increasing
,
which
has
a
negative
impact.
Some
of
the
negative
impacts
include
noise,
air
pol
lution,
traffic
congestion,
and the
need fo
r
parking spa
ces. D
rivers
that drive
around
looking
for
parking
places
increase
the
negative
impact
as
well
as
bo
redom
and even wo
rry for th
e
driver.
Therefore,
the dri
ver needs
this
informat
ion on
the
availability
of
parking
spaces.
A
convolutional
neural
network
(CNN)
usin
g
a
camera
is
one
of
the
best
methods
that
can
be
used
to
sol
ve
this
problem.
We
built
a
more
efficient
CNN
architecture
for
classifying
p
arking
spaces,
which
was
named
ConciseCarNet.
ConciseCarNet
uses
3
3
a
nd
1
1
convolut
ion
filters,
which
cause
fewer
parameters
than
previous
archite
ctures.
ConciseCa
rNet ha
s two bra
nches,
each
with
a diff
erent b
ranch
structur
e. This
branch
is
designed
to
generate
additional
feature
variations,
which
wi
ll
help
improve
the
accura
cy
.
Based
on
testing,
the
accuracy
of
Concise
Ca
rNet2x
outperforms the accura
cy of mAlexnet, Carnet, Efficie
ntParkingNet, and you
look
once
(YOLO)+
MobilNet
architectures,
which
is
99.37%.
Concise
CarNet
has
fewer
parameters,
file
sizes,
and
floating
point
operations
(
F
LOPs)
compared t
o other ar
chitectur
es.
K
e
y
w
o
r
ds
:
C
onc
is
e
c
a
r
ne
twor
k
C
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
D
e
e
p l
e
a
r
ni
ng
P
a
r
ki
ng s
pa
c
e
S
m
a
r
t
pa
r
ki
ng s
ys
te
m
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
S
a
yut
i
R
a
hm
a
n
D
e
pa
r
tm
e
nt
of
I
nf
or
m
a
ti
c
s
, E
ngi
ne
e
r
in
g F
a
c
ul
ty
, U
ni
ve
r
s
it
a
s
M
e
da
n A
r
e
a
J
a
la
n K
ol
a
m
N
o. 1 M
e
da
n
E
s
ta
te
,
M
e
da
n, I
ndone
s
ia
E
m
a
il
:
s
a
yut
ir
a
hm
a
n@
s
ta
f
f
.um
a
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
N
ow
a
da
ys
,
th
e
u
s
e
of
c
a
r
s
f
or
tr
a
ns
por
ta
ti
on,
bot
h
w
it
hi
n
ur
ba
n
a
r
e
a
s
a
nd
be
yond,
ha
s
b
e
c
om
e
e
s
s
e
nt
ia
l.
T
he
popula
r
it
y
of
c
a
r
s
is
on
th
e
r
is
e
du
e
to
th
e
ir
gr
e
a
t
e
r
c
a
pa
c
it
y
c
om
pa
r
e
d
to
m
ot
or
c
yc
le
s
a
nd
th
e
ir
a
bi
li
ty
to
s
hi
e
ld
o
c
c
upa
nt
s
f
r
om
e
xt
e
r
na
l
a
ir
pol
l
ut
io
n.
D
e
s
pi
te
th
e
a
dva
nt
a
ge
s
,
th
e
s
ur
ge
in
th
e
pr
e
va
le
nc
e
of
c
a
r
s
br
in
gs
a
bout
a
dve
r
s
e
e
f
f
e
c
ts
,
in
c
lu
di
ng
a
ir
pol
lu
ti
on,
noi
s
e
,
li
m
it
e
d
pa
r
ki
ng
a
va
il
a
bi
li
ty
,
tr
a
f
f
ic
c
onge
s
ti
on,
a
nd
he
ig
ht
e
ne
d
f
ue
l
c
ons
um
pt
io
n
[
1]
.
V
e
hi
c
l
e
e
xha
u
s
t
e
m
i
s
s
io
ns
c
ont
r
ib
ut
e
to
a
ir
pol
lu
ti
on,
po
s
in
g
r
is
ks
to
publ
ic
he
a
lt
h,
a
nd
a
r
e
a
s
ig
ni
f
ic
a
nt
s
our
c
e
of
a
nt
hr
opoge
ni
c
c
a
r
bon
di
oxi
de
(
C
O
2
)
[
2]
.
F
ur
th
e
r
m
or
e
,
th
e
e
s
c
a
la
ti
ng
num
be
r
of
c
a
r
s
h
a
s
in
te
n
s
if
ie
d
th
e
de
m
a
nd
f
or
pa
r
ki
ng
s
pa
c
e
s
,
le
a
di
ng
to
c
h
a
ll
e
nge
s
s
uc
h
a
s
tr
a
f
f
ic
c
onge
s
ti
on
c
a
u
s
e
d
by
dr
iv
e
r
s
s
e
a
r
c
hi
ng
f
or
pa
r
ki
ng
s
pa
c
e
.
C
on
s
e
que
nt
ly
,
im
pl
e
m
e
nt
in
g
a
n
in
te
ll
ig
e
nt
pa
r
ki
ng
s
ys
te
m
e
m
e
r
ge
s
a
s
a
vi
a
bl
e
s
ol
ut
io
n t
o a
ddr
e
s
s
t
hi
s
i
s
s
ue
.
S
m
a
r
t
pa
r
ki
ng
s
ys
te
m
s
of
f
e
r
dr
iv
e
r
s
va
r
io
us
be
ne
f
it
s
,
a
id
in
g
th
e
m
in
a
voi
di
ng
tr
a
f
f
ic
c
onge
s
ti
on,
in
e
f
f
ic
ie
nt
f
ue
l
us
a
ge
,
lo
ng
ve
hi
c
le
que
ue
s
,
a
nxi
e
ty
,
a
nd
e
nvi
r
onm
e
nt
a
l
pol
lu
ti
on.
R
e
s
e
a
r
c
h
in
th
e
r
e
a
lm
o
f
s
m
a
r
t
pa
r
ki
ng
s
ys
te
m
s
e
nc
om
pa
s
s
e
s
a
s
pe
c
ts
s
uc
h
a
s
pa
r
ki
ng
s
pa
c
e
a
va
il
a
bi
li
ty
de
te
c
ti
on,
pa
r
ki
ng
m
e
te
r
s
,
c
r
ow
ds
e
ns
in
g,
in
t
e
gr
a
te
d
ve
hi
c
l
e
de
v
e
lo
pm
e
nt
,
s
im
ul
a
ti
on,
p
a
r
ki
ng
da
ta
a
n
a
ly
s
is
,
a
nd
c
om
pe
ti
ti
on
in
pa
r
ki
ng
s
pa
c
e
r
e
s
e
r
va
ti
ons
[
3]
.
T
he
id
e
nt
if
ic
a
ti
on
of
va
c
a
nt
pa
r
ki
ng
s
p
a
c
e
s
s
ta
nds
out
a
s
a
c
r
uc
ia
l
e
le
m
e
nt
of
s
m
a
r
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
onc
is
e
C
a
r
N
e
t:
c
on
v
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
pa
r
k
in
g s
pac
e
c
la
s
s
if
ic
at
io
n (
M
ar
w
an R
am
li
)
4159
pa
r
ki
ng s
ys
te
m
s
. T
hi
s
i
nvol
ve
s
ga
th
e
r
in
g da
ta
t
o i
nf
or
m
dr
iv
e
r
s
a
bout
a
va
il
a
bl
e
pa
r
ki
ng s
pot
s
, w
it
h s
e
ns
or
s
or
c
om
put
e
r
vi
s
io
n
be
in
g
c
om
m
on
m
e
th
ods
f
or
pa
r
ki
ng
s
pa
c
e
de
t
e
c
ti
on.
C
om
put
e
r
vi
s
io
n
c
a
m
e
r
a
s
,
f
or
in
s
ta
nc
e
,
c
a
n
s
im
ul
ta
ne
ous
ly
c
l
a
s
s
if
y
num
e
r
ous
pa
r
ki
ng
s
p
a
c
e
s
,
pr
e
s
e
nt
i
ng
a
m
or
e
c
os
t
-
e
f
f
e
c
ti
ve
a
lt
e
r
na
ti
ve
c
om
pa
r
e
d
to
i
ns
ta
ll
in
g a
nd ma
in
ta
in
in
g s
e
ns
or
s
i
n
e
a
c
h i
ndi
vi
dua
l
pa
r
ki
ng
s
pa
c
e
.
M
a
ny
a
ppr
oa
c
h
e
s
f
or
de
te
r
m
in
in
g
th
e
a
va
il
a
bi
li
ty
of
pa
r
ki
ng
pl
a
c
e
s
w
it
h
good
c
om
put
e
r
vi
s
io
n
ha
v
e
be
e
n
de
ve
lo
pe
d
by
r
e
s
e
a
r
c
he
r
s
.
S
om
e
of
th
e
m
e
th
ods
pr
oduc
e
d
in
c
lu
de
de
te
c
ti
ng
pa
r
ki
ng
s
pa
c
e
s
us
in
g
im
a
ge
s
ubt
r
a
c
ti
on
[
4]
,
gi
vi
ng
m
a
r
ke
r
s
to
pa
r
ki
ng
s
pa
c
e
s
to
be
r
e
c
ogni
z
e
d
[
5]
,
[
6
]
,
a
da
pt
iv
e
a
ppr
oa
c
h
to
s
pa
c
e
c
ons
tr
a
in
ts
w
it
h c
ube
s
[
7]
, a
nd l
e
a
r
ni
ng ba
s
e
[
8]
, [
9]
.
A
m
ong the
s
e
, l
e
a
r
ni
ng
-
ba
s
e
d m
e
th
ods
, pa
r
ti
c
ul
a
r
ly
de
e
p
le
a
r
ni
ng, a
r
e
f
r
e
que
nt
ly
de
ve
lo
pe
d t
o a
ddr
e
s
s
t
hi
s
pr
obl
e
m
.
D
e
e
p l
e
a
r
ni
ng i
s
a
pa
r
ti
c
ul
a
r
ly
s
uc
c
e
s
s
f
ul
m
e
th
od f
or
ha
ndl
in
g c
om
put
e
r
vi
s
io
n di
f
f
ic
ul
ti
e
s
. T
he
de
e
p
le
a
r
ni
ng
m
e
th
od
th
a
t
is
of
te
n
u
s
e
d
to
c
l
a
s
s
if
y
obj
e
c
ts
in
im
a
g
e
s
i
s
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
)
[
10]
.
R
e
s
e
a
r
c
h
e
r
s
ha
ve
s
uc
c
e
s
s
f
ul
ly
us
e
d
C
N
N
to
c
la
s
s
if
y
obj
e
c
ts
in
im
a
ge
s
.
E
xi
s
ti
ng
C
N
N
w
it
h
la
r
ge
e
nough
pa
r
a
m
e
te
r
s
i
nc
lu
de
A
le
xne
t
[
10]
, vi
s
ua
l
g
e
om
e
tr
y gr
oup ne
twor
k (
V
G
G
N
e
t
)
[
11]
, G
oogl
e
N
e
t
[
12]
, a
nd R
e
s
n
e
t
[
13]
.
C
ur
r
e
nt
ly
,
C
N
N
a
r
c
hi
te
c
tu
r
e
s
f
or
m
obi
le
a
ppl
ic
a
ti
ons
ha
ve
be
e
n
f
ound,
in
c
lu
di
ng
M
obi
le
N
e
t
V
2
[
14
]
,
a
nd
S
huf
f
le
N
e
t
[
15
]
w
hi
c
h
ha
ve
a
s
m
a
ll
e
r
num
be
r
o
f
pa
r
a
m
e
te
r
s
.
T
hi
s
e
xi
s
ti
ng
C
N
N
is
a
im
e
d
a
t
c
la
s
s
if
yi
ng
1,000
obj
e
c
ts
f
r
om
th
e
I
m
a
ge
ne
t
da
ta
s
e
t.
T
hus
,
th
e
e
xi
s
ti
ng
C
N
N
ha
s
a
la
r
ge
a
m
ount
of
c
om
put
in
g
a
nd
s
to
r
a
ge
s
pa
c
e
.
C
la
s
s
if
ic
a
ti
on
of
obj
e
c
ts
w
it
h
f
e
w
e
r
c
l
a
s
s
e
s
th
a
n
I
m
a
ge
ne
t,
th
e
ne
w
a
r
c
hi
te
c
tu
r
e
a
s
ne
e
de
d
w
il
l
be
b
e
tt
e
r
.
how
e
ve
r
,
s
om
e
r
e
s
e
a
r
c
he
r
s
pr
e
f
e
r
to
e
xpl
oi
t
C
N
N
,
s
u
c
h
a
s
tr
a
n
s
f
e
r
le
a
r
ni
ng
[
16]
,
f
in
e
-
tu
ni
ng
[
17]
,
or
p
r
uni
ng
[
18]
a
ga
in
s
t
e
xi
s
ti
ng
C
N
N
.
m
A
le
xne
t
a
nd mL
e
ne
t
a
r
e
C
N
N
a
r
c
hi
te
c
tu
r
e
s
s
pe
c
if
ic
a
ll
y de
ve
l
ope
d f
or
pa
r
ki
ng s
pa
c
e
c
la
s
s
if
ic
a
ti
on.
m
A
le
xne
t
a
nd
m
L
e
ne
t
a
dopt
th
e
A
le
xne
t
[
10]
a
nd
L
e
ne
t
[
19]
a
r
c
hi
te
c
tu
r
e
s
,
but
a
r
e
m
or
e
c
om
pa
c
t.
m
A
le
xne
t
ha
s
be
tt
e
r
a
c
c
ur
a
c
y
th
a
n
m
L
e
ne
t.
I
n
th
e
pa
r
ki
ng
lo
t
(
P
K
L
ot
)
a
nd
c
ons
ig
li
o
na
z
io
na
le
d
e
ll
e
r
ic
e
r
c
he
pa
r
ki
ng
(
C
N
R
P
a
r
kE
X
T
)
da
ta
s
e
t
s
,
m
A
le
xne
t
pr
oduc
e
s
m
or
e
th
a
n
90%
a
c
c
ur
a
c
y.
A
le
xne
t
a
nd
m
A
le
xne
t
e
xhi
bi
t
ne
a
r
ly
id
e
nt
ic
a
l
a
c
c
ur
a
c
y
le
ve
l
s
,
w
it
h
a
m
a
r
gi
na
l
di
f
f
e
r
e
nc
e
of
a
ppr
oxi
m
a
te
ly
1%
[
8]
.
S
ubs
e
que
nt
ly
,
C
a
r
N
e
t
w
a
s
de
ve
lo
pe
d
[
9]
,
a
nd
it
e
f
f
e
c
ti
ve
ly
e
nha
nc
e
d
th
e
a
c
c
ur
a
c
y
of
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on
w
it
hi
n
th
e
s
a
m
e
d
a
ta
s
e
t.
H
ow
e
ve
r
,
due
to
it
s
e
xt
e
n
s
iv
e
a
r
c
hi
te
c
tu
r
e
,
C
a
r
N
e
t
pos
s
e
s
s
e
s
a
hi
gh
num
be
r
of
pa
r
a
m
e
te
r
s
,
m
a
ki
ng
it
uns
ui
ta
bl
e
f
or
lo
w
-
c
om
put
in
g
de
vi
c
e
s
li
ke
th
e
R
a
s
pbe
r
r
y
P
i.
N
ot
s
a
ti
s
f
ie
d
w
it
h
th
is
,
be
c
a
us
e
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
f
or
m
A
le
xne
t
is
s
ti
ll
m
uc
h s
m
a
ll
e
r
t
ha
n t
ha
t
o
f
C
a
r
N
e
t,
f
in
e
-
tu
ni
ng
is
done
on mAle
xne
t.
F
in
e
-
tu
ni
ng
t
he
f
il
te
r
on
th
e
c
onvolut
io
n
la
ye
r
a
nd
th
e
a
c
ti
va
ti
on
f
unc
ti
on
c
a
n
in
c
r
e
a
s
e
th
e
a
c
c
ur
a
c
y
of
m
A
l
e
xne
t
[
20]
,
a
lt
hough
not
s
ig
ni
f
ic
a
nt
ly
.
A
f
te
r
th
a
t,
it
w
a
s
f
ound
th
a
t
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
[
21]
w
a
s
be
tt
e
r
in
a
c
c
ur
a
c
y
a
nd
s
pe
e
d
f
or
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on.
H
ow
e
ve
r
,
f
or
r
e
a
l
-
ti
m
e
in
f
or
m
a
ti
on
s
pe
e
d,
it
is
ne
c
e
s
s
a
r
y
to
de
ve
lo
p
C
N
N
a
r
c
hi
te
c
tu
r
e
th
a
t
c
a
n c
la
s
s
if
y pa
r
ki
ng s
pa
c
e
s
m
or
e
qui
c
kl
y a
nd a
c
c
ur
a
te
ly
.
T
he
C
N
N
c
om
put
a
ti
ona
l
pr
oc
e
s
s
is
ve
r
y
la
r
ge
,
e
s
pe
c
ia
ll
y
dur
in
g
tr
a
in
in
g,
s
o
i
t
c
a
n
ta
ke
qui
te
a
lo
ng
ti
m
e
[
22]
.
A
s
a
r
e
s
ul
t,
s
e
ve
r
a
l
r
e
s
e
a
r
c
he
r
s
ha
ve
f
oc
us
e
d
on
s
pe
e
di
ng
up
C
N
N
tr
a
in
in
g
a
nd
te
s
ti
ng
ti
m
e
s
[
23]
,
[
24]
.
M
ot
iv
a
te
d
by
th
e
ne
e
d
f
or
in
f
or
m
a
ti
on
on
th
e
a
va
il
a
bi
li
ty
of
pa
r
ki
ng
s
pa
c
e
s
,
th
e
r
e
s
e
a
r
c
h
pr
opos
e
d
in
th
is
pa
pe
r
is
to
pr
opos
e
a
ne
w
,
m
or
e
e
f
f
ic
ie
nt
a
r
c
hi
te
c
tu
r
e
.
T
he
pr
o
pos
e
d
a
r
c
hi
te
c
tu
r
e
c
a
n
be
a
ppl
ie
d
to
lo
w
-
c
os
t
c
om
put
in
g
de
vi
c
e
s
,
to
r
e
duc
e
c
la
s
s
if
ic
a
ti
on
c
os
ts
.
T
h
e
pr
opos
e
d
a
r
c
hi
te
c
tu
r
e
a
ls
o
ha
s
be
tt
e
r
a
c
c
ur
a
c
y
th
a
n
pr
e
vi
ous
s
tu
di
e
s
. I
n a
ddi
ti
on, t
hi
s
a
r
c
hi
te
c
tu
r
e
ha
s
a
numbe
r
of
pa
r
a
m
e
te
r
s
, f
lo
a
ti
ng point
ope
r
a
ti
ons
(
F
L
O
P
s
)
,
a
nd
s
m
a
ll
e
r
f
il
e
s
iz
e
s
o
th
a
t
it
is
f
a
s
te
r
in
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on.
T
hi
s
a
r
c
hi
te
c
tu
r
e
c
a
n
pr
ovi
de
in
f
or
m
a
ti
on
on t
he
a
va
il
a
bi
li
ty
of
pa
r
ki
ng s
pa
c
e
s
i
n r
e
a
l
-
ti
m
e
a
t
a
l
ow
c
o
s
t.
T
he
a
va
il
a
bi
li
ty
of
pa
r
ki
ng
s
pa
c
e
s
is
a
n
im
por
ta
nt
pa
r
t
of
a
s
m
a
r
t
pa
r
ki
ng
s
ys
te
m
.
D
r
iv
e
r
s
r
e
qui
r
e
qui
c
k
a
nd
a
c
c
ur
a
te
pa
r
ki
ng
s
pa
c
e
a
va
il
a
bi
li
ty
in
f
or
m
a
ti
on.
D
r
iv
e
r
s
m
a
y
qui
c
kl
y
s
e
le
c
t
th
e
ir
pr
e
f
e
r
r
e
d
pa
r
ki
ng
a
r
e
a
,
r
e
duc
in
g
th
e
ti
m
e
it
ta
ke
s
to
f
in
d
a
pa
r
k
in
g
s
pa
c
e
.
S
e
ns
or
s
or
vi
s
io
n
-
ba
s
e
d
te
c
hnol
ogy
a
r
e
of
te
n
ut
i
li
z
e
d
in
th
e
c
la
s
s
if
ic
a
ti
on
of
pa
r
ki
ng
s
pa
c
e
s
[
25]
.
B
a
s
e
d
on
budge
t
us
a
ge
,
th
e
a
ppl
ic
a
ti
on
of
s
e
ns
or
-
ba
s
e
d
te
c
hnol
ogy
f
or
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on
r
e
qui
r
e
s
th
e
m
or
e
e
xpe
n
s
iv
e
i
ns
ta
ll
a
ti
on
a
nd
m
a
in
te
na
nc
e
c
os
ts
th
a
n
u
s
in
g
vi
s
io
n
-
ba
s
e
d
te
c
hnol
ogy.
A
s
m
a
r
t
c
a
m
e
r
a
th
a
t
u
s
e
s
c
om
put
e
r
vi
s
io
n
c
a
n
c
la
s
s
if
y
m
u
lt
ip
le
pa
r
ki
ng
s
p
a
c
e
s
a
t
onc
e
. I
n a
ddi
ti
on, c
om
put
e
r
vi
s
io
n c
a
n ge
ne
r
a
te
s
om
e
a
ddi
ti
ona
l
in
f
or
m
a
ti
on.
C
om
put
e
r
vi
s
io
n
is
w
id
e
ly
a
ppl
ie
d
to
he
lp
hum
a
ns
in
va
r
io
us
f
i
e
ld
s
.
C
om
put
e
r
vi
s
io
n
w
or
ks
li
ke
th
e
hum
a
n
e
ye
.
C
om
put
e
r
s
c
a
n
s
e
e
a
nd
pr
ovi
de
in
f
or
m
a
ti
on
a
bout
w
ha
t
th
e
y
s
e
e
to
hum
a
ns
.
T
he
r
e
f
or
e
,
r
e
s
e
a
r
c
he
r
s
ha
ve
a
ppl
ie
d
c
om
put
e
r
vi
s
io
n
to
c
la
s
s
if
y
p
a
r
ki
ng
s
pa
c
e
s
.
R
e
s
e
a
r
c
he
r
s
de
v
e
lo
p
r
e
s
e
a
r
c
h
to
pr
ovi
de
f
a
s
t
a
nd
a
c
c
ur
a
te
pa
r
ki
ng
s
p
a
c
e
a
va
il
a
bi
li
ty
in
f
or
m
a
ti
on.
S
e
ve
r
a
l
c
om
put
e
r
vi
s
io
n
m
e
th
ods
f
or
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on
in
c
lu
de
m
a
r
ki
ng
pa
r
ki
ng
s
pa
c
e
s
,
li
ne
-
ba
s
e
d
de
t
e
c
ti
on,
a
nd
le
a
r
ni
ng
ba
s
e
.
R
e
s
e
a
r
c
h
r
e
la
te
d
to
pa
r
ki
ng s
pa
c
e
c
la
s
s
if
ic
a
ti
on i
s
pr
e
s
e
nt
e
d i
n T
a
bl
e
1
[
8]
, [
9]
, [
21
]
,
[
26]
‒[
34
]
.
T
a
bl
e
1
s
how
s
s
e
v
e
r
a
l
m
e
th
ods
f
or
c
om
put
e
r
vi
s
io
n
-
ba
s
e
d
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on.
m
A
le
xne
t
is
th
e
f
ir
s
t
C
N
N
us
e
d
f
or
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on
[
8]
.
A
f
te
r
th
a
t,
C
a
r
N
e
t
a
ppe
a
r
e
d
to
im
pr
ove
th
e
a
c
c
ur
a
c
y
of
m
A
le
xne
t,
but
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
a
nd
F
L
O
P
s
is
va
s
t.
T
he
c
la
s
s
if
ic
a
ti
on
a
c
c
ur
a
c
y
of
Y
ol
o+
M
obi
le
N
e
t
on
th
e
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t
r
e
a
c
he
s
up
to
98.97%
[
31]
.
G
D
S
N
-
I
C
c
la
s
s
if
ie
d
th
e
C
N
R
P
a
r
kE
X
T
s
ub
-
da
ta
s
e
t
w
it
h
a
n
a
c
c
ur
a
c
y
of
97.10
%
[
32
]
.
M
in
i
S
huf
f
le
ne
t
[
33]
by
p
r
uni
ng
S
huf
f
le
N
e
t
V
2
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
of
98.37. E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
i
s
a
n a
r
c
hi
te
c
tu
r
e
t
ha
t
c
a
n be
i
m
pl
e
m
e
nt
e
d on low
-
e
nd c
om
put
in
g de
vi
c
e
s
[
21]
. I
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4158
-
4168
4160
a
ddi
ti
on
to
it
s
a
c
c
ur
a
c
y
r
e
a
c
hi
ng
98.44%
,
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
a
ls
o
ha
s
f
e
w
e
r
pa
r
a
m
e
te
r
s
a
nd
F
L
O
P
s
.
H
ow
e
ve
r
,
a
c
c
ur
a
c
y s
ti
ll
ne
e
d
s
t
o be
i
m
pr
ove
d t
o pr
oduc
e
m
or
e
a
c
c
ur
a
te
a
nd f
a
s
te
r
i
nf
or
m
a
ti
on.
T
a
bl
e
1. R
e
s
e
a
r
c
h r
e
la
t
e
d t
o pa
r
ki
ng s
pa
c
e
c
la
s
s
if
ic
a
ti
on
M
e
t
hods
D
a
t
a
s
e
t
A
c
c
ur
a
c
y (
%
)
F
e
a
t
ur
e
poi
nt
de
t
e
c
t
i
on a
nd c
ol
or
hi
s
t
ogr
a
m
c
l
a
s
s
i
f
i
c
a
t
i
on
[
26]
S
e
l
f
C
ol
l
e
c
t
i
ng
85.20
S
uppor
t
V
e
c
t
or
M
a
c
hi
ne
[
27]
S
e
l
f
C
ol
l
e
c
t
i
ng
85.57
R
ot
a
t
i
on
-
ba
s
e
d m
e
t
hod a
nd m
os
a
i
c
3D
s
t
r
uc
t
ur
e
s
[
28]
S
e
l
f
C
ol
l
e
c
t
i
ng
90.00
T
hr
e
s
hol
d va
l
ue
[
29]
S
e
l
f
C
ol
l
e
c
t
i
ng
98.00
L
oc
a
l
B
i
na
r
y P
a
t
t
e
r
ns
(
L
B
P
)
a
nd L
oc
a
l
P
ha
s
e
Q
ua
nt
i
z
a
t
i
on (
L
P
Q
[
30]
P
K
L
ot
< 89.83
m
A
l
e
xne
t
[
8]
P
K
L
ot
a
nd C
N
R
P
a
r
kE
X
T
91.34
C
a
r
N
e
t
[
9]
C
N
R
P
a
r
kE
X
T
97.24
Y
ou L
ook O
nl
y O
nc
e
(
Y
ol
o)
a
nd M
obi
l
e
N
e
t
[
31
]
C
N
R
P
a
r
kE
X
T
98.97
G
r
a
s
s
m
a
nni
a
n D
e
e
p S
t
a
c
ki
ng N
e
t
w
or
k w
i
t
h I
l
l
um
i
na
t
i
on C
or
r
e
c
t
i
on
(
G
D
S
N
-
I
C
)
[
32]
C
N
R
P
a
r
kE
X
T
97.10
E
f
f
i
c
i
e
nt
P
a
r
k
i
ngN
e
t
[
21]
C
N
R
P
a
r
kE
X
T
98.44
M
i
ni
S
huf
f
l
e
ne
t
[
33]
C
N
R
P
a
r
kE
X
T
98.37
Y
ol
oV
5
-
P
r
e
t
r
a
i
ne
d
[
34]
P
K
L
ot
99.50
O
ur
m
ot
iv
a
ti
on
in
th
is
r
e
s
e
a
r
c
h
is
to
f
in
d
a
ne
w
a
r
c
hi
te
c
tu
r
e
t
ha
t
c
a
n
im
pr
ove
th
e
a
c
c
ur
a
c
y
P
K
L
o
t
c
la
s
s
if
ic
a
ti
on.
T
hi
s
a
r
c
hi
te
c
tu
r
e
a
ls
o
h
a
s
a
s
m
a
ll
e
r
num
be
r
of
pa
r
a
m
e
te
r
s
a
nd
F
L
O
P
s
s
o
it
i
s
f
a
s
te
r
in
c
la
s
s
if
ic
a
ti
on. T
hi
s
a
r
c
hi
te
c
tu
r
e
i
s
s
ui
ta
bl
e
f
or
im
pl
e
m
e
nt
a
ti
on on
budge
t
-
f
r
ie
ndl
y
c
om
put
in
g
de
vi
c
e
s
,
s
uc
h
a
s
th
e
R
a
s
pb
e
r
r
y
P
i,
s
o
it
c
a
n
be
im
pl
e
m
e
nt
e
d
in
lo
w
-
c
os
t
p
a
r
ki
ng
s
ys
te
m
s
a
nd
be
a
s
ol
ut
io
n
to
pr
obl
e
m
s
th
a
t
a
r
is
e
w
he
n dr
iv
in
g a
r
ound looki
ng f
or
a
pa
r
ki
ng s
pa
c
e
.
2.
M
E
T
H
O
D
T
hi
s
s
tu
dy
f
oc
us
e
s
on
c
r
e
a
ti
ng
a
m
or
e
e
f
f
ic
ie
nt
C
N
N
a
r
c
hi
te
c
tu
r
e
f
or
c
la
s
s
if
yi
ng
pa
r
ki
ng
s
pa
c
e
s
.
I
n
th
is
w
or
k, w
e
c
om
pa
r
e
d t
he
a
r
c
hi
te
c
tu
r
e
di
s
c
ove
r
e
d w
it
h e
xi
s
ti
ng a
ppr
oa
c
he
s
. W
e
m
a
ke
obs
e
r
va
ti
ons
i
n or
de
r
to
id
e
nt
if
y
a
r
c
hi
te
c
tu
r
e
s
th
a
t
a
r
e
m
or
e
e
f
f
ic
ie
nt
in
te
r
m
s
o
f
a
c
c
ur
a
c
y
a
nd
s
pe
e
d.
T
h
e
pr
opos
e
d
C
N
N
da
ta
s
e
t
a
nd
a
r
c
hi
te
c
tu
r
e
a
r
e
f
ur
th
e
r
de
s
c
r
ib
e
d i
n s
e
c
ti
ons
2.1
a
nd 2.2
.
2.1. Dat
as
e
t
s
C
N
N
a
r
c
hi
te
c
tu
r
e
te
s
ti
ng
us
in
g
th
e
P
K
L
ot
da
ta
s
e
t
a
nd
C
N
R
P
a
r
kE
X
T
.
T
he
P
K
L
ot
a
nd
C
N
R
P
a
r
k
-
E
X
T
da
ta
s
e
t
s
r
e
pr
e
s
e
nt
th
e
r
e
a
l
s
it
ua
ti
on
of
pa
r
ki
ng
s
pa
c
e
s
in
te
r
m
s
of
w
e
a
th
e
r
c
ondi
ti
on
s
,
c
a
m
e
r
a
a
ngl
e
s
,
a
nd
obs
ta
c
le
s
in
th
e
pa
r
ki
ng
s
pa
c
e
.
P
K
L
ot
ha
s
12,417
f
ul
l
im
a
ge
s
w
it
h
a
r
e
s
ol
ut
io
n
of
1280
720
pi
xe
ls
.
P
K
L
ot
c
ont
a
in
s
da
ta
s
e
t
s
ta
ke
n
f
r
om
two
pa
r
ki
ng
a
r
e
a
s
,
na
m
e
d
pa
r
ki
ng1
a
nd
pa
r
ki
ng2.
P
a
r
ki
ng1
is
a
pa
r
ki
ng
s
pa
c
e
da
ta
s
e
t
f
r
om
two
c
a
m
e
r
a
s
,
na
m
e
ly
pa
r
ki
ng1a
a
nd
p
a
r
ki
ng1b.
P
a
r
ki
ng2
is
a
pa
r
ki
ng
s
pa
c
e
da
ta
s
e
t
c
a
pt
ur
e
d
f
r
om
a
c
a
m
e
r
a
.
T
he
P
K
L
ot
da
ta
s
e
t
c
ons
i
s
ts
of
th
r
e
e
w
e
a
th
e
r
c
ondi
ti
ons
s
uc
h
a
s
s
unny,
r
a
in
y,
a
nd
c
lo
udy.
T
h
e
da
ta
s
e
t
c
ons
is
ts
of
two
la
be
ls
(
f
r
e
e
/b
us
y)
c
on
s
is
ti
ng
of
695,900
pa
r
ki
ng
s
pa
c
e
im
a
ge
s
.
T
he
im
a
g
e
on
pa
r
ki
ng1a
,
pa
r
ki
ng1b, a
nd pa
r
ki
ng2 a
r
e
a
s
s
how
n
in
F
ig
ur
e
s
1(
a
)
t
o 1(
c
)
:
(
a
)
(
b)
(
c
)
F
ig
ur
e
1. P
K
L
ot
da
ta
s
e
t
s
how
in
g t
w
o pa
r
ki
ng a
r
e
a
s
a
nd t
hr
e
e
c
a
m
e
r
a
poi
nt
s
, na
m
e
ly
(
a
)
P
a
r
ki
ng1a
,
(
b)
P
a
r
ki
ng1b, a
nd (
c
)
P
a
r
ki
ng2
[
30]
T
he
s
e
c
ond
da
ta
s
e
t
is
C
N
R
P
a
r
kE
X
T
c
on
s
is
ti
ng
of
C
N
R
P
a
r
k
a
nd
C
N
R
-
E
X
T
.
T
he
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t
or
ig
in
a
te
s
f
r
om
C
N
R
R
e
s
e
a
r
c
h i
n t
he
C
it
y of
P
is
a
, I
ta
ly
.
I
t
c
om
pr
is
e
s
242
c
om
pl
e
te
i
m
a
ge
s
a
nd 12,584
im
a
ge
s
s
pe
c
if
ic
a
ll
y
f
oc
us
in
g
on
pa
r
ki
ng
s
pa
c
e
s
in
th
e
C
N
R
P
a
r
k
s
u
bs
e
t.
A
ddi
ti
ona
ll
y,
C
N
R
+
E
X
T
in
c
lu
de
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
onc
is
e
C
a
r
N
e
t:
c
on
v
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
pa
r
k
in
g s
pac
e
c
la
s
s
if
ic
at
io
n (
M
ar
w
an R
am
li
)
4161
4,081 c
om
pl
e
te
i
m
a
ge
s
a
nd a
s
ubs
ta
nt
ia
l
144,965 pa
r
ki
ng s
pa
c
e
i
m
a
ge
s
c
a
pt
ur
e
d f
r
om
ni
ne
di
f
f
e
r
e
nt
c
a
m
e
r
a
s
.
C
N
R
P
a
r
k
c
ont
a
in
s
da
ta
s
e
ts
t
a
ke
n
dur
in
g
s
unny
w
e
a
th
e
r
,
w
hi
le
C
N
R
+
E
X
T
c
ont
a
in
s
da
ta
s
e
ts
c
a
pt
ur
e
d
in
s
unny,
r
a
in
y,
a
nd
c
lo
udy
w
e
a
th
e
r
.
C
N
R
P
a
r
kE
X
T
is
a
d
a
ta
s
e
t
w
it
h
m
a
ny
obs
ta
c
le
s
s
uc
h
a
s
ba
ll
s
,
tr
e
e
s
,
a
nd
la
m
p
pos
ts
.
C
N
R
P
a
r
kE
X
T
i
s
pr
e
s
e
nt
e
d i
n F
ig
ur
e
2
.
F
ig
ur
e
2. T
w
o out of
ni
ne
c
a
m
e
r
a
poi
nt
s
C
N
R
P
a
r
kE
X
T
[
8]
F
ig
ur
e
s
1
a
nd
2
a
r
e
f
ul
l
im
a
ge
s
.
T
he
f
ul
l
im
a
ge
is
c
r
oppe
d
a
c
c
or
di
ng
to
th
e
s
ha
pe
of
th
e
pa
r
ki
ng
s
pa
c
e
a
nd r
e
s
iz
e
d
t
o a
s
iz
e
of
224
224
pi
xe
ls
f
or
C
N
N
i
nput
.
T
he
c
r
o
ppe
d i
m
a
ge
s
r
e
s
ul
ts
a
r
e
s
how
n i
n
F
ig
ur
e
3.
T
he
e
va
lu
a
ti
on w
a
s
pe
r
f
or
m
e
d on a
c
om
put
e
r
e
qui
ppe
d w
it
h
th
e
f
ol
l
ow
in
g s
pe
c
if
ic
a
ti
ons
:
4 gi
ga
byt
e
s
of
R
A
M
, a
n
I
nt
e
l
C
or
e
i3
pr
oc
e
s
s
or
c
lo
c
ke
d
a
t
3.30
G
H
z
,
r
unni
ng
on
a
64
-
bi
t
ope
r
a
ti
ng
s
ys
te
m
.
T
he
te
s
ti
ng
pr
oc
e
s
s
e
m
pl
oye
d P
yt
hon, uti
li
z
in
g t
he
P
yT
or
c
h f
r
a
m
e
w
or
k.
F
ig
ur
e
3. R
e
s
ul
ts
of
i
m
a
ge
c
r
oppi
ng f
or
pa
r
ki
ng s
pa
c
e
s
T
he
da
t
a
s
e
t
i
s
di
vi
de
d
in
to
s
ub
s
e
ts
th
a
t
w
il
l
be
ut
il
iz
e
d
f
or
da
ta
tr
a
in
in
g
a
nd
te
s
ti
ng.
T
hi
s
s
tu
dy
e
m
pl
oye
d
id
e
nt
ic
a
l
da
ta
s
ha
r
in
g
a
nd
te
s
ti
ng
a
ppr
oa
c
he
s
to
pr
i
or
e
xpe
r
im
e
nt
s
on
th
e
m
A
le
xne
t
a
nd
C
a
r
N
e
t
m
ode
ls
.
T
a
bl
e
2
c
ont
a
in
s
de
ta
il
e
d i
nf
or
m
a
ti
on a
bout
t
he
da
ta
s
e
t
.
T
a
bl
e
2. D
e
ta
il
s
of
t
he
pa
r
ki
ng s
pa
c
e
d
a
ta
s
e
t
u
s
e
d a
s
t
r
a
in
in
g a
n
d t
e
s
t
da
ta
S
ubs
e
t
F
r
e
e
s
pa
c
e
B
us
y s
pa
c
e
T
ot
a
l
C
N
R
P
a
r
k A
2,549
3,622
6,171
C
N
R
P
a
r
k B
1,632
4,781
6,413
C
N
R
P
a
r
k
4,181
8,403
12,584
C
N
R
P
a
r
kE
X
T
T
r
a
i
n
46,877
47,616
94,493
C
N
R
P
a
r
kE
X
T
T
r
a
i
n C
1 C
8
21,769
16,784
38,553
C
N
R
P
a
r
kE
X
T
T
e
s
t
13,589
18,276
31,825
C
N
R
P
a
r
kE
X
T
65,658
79,307
144,965
P
K
L
ot
U
F
P
R
04
59,718
46,125
105,843
P
K
L
ot
U
F
P
R
05
68,359
97,426
165,785
P
K
L
ot
P
U
C
130,040
194,229
324,269
2.2. P
r
op
os
e
d
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
ar
c
h
it
e
c
t
u
r
e
T
he
C
N
N
a
r
c
hi
te
c
tu
r
e
w
a
s
bui
lt
to
de
te
c
t
c
a
r
s
in
pa
r
ki
ng
s
pa
c
e
s
a
nd
w
a
s
na
m
e
d
C
onc
is
e
C
a
r
N
e
t.
T
hi
s
a
r
c
hi
te
c
tu
r
e
is
c
a
ll
e
d
C
onc
is
e
C
a
r
N
e
t
be
c
a
us
e
it
is
th
e
s
a
m
e
a
s
it
s
f
unc
ti
on
a
s
a
c
a
r
pa
r
ki
ng
s
pa
c
e
de
te
c
to
r
.
C
onc
is
e
C
a
r
N
e
t
is
a
l
s
o
m
or
e
e
f
f
ic
ie
nt
th
a
n
C
a
r
N
e
t
[
31]
in
te
r
m
s
of
a
c
c
ur
a
c
y,
m
a
ny
pa
r
a
m
e
te
r
s
,
a
nd
F
L
O
P
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4158
-
4168
4162
T
he
C
onc
is
e
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
c
om
pr
is
e
s
s
e
ve
r
a
l
vi
ta
l
c
om
pone
nt
s
a
im
e
d
a
t
e
nha
nc
in
g
bot
h
a
c
c
ur
a
c
y
a
nd
c
la
s
s
if
ic
a
ti
on s
p
e
e
d. S
om
e
of
t
he
m
a
in
C
N
N
c
om
pon
e
nt
s
i
nc
lu
d
e
t
he
f
ol
lo
w
in
g:
−
T
he
a
r
c
hi
te
c
tu
r
e
e
m
pl
oys
3
3
a
nd
1
1
f
il
te
r
in
e
a
c
h
c
onvolut
io
n
to
e
f
f
e
c
ti
ve
ly
de
c
r
e
a
s
e
th
e
num
be
r
o
f
pa
r
a
m
e
te
r
s
.
−
A
r
c
hi
te
c
tu
r
e
ha
s
two
br
a
nc
he
s
w
it
h
di
f
f
e
r
e
nt
s
tr
uc
tu
r
e
s
.
T
hi
s
br
a
nc
h
s
e
r
ve
s
to
obt
a
in
f
e
a
tu
r
e
va
r
ia
ti
ons
t
o
im
pr
ove
C
N
N
a
c
c
ur
a
c
y.
−
A
ve
r
a
ge
P
ool
in
g
7
7
is
us
e
d
b
e
f
or
e
f
ul
ly
-
c
onne
c
te
d
to
r
e
duc
e
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
,
w
it
hout
lo
s
in
g
f
e
a
tu
r
e
s
.
C
onc
is
e
C
a
r
N
e
t
us
e
s
3
3
a
nd
1
1
c
onvolut
io
n
f
il
te
r
s
to
e
xt
r
a
c
t
s
pa
ti
a
l
f
e
a
tu
r
e
s
.
R
e
s
e
a
r
c
he
r
s
a
lwa
ys
us
e
3
3
c
onvolut
io
n
f
il
te
r
in
de
ve
lo
pi
ng
C
N
N
a
r
c
hi
te
c
tu
r
e
.
T
h
e
f
il
te
r
s
iz
e
of
3
3
is
be
tt
e
r
th
a
n
5
5
a
nd
7
7
[
35]
–
[
37]
in
m
os
t
c
a
s
e
s
. W
hi
le
t
he
1
1 f
il
te
r
i
s
us
e
d t
o r
e
tr
ie
ve
l
oc
a
l
f
e
a
tu
r
e
s
w
it
hout
i
nvol
vi
ng i
ts
ne
ig
hbor
s
.
T
he
1
1
f
il
te
r
is
li
ne
a
r
w
e
ig
ht
e
d
or
a
s
in
put
pr
oj
e
c
ti
on.
T
hi
s
f
il
te
r
is
a
ls
o
us
e
f
ul
f
or
in
c
r
e
a
s
in
g
or
de
c
r
e
a
s
in
g
th
e
num
be
r
of
out
put
m
a
p
f
e
a
tu
r
e
s
w
it
hout
c
h
a
ngi
ng
th
e
he
ig
ht
a
nd
w
id
th
of
th
e
f
e
a
tu
r
e
. C
onc
is
e
C
a
r
N
e
t
u
s
e
s
a
1
1
c
onvolut
io
n
f
il
te
r
to
ba
la
nc
e
th
e
out
put
of
e
a
c
h
br
a
nc
h
.
T
hi
s
1
1
f
il
te
r
ha
s
be
e
n
us
e
d
on
ne
twor
k
in
ne
twor
k (
N
I
N
)
[
38]
,
G
oogl
e
N
e
t
[
12]
, ne
ur
a
l
a
r
c
hi
te
c
tu
r
e
s
e
a
r
c
h ne
twor
k (
N
a
s
N
e
t)
[
39]
, a
nd unti
l
now
s
uc
h
a
s
m
ul
ti
pl
e
xi
ng ne
twor
k (
M
U
X
N
e
t)
[
40]
.
B
r
a
nc
he
s
on
th
e
C
onc
is
e
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
c
a
n
be
us
e
d
to
c
a
pt
ur
e
s
pa
ti
a
l
in
f
or
m
a
ti
on
a
t
va
r
io
us
s
pa
ti
a
l
r
e
s
ol
ut
io
n
r
a
nge
s
in
or
de
r
to
ge
ne
r
a
te
f
e
a
tu
r
e
va
r
ia
nt
s
.
T
he
popula
r
e
xi
s
ti
ng
C
N
N
w
it
h
br
a
nc
he
s
ha
s
s
uc
c
e
e
d
e
d
in
in
c
r
e
a
s
in
g
th
e
a
c
c
ur
a
c
y
f
or
c
la
s
s
if
yi
n
g
obj
e
c
ts
in
i
m
a
ge
s
s
uc
h a
s
G
oogl
e
N
e
t
[
12]
,
R
e
s
ne
xt
[
41]
,
s
que
e
z
e
-
a
nd
-
e
x
c
it
a
ti
on
ne
twor
k
(
S
E
N
e
t)
[
38]
,
a
nd
pa
th
w
a
y
a
s
s
oc
ia
te
d
s
pa
r
s
e
ne
twor
k
(
P
a
s
N
e
t)
[
42]
.
T
h
e
a
c
c
ur
a
c
y
of
C
N
N
a
r
c
hi
te
c
tu
r
e
s
w
it
h
br
a
nc
he
s
is
a
lwa
y
s
be
tt
e
r
th
a
n
br
a
nc
hl
e
s
s
a
r
c
hi
te
c
tu
r
e
s
s
uc
h
a
s
V
G
G
N
e
t
a
nd
A
le
xne
t.
T
he
br
a
nc
h
out
put
s
a
r
e
c
onc
a
te
na
te
d
be
f
or
e
be
in
g
pa
s
s
e
d
to
th
e
ne
xt
la
ye
r
.
I
n
a
ddi
t
io
n,
C
onc
is
e
C
a
r
N
e
t
a
ls
o u
s
e
s
t
he
r
e
c
ti
f
ie
d l
in
e
a
r
uni
t
(
R
e
L
U
)
a
c
ti
va
ti
on f
unc
ti
on a
nd ba
tc
h nor
m
a
li
z
a
ti
on (
B
N
)
.
O
ve
r
f
it
ti
ng
c
a
us
e
s
a
de
c
r
e
a
s
e
in
th
e
a
c
c
ur
a
c
y
of
C
N
N
f
or
c
la
s
s
if
ic
a
ti
on.
T
o
a
voi
d
ove
r
f
it
ti
ng,
th
e
r
e
s
e
a
r
c
he
r
s
u
s
e
d f
unc
ti
ons
a
f
te
r
a
c
ti
va
ti
on f
unc
ti
ons
s
u
c
h a
s
l
oc
a
l
r
e
s
pons
e
nor
m
a
li
z
a
ti
on (
L
R
N
)
, dr
opout a
nd
B
N
.
L
R
N
w
a
s
f
ir
s
t
in
tr
oduc
e
d
to
A
le
xne
t
[
10]
.
L
R
N
ha
s
s
uc
c
e
e
de
d
in
im
pr
ovi
ng
C
N
N
'
s
pe
r
f
or
m
a
nc
e
.
A
f
te
r
th
a
t,
"
dr
opout"
w
a
s
in
tr
oduc
e
d
in
2014
to
a
voi
d
ove
r
f
it
ti
ng
a
nd
i
m
pr
ove
C
N
N
pe
r
f
or
m
a
nc
e
[
43]
.
L
ik
e
d
r
opout
,
B
N
w
a
s
in
tr
oduc
e
d
in
2015
to
im
p
r
ove
C
N
N
a
c
c
ur
a
c
y
[
44]
.
B
N
im
pa
c
te
d
C
N
N
tr
a
in
in
g
f
a
s
te
r
w
it
h
hi
gh
a
c
c
ur
a
c
y. T
h
e
C
onc
i
s
e
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
i
s
s
how
n i
n F
ig
ur
e
4
.
F
ig
ur
e
4. C
onc
is
e
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
A
s
s
ho
w
n
in
F
ig
ur
e
4,
C
onv1
(
3/
2
6)
m
e
a
n
s
th
a
t
th
e
f
ir
s
t
c
onvol
ut
io
n
la
ye
r
us
e
s
a
3
3
f
il
te
r
w
it
h
s
tr
id
e
2
a
nd
th
e
num
be
r
of
out
put
s
6.
M
a
xpool
(
3/
2)
m
e
a
ns
M
a
xP
ool
in
g
w
it
h
a
f
il
te
r
s
iz
e
of
3
3
a
nd
th
e
num
be
r
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
onc
is
e
C
a
r
N
e
t:
c
on
v
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
pa
r
k
in
g s
pac
e
c
la
s
s
if
ic
at
io
n (
M
ar
w
an R
am
li
)
4163
s
tr
id
e
s
2.
T
he
C
onc
is
e
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
ha
s
th
r
e
e
c
onvolut
i
on
la
ye
r
s
,
na
m
e
ly
c
onv1,
c
onv2,
a
nd
c
onv3.
B
e
twe
e
n
c
onv1
a
nd
c
onv2
th
e
r
e
a
r
e
two
br
a
nc
h
e
s
,
n
a
m
e
ly
A
-
B
r
a
nc
h
a
nd
B
-
B
r
a
n
c
h.
A
-
B
r
a
nc
h
ha
s
th
r
e
e
c
onvolut
io
n
la
ye
r
s
,
na
m
e
ly
A
-
c
onv1,
A
-
c
onv2,
a
nd
A
-
c
onv
3.
E
a
c
h
c
onvolut
io
n
la
ye
r
in
th
e
A
-
B
r
a
nc
h
u
s
e
s
f
il
te
r
s
1
1,
3
3,
a
nd
1
1.
I
n
B
-
B
r
a
nc
h,
th
e
r
e
a
r
e
two
c
onvolut
io
n
la
ye
r
s
,
na
m
e
ly
B
-
c
onv1
a
nd
B
-
c
onv2.
B
-
c
onv1
us
e
s
a
1
1
f
il
te
r
a
nd B
-
c
onv1
us
e
s
a
3
3
f
il
te
r
.
T
he
out
p
ut
s
of
bot
h
br
a
nc
he
s
a
r
e
c
onc
a
t
e
na
te
d a
s
in
put
c
onv2.
C
onv3
us
e
s
B
N
a
nd
th
e
R
e
L
U
a
c
ti
va
ti
on
f
unc
ti
on.
C
onv3
out
put
is
poo
le
d
us
in
g
A
vgP
ool
in
g
w
it
h
a
f
il
te
r
s
iz
e
of
7
7
a
nd
s
tr
id
e
7.
P
ool
in
g
out
put
s
a
r
e
c
la
s
s
if
ie
d
a
s
th
e
f
ir
s
t
f
ul
ly
c
onne
c
te
d
(
F
C
1)
w
it
h
c
la
s
s
80
out
put
s
. F
C
1 f
ol
lo
w
e
d by F
C
2 w
it
h t
w
o c
la
s
s
out
put
s
a
c
c
or
di
ng
t
o t
he
de
s
ir
e
d pa
r
ki
ng s
pa
c
e
i
nf
or
m
a
ti
on.
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
m
A
le
xne
t
is
a
m
in
i
a
r
c
hi
te
c
tu
r
e
pr
une
d by
A
le
xne
t
[
8]
. m
A
le
xn
e
t
ha
s
a
n
a
ve
r
a
ge
a
c
c
ur
a
c
y
of
91.34%
f
or
t
he
P
K
L
ot
a
nd C
N
R
P
a
r
k
da
ta
s
e
ts
b
a
s
e
d
on dif
f
e
r
e
nt
t
e
s
t
s
e
t
s
. T
r
a
in
in
g on
t
he
P
K
L
ot
da
ta
s
e
t
a
nd
t
e
s
ti
ng
on
th
e
C
N
R
P
a
r
k
da
ta
s
e
t
or
vi
c
e
ve
r
s
a
r
e
s
ul
ts
in
lo
w
a
c
c
ur
a
c
y.
O
ve
r
f
it
ti
ng
a
nd
unde
r
f
it
ti
ng
c
a
us
e
lo
w
a
c
c
ur
a
c
y.
O
ve
r
f
it
ti
ng
is
w
he
n
C
N
N
r
e
c
ogni
z
e
s
tr
a
in
in
g
da
ta
to
o
m
uc
h
s
o
th
a
t
it
c
a
n'
t
r
e
c
ogni
z
e
te
s
t
da
ta
,
w
hi
le
unde
r
f
it
ti
ng
is
t
r
a
in
in
g
da
ta
t
ha
t
doe
s
not
r
e
pr
e
s
e
nt
a
ll
te
s
t
da
ta
[
45]
.
T
e
s
ti
ng
by
di
vi
di
ng t
he
da
ta
s
e
t
a
s
tr
a
in
in
g
a
nd t
e
s
t
da
ta
, m
A
le
xne
t
pr
oduc
e
s
hi
ghe
r
a
c
c
ur
a
c
y t
h
a
n t
r
a
in
in
g a
nd t
e
s
ti
ng on dif
f
e
r
e
nt
da
ta
s
e
ts
.
T
he
r
e
f
or
e
, i
n
th
is
t
e
s
t,
w
e
di
d not us
e
a
di
f
f
e
r
e
nt
da
ta
s
e
t.
W
e
te
s
te
d
th
e
a
r
c
hi
te
c
tu
r
a
l
c
a
pa
bi
li
ti
e
s
s
te
p
by
s
te
p.
I
n
th
e
f
ir
s
t
s
ta
ge
,
w
e
te
s
te
d
C
onc
i
s
e
C
a
r
N
e
t
w
it
h
th
e
C
N
R
P
a
r
kE
X
T
d
a
ta
s
e
t.
A
t
th
is
s
ta
ge
,
w
e
c
om
pa
r
e
th
e
a
c
c
u
r
a
c
y
of
C
onc
is
e
C
a
r
N
e
t
w
it
h
m
A
le
xne
t.
I
n
th
e
s
e
c
ond
s
ta
ge
,
w
e
tr
y
to
in
c
r
e
a
s
e
th
e
num
be
r
of
out
put
s
in
s
e
ve
r
a
l
la
ye
r
s
of
C
onc
is
e
C
a
r
N
e
t
c
onvolut
io
n.
C
onc
is
e
C
a
r
N
e
t
f
in
e
-
tu
ni
ng
s
e
r
ve
s
to
im
p
r
ove
C
N
N
a
c
c
ur
a
c
y.
I
n
th
e
th
ir
d
s
ta
ge
,
w
e
us
e
th
e
be
s
t
C
onc
is
e
C
a
r
N
e
t
a
nd
c
om
pa
r
e
it
a
ga
in
s
t
s
e
ve
r
a
l
ot
he
r
m
e
th
ods
w
it
h
a
ll
da
ta
s
e
ts
.
I
n
th
e
f
our
th
s
te
p,
w
e
c
om
pa
r
e
th
e
e
f
f
ic
ie
nc
y
of
e
a
c
h a
r
c
hi
te
c
tu
r
e
by c
a
l
c
ul
a
ti
ng t
he
numbe
r
of
pa
r
a
m
e
te
r
s
a
nd
F
L
O
P
s
.
3.1. Com
p
ar
is
on
o
f
C
on
c
is
e
C
a
r
N
e
t
an
d
m
A
le
xn
e
t
T
he
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t
is
th
e
da
ta
u
s
e
d
to
te
s
t
C
onc
is
e
C
a
r
N
e
t.
T
h
e
r
e
s
ul
ti
ng
a
c
c
ur
a
c
y
of
C
onc
is
e
C
a
r
N
e
t
is
c
om
pa
r
e
d
w
it
h
m
A
le
xne
t
to
s
e
e
th
e
in
c
r
e
a
s
e
in
a
c
c
ur
a
c
y
obt
a
in
e
d.
T
he
c
om
pos
it
io
n
of
th
e
tr
a
in
in
g
da
ta
us
e
d
is
C
N
R
P
a
r
k,
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
1
-
C
8,
a
nd
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
. T
he
te
s
t
da
t
a
us
e
d
is
C
N
R
P
a
r
kE
X
T
T
E
S
T
.
T
he
c
om
pa
r
is
on
of
th
e
a
c
c
ur
a
c
y
of
C
onc
is
e
C
a
r
N
e
t
a
nd
m
A
l
e
xne
t
is
s
how
n
in
T
a
bl
e
3.
T
a
bl
e
3. A
c
c
ur
a
c
y C
om
pa
r
is
on be
tw
e
e
n C
onc
i
s
e
C
a
r
N
e
t
a
nd m
A
le
xN
e
t
T
r
a
i
n da
t
a
s
e
t
T
e
s
t
da
t
a
s
e
t
M
e
t
hods
m
A
l
e
xne
t
(
%
)
C
onc
i
s
e
C
a
r
N
e
t
(
%
)
C
N
R
P
a
r
k
C
N
R
P
a
r
kE
X
T
T
E
S
T
93.52
92.93
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
1
-
C8
C
N
R
P
a
r
kE
X
T
T
E
S
T
95.88
96.93
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
N
R
P
a
r
kE
X
T
T
E
S
T
97.70
98.30
A
ve
r
a
ge
a
c
c
ur
a
c
y
95.70
96.05
B
a
s
e
d
on
T
a
bl
e
3,
th
e
a
c
c
ur
a
c
y
of
C
on
c
is
e
C
a
r
N
e
t
is
be
tt
e
r
th
a
n
m
A
le
xne
t
in
two
t
e
s
ts
,
na
m
e
ly
th
e
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
1
-
C
8
a
nd
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
tr
a
i
ni
ng
da
ta
.
T
e
s
ti
ng
on
C
N
R
P
a
r
k
tr
a
in
in
g
da
ta
,
m
A
le
xne
t'
s
a
c
c
ur
a
c
y
i
s
be
tt
e
r
th
a
n
C
onc
i
s
e
C
a
r
N
e
t.
C
onc
i
s
e
C
a
r
N
e
t
is
0.35%
b
e
tt
e
r
th
a
n
m
A
l
e
xne
t
on a
ve
r
a
ge
a
c
c
ur
a
c
y.
C
onc
is
e
C
a
r
N
e
t'
s
a
c
c
ur
a
c
y
is
be
tt
e
r
,
but
it
s
ti
ll
ne
e
d
s
to
be
m
or
e
e
f
f
ic
ie
nt
.
T
he
r
e
f
or
e
,
f
in
e
-
tu
ni
ng
C
onc
is
e
C
a
r
N
e
t
is
ne
c
e
s
s
a
r
y.
3.2. F
in
e
t
u
n
i
n
g C
on
c
is
e
C
ar
N
e
t
F
in
e
t
uni
ng
C
onc
is
e
C
a
r
N
e
t
is
a
w
a
y
to
im
pr
ove
th
e
a
c
c
ur
a
c
y
of
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on.
W
e
in
c
r
e
a
s
e
t
he
numbe
r
of
f
il
te
r
s
i
n A
-
B
r
a
nc
h a
nd B
-
B
r
a
nc
h t
o 15
f
il
te
r
s
a
nd t
he
numbe
r
of
f
il
te
r
s
i
n c
onv2
to
30
f
il
te
r
s
.
T
hi
s
f
in
e
-
tu
ni
ng
c
e
r
ta
in
ly
in
c
r
e
a
s
e
s
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
by
a
lm
os
t
double
th
e
or
ig
in
a
l,
s
o
w
e
na
m
e
d
th
is
a
r
c
hi
te
c
tu
r
e
C
on
c
is
e
C
a
r
N
e
t
2x.
C
onc
i
s
e
C
a
r
N
e
t
2x
a
r
c
hi
te
c
tu
r
e
is
s
how
n
in
F
ig
ur
e
5.
A
s
s
e
e
n
in
F
ig
ur
e
5,
th
e
a
ddi
ti
on
of
th
e
num
be
r
of
f
il
te
r
s
is
e
xpe
c
te
d
to
pr
oduc
e
a
la
r
ge
f
e
a
tu
r
e
va
r
ia
n
c
e
,
to
in
c
r
e
a
s
e
a
c
c
ur
a
c
y.
C
onc
is
e
C
a
r
N
e
t2
x
doe
s
not
c
ha
nge
th
e
s
tr
uc
tu
r
e
of
C
onc
is
e
C
a
r
N
e
t.
T
he
di
f
f
e
r
e
nc
e
b
e
twe
e
n
C
onc
is
e
C
a
r
N
e
t2
x a
nd
C
onc
is
e
C
a
r
N
e
t
onl
y
l
ie
s
i
n t
h
e
numbe
r
o
f
f
il
te
r
s
us
e
d i
n br
a
nc
h
c
onvolut
io
n a
nd c
onv2.
T
he
te
s
t
s
c
onduc
te
d
ut
il
iz
e
d
id
e
nt
ic
a
l
da
ta
to
th
a
t
of
th
e
pr
e
vi
ous
r
e
s
e
a
r
c
he
r
s
.
T
he
tr
a
in
in
g
da
ta
c
om
pr
is
e
d C
N
R
P
a
r
k, C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
1
-
C
8, a
nd
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
, w
hi
le
t
he
t
e
s
t
da
ta
c
on
s
is
te
d
of
C
N
R
P
a
r
kE
X
T
te
s
t.
C
onc
is
e
C
a
r
N
e
t2
x
de
m
ons
tr
a
te
d
s
upe
r
io
r
a
c
c
u
r
a
c
y
c
om
pa
r
e
d
to
a
ll
te
s
t
s
c
he
m
e
s
.
C
onc
is
e
C
a
r
N
e
t2
x
ha
s
a
n
a
ve
r
a
ge
a
c
c
ur
a
c
y
of
96.82%
,
1.
12%
be
tt
e
r
th
a
n
m
A
le
xne
t
a
nd
0.4%
th
a
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4158
-
4168
4164
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t.
A
c
om
pa
r
is
on
of
th
e
a
c
c
ur
a
c
y
of
C
on
c
is
e
C
a
r
N
e
t2
x,
C
onc
is
e
C
a
r
N
e
t,
m
A
le
xne
t,
a
nd
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
is
pr
e
s
e
nt
e
d i
n T
a
bl
e
4.
F
ig
ur
e
5. C
onc
is
e
C
a
r
N
e
t
2x a
r
c
hi
te
c
tu
r
e
T
a
bl
e
4. A
c
c
ur
a
c
y c
om
pa
r
is
on of
C
onc
i
s
e
C
a
r
N
e
t2
x, C
onc
is
e
C
a
r
N
e
t,
m
A
le
xN
e
t,
a
nd E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
T
r
a
i
n da
t
a
s
e
t
M
e
t
hods
m
A
l
e
xne
t
(%)
E
f
f
i
c
i
e
nt
P
a
r
k
i
ngN
e
t
(%)
C
onc
i
s
e
C
a
r
N
e
t
(%)
C
onc
i
s
e
C
a
r
N
e
t
2x
(%)
C
N
R
P
a
r
k
93.52
93.97
92.93
94.30
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
C
1
-
C8
95.88
96.85
96.93
97.55
C
N
R
P
a
r
kE
X
T
T
R
A
I
N
97.70
98.54
98.30
98.60
A
ve
r
a
ge
95.70
96.42
96.05
96.82
T
he
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t
is
a
da
ta
s
e
t
ta
ke
n
f
r
om
ni
ne
di
f
f
e
r
e
nt
c
a
m
e
r
a
s
,
s
o
it
ha
s
s
li
ght
ly
di
f
f
e
r
e
n
t
f
e
a
tu
r
e
s
.
T
r
a
in
in
g
w
it
h
one
s
ubs
e
t
a
nd
te
s
ti
ng
w
it
h
a
not
he
r
s
ubs
e
t
is
not
th
e
r
ig
ht
s
ol
ut
io
n
f
or
C
N
N
im
pl
e
m
e
nt
a
ti
on,
th
is
c
a
us
e
s
ove
r
f
it
ti
ng
or
unde
r
f
it
ti
ng.
T
o
ov
e
r
c
om
e
th
is
,
w
e
us
e
a
ll
th
e
da
ta
to
be
a
bl
e
to
r
e
pr
e
s
e
nt
a
ll
th
e
c
a
m
e
r
a
s
.
A
ll
C
N
R
P
a
r
kE
X
T
da
ta
s
e
ts
w
e
r
e
c
o
m
bi
ne
d
a
nd
di
vi
de
d
in
to
80%
tr
a
in
in
g
d
a
ta
a
nd
20%
a
s
va
li
da
ti
on da
ta
. T
hi
s
t
e
s
t
is
c
lo
s
e
r
t
o t
he
r
e
a
l
im
pl
e
m
e
nt
a
ti
on of
pa
r
ki
ng.
T
he
te
s
t
r
e
s
ul
ts
s
how
th
a
t
C
onc
i
s
e
C
a
r
N
e
t2
x
ha
s
b
e
tt
e
r
a
c
c
u
r
a
c
y
th
a
n
s
om
e
ot
he
r
m
e
th
ods
.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
r
a
t
e
on
C
on
c
is
e
C
a
r
N
e
t
r
e
a
c
h
e
s
99.37%
,
b
e
tt
e
r
th
a
n
m
A
le
xne
t
a
nd
m
uc
h
be
tt
e
r
th
a
n
C
a
r
N
e
t.
C
onc
is
e
C
a
r
N
e
t
m
a
tc
he
s
t
he
a
c
c
ur
a
c
y of
Y
ol
o+
M
obi
le
N
e
t
w
hi
c
h i
s
a
w
e
ll
-
known e
xi
s
ti
ng C
N
N
, e
ve
n s
li
ght
ly
be
tt
e
r
. A
c
om
pa
r
is
on of
t
he
a
c
c
ur
a
c
y of
s
e
ve
r
a
l
m
e
th
ods
i
s
s
ho
w
n i
n T
a
bl
e
5.
T
a
bl
e
5. V
a
li
da
ti
on a
c
c
ur
a
c
y r
e
s
ul
t
s
on t
he
C
N
R
P
a
r
kE
X
T
d
a
ta
s
e
t
M
e
t
hods
A
c
c
ur
a
c
y (
%
)
Re
-
t
r
a
i
ne
d m
A
l
e
xne
t
98.13
Y
ol
o+M
obi
l
e
N
e
t
[
31]
98.97
C
a
r
N
e
t
[
9]
97.24
E
f
f
i
c
i
e
nt
P
a
r
k
i
ngN
e
t
[
21]
98.44
M
i
ni
S
huf
f
l
e
ne
t
[
33]
98.37
C
onc
i
s
e
C
a
r
N
e
t
2x
99.37
3.3. Con
c
is
e
C
ar
N
e
t
2x on
P
K
L
ot
d
at
as
e
t
T
e
s
ti
ng
on
th
e
C
N
R
P
a
r
kE
X
T
da
t
a
s
e
t
s
ho
w
s
th
a
t
C
onc
i
s
e
C
a
r
N
e
t
is
be
tt
e
r
th
a
n
m
A
le
xne
t
a
nd
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t.
T
he
ne
xt
te
s
t
us
e
s
th
e
P
K
L
ot
da
ta
s
e
t
[
30]
w
hi
c
h
ha
s
th
r
e
e
s
ubs
e
ts
.
E
a
c
h
s
ubs
e
t
is
c
a
pt
ur
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
onc
is
e
C
a
r
N
e
t:
c
on
v
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
pa
r
k
in
g s
pac
e
c
la
s
s
if
ic
at
io
n (
M
ar
w
an R
am
li
)
4165
by
a
c
a
m
e
r
a
.
T
he
P
K
L
ot
s
ubs
e
ts
na
m
e
d
U
F
P
B
R
04,
U
F
P
B
R
0
5,
a
nd
P
U
C
a
r
e
de
r
iv
e
d
f
r
om
th
r
e
e
di
f
f
e
r
e
n
t
c
a
m
e
r
a
s
f
r
om
two
pa
r
ki
ng
a
r
e
a
s
.
P
K
L
ot
r
e
pr
e
s
e
nt
s
a
s
c
e
na
r
io
w
h
e
r
e
one
pa
r
ki
ng
s
pa
c
e
is
ut
il
iz
e
d
a
s
th
e
tr
a
in
in
g
da
ta
s
e
t,
w
hi
le
a
not
he
r
pa
r
ki
ng
s
p
a
c
e
s
e
r
ve
s
a
s
t
he
te
s
ti
ng
da
ta
s
e
t.
T
he
P
K
L
ot
da
ta
s
e
t
i
s
e
a
s
ie
r
to
id
e
nt
if
y
th
a
n
th
e
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t.
C
onc
is
e
C
a
r
N
e
t2
x
a
c
c
ur
a
c
y
is
gr
e
a
te
r
th
a
n
99.9%
f
or
da
ta
s
e
ts
w
it
h
th
e
s
a
m
e
c
a
m
e
r
a
s
our
c
e
. T
he
a
c
c
ur
a
c
y
of
C
onc
i
s
e
C
a
r
N
e
t2
x
is
b
e
tt
e
r
th
a
n
th
e
a
c
c
ur
a
c
y
of
m
A
le
xne
t
a
nd
C
a
r
N
e
t.
T
h
e
r
e
s
ul
t
s
of
th
e
c
om
pa
r
is
on of
a
c
c
ur
a
c
y a
r
e
pr
e
s
e
nt
e
d i
n T
a
bl
e
6.
T
a
bl
e
6. A
c
c
ur
a
c
y c
om
pa
r
is
on on the
P
K
L
ot
da
ta
s
e
t
T
r
a
i
n D
a
t
a
s
e
t
T
e
s
t
da
t
a
s
e
t
M
e
t
hods
M
A
l
e
xne
t
(%)
C
onc
i
s
e
C
a
r
N
e
t
2x
(%)
C
a
r
N
e
t
(%)
U
F
P
R
04
U
F
P
R
04
99.54
99.93
95.60
U
F
P
R
05
93.23
97.68
97.60
P
U
C
98.27
99.02
98.30
U
F
P
R
05
U
F
P
R
04
93.69
93.90
95.20
U
F
P
R
05
99.49
99.93
97.50
P
U
C
92.72
96.86
98.40
P
U
C
U
F
P
R
04
98.03
98.92
94.40
U
F
P
R
05
96.00
96.07
97.60
P
U
C
99.90
99.92
98.80
A
ve
r
a
ge
a
c
c
ur
a
c
y
96.74
97.27
97.04
A
s
s
e
e
n
i
n T
a
bl
e
6
,
t
e
s
ti
ng w
it
h
9 s
c
h
e
m
e
s
, C
o
nc
i
s
e
C
a
r
N
e
t
2x
6
t
im
e
s
a
nd C
a
r
N
e
t
3
ti
m
e
s
be
i
ng be
tt
e
r
th
a
n
a
ll
m
e
t
hod
s
.
C
o
nc
is
e
C
a
r
N
e
t2
x
ha
s
b
e
tt
e
r
a
v
e
r
a
g
e
a
c
c
ur
a
c
y
th
a
n
m
A
le
xn
e
t
a
nd
C
a
r
N
e
t
.
C
o
n
c
i
s
e
C
a
r
N
e
t
2x
ha
s
a
n
a
v
e
r
a
g
e
a
c
c
ur
a
c
y
of
9
7.
27%
,
w
hi
l
e
C
a
r
N
e
t
i
s
97
.04%
a
nd
m
A
le
xn
e
t
is
on
ly
96.
74%
.
C
on
c
i
s
e
C
a
r
N
e
t
a
c
hi
e
v
e
s
a
m
a
xi
m
um
a
c
c
ur
a
c
y
o
f
9
9.9
3%
,
s
ur
pa
s
s
i
ng
l
a
r
g
e
r
a
r
c
h
it
e
c
tu
r
e
s
s
u
c
h
a
s
Y
O
L
O
v
5
-
p
r
e
tr
a
i
ne
d
[
34]
,
w
h
ic
h
a
tt
a
i
ns
9
9.5
0%
a
c
c
ur
a
c
y
.
T
he
r
e
f
or
e
,
C
on
c
i
s
e
C
a
r
N
e
t
i
s
b
e
tt
e
r
a
t
c
l
a
s
s
if
yi
ng
p
a
r
ki
ng
s
pa
c
e
s
o
n t
h
e
P
K
L
ot
d
a
t
a
s
e
t
.
3.4. Com
p
ar
is
on
o
f
n
u
m
b
e
r
o
f
p
ar
am
e
t
e
r
s
an
d
f
lo
at
in
g p
oi
n
t
op
e
r
at
io
n
s
T
he
m
A
le
xn
e
t
a
r
c
hi
te
c
tu
r
e
h
a
s
a
to
ta
l
of
32,332
pa
r
a
m
e
te
r
s
,
s
o t
he
c
la
s
s
if
ic
a
ti
on
pr
oc
e
s
s
i
s
f
a
s
te
r
.
I
n
c
ont
r
a
s
t
to
C
a
r
N
e
t,
w
hi
c
h
ha
s
a
ve
r
y
w
id
e
a
r
c
hi
te
c
tu
r
e
,
s
o
th
e
pa
r
a
m
e
te
r
s
i
z
e
i
s
to
o
l
a
r
ge
.
C
a
r
ne
t
ha
s
th
r
e
e
c
onvolut
io
n
la
ye
r
s
,
th
e
f
ir
s
t
c
onvolut
io
n
l
a
ye
r
us
e
s
96
f
il
te
r
s
w
it
h
a
s
iz
e
of
11
11
a
nd
is
f
ol
lo
w
e
d
by
m
a
x
-
pool
in
g
w
it
h
2
2
f
il
te
r
s
.
T
he
s
e
c
ond
a
nd
th
ir
d
c
onvolut
io
n
la
ye
r
s
a
r
e
th
e
s
a
m
e
a
s
th
e
f
ir
s
t
la
ye
r
,
but
th
e
num
be
r
of
f
il
te
r
s
i
s
192 in t
he
s
e
c
ond c
onvolut
io
n l
a
ye
r
a
nd 384 in t
he
c
onvolut
io
n l
a
ye
r
. T
hi
r
d
C
a
r
N
e
t
ha
s
t
hr
e
e
f
ul
ly
c
onne
c
te
d l
a
ye
r
s
, t
he
f
ir
s
t
a
nd
s
e
c
ond la
y
e
r
s
us
e
4096 tot
a
l
out
p
ut
s
a
nd t
he
t
hi
r
d f
ul
ly
c
onne
c
te
d l
a
ye
r
ha
s
t
w
o
out
put
s
[
9]
.
T
he
C
a
r
N
e
t
a
r
c
hi
te
c
tu
r
e
ha
s
64,139,354
p
a
r
a
m
e
te
r
s
.
C
a
r
N
e
t
ha
s
ve
r
y
la
r
ge
pa
r
a
m
e
te
r
s
,
c
a
us
in
g
r
e
la
ti
ve
ly
la
r
ge
f
il
e
s
iz
e
s
.
I
n
a
ddi
ti
on,
th
e
c
om
put
a
ti
ona
l
pr
oc
e
s
s
is
ve
r
y
la
r
ge
a
nd
r
e
qui
r
e
s
c
om
put
in
g
de
vi
c
e
s
w
it
h
a
de
qua
te
s
pe
c
if
ic
a
ti
ons
to
c
a
r
r
y
out
th
e
tr
a
in
in
g
a
nd
te
s
ti
ng
pr
oc
e
s
s
.
C
a
r
N
e
t
ha
s
a
l
a
r
ge
r
num
be
r
of
pa
r
a
m
e
te
r
s
th
a
n
a
ll
a
r
c
hi
te
c
tu
r
e
s
f
or
pa
r
ki
ng
s
pa
c
e
c
l
a
s
s
if
ic
a
ti
on,
e
ve
n
a
ga
in
s
t
Y
ol
o
+
M
obi
le
N
e
t.
M
obi
le
N
e
t
pa
r
a
m
e
te
r
s
onl
y
r
a
nge
f
r
om
4,000,000
pa
r
a
m
e
te
r
s
[
31]
.
C
om
pa
r
e
d
to
ot
he
r
a
r
c
hi
te
c
tu
r
e
s
,
C
onc
is
e
C
a
r
N
e
t2
x
ha
s
th
e
s
m
a
ll
e
s
t
pa
r
a
m
e
te
r
s
.
C
onc
is
e
C
a
r
N
e
t2
x
h
a
s
14,642
pa
r
a
m
e
te
r
s
a
nd
is
s
m
a
ll
e
r
th
a
n
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t
w
it
h
21,834
pa
r
a
m
e
te
r
s
[
21]
.
T
hi
s
is
c
a
us
e
d
by
th
e
s
iz
e
of
th
e
f
il
te
r
us
e
d
3
3
a
nd
1
1.
T
he
c
om
pa
r
is
on
of
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
i
s
pr
e
s
e
nt
e
d i
n
T
a
bl
e
7.
T
a
bl
e
7. P
a
r
a
m
e
te
r
c
om
pa
r
is
on
M
e
t
hods
P
a
r
a
m
e
t
e
r
s
(
m
i
l
l
i
on)
m
A
l
e
xne
t
0.032
C
a
r
N
e
t
64.139
Y
ol
o+M
obi
l
e
N
e
t
4.000
E
f
f
i
c
i
e
nt
P
a
r
k
i
ngN
e
t
0.022
C
onc
i
s
e
C
a
r
N
e
t
2 x
0.015
C
onc
i
s
e
C
a
r
N
e
t
0.008
B
a
s
e
d
on
T
a
bl
e
7,
C
onc
is
e
C
a
r
N
e
t
ha
s
th
e
f
e
w
e
s
t
pa
r
a
m
e
te
r
s
.
C
onc
is
e
C
a
r
N
e
t2
x
is
onl
y
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
f
or
m
A
le
xne
t,
1/
273
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
f
or
Y
ol
o+
m
obi
le
ne
t,
a
nd
2/
3
th
e
num
be
r
of
pa
r
a
m
e
te
r
s
f
or
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t.
C
a
r
N
e
t
ha
s
th
e
la
r
ge
s
t
nu
m
be
r
of
pa
r
a
m
e
te
r
s
,
it
is
di
f
f
ic
ul
t
to
im
pl
e
m
e
nt
on l
ow
-
c
om
put
in
g de
vi
c
e
s
s
uc
h
a
s
R
a
s
pb
e
r
r
y P
i.
B
a
s
e
d on the
num
be
r
of
pa
r
a
m
e
te
r
s
, C
onc
is
e
C
a
r
N
e
t
is
b
e
tt
e
r
th
a
n
ot
he
r
a
r
c
hi
te
c
tu
r
e
s
to
b
e
im
pl
e
m
e
nt
e
d
on
lo
w
-
e
nd
c
om
put
in
g
de
vi
c
e
s
.
T
h
e
f
ol
lo
w
in
g
c
om
pa
r
is
on
of
f
il
e
s
iz
e
s
a
nd F
L
O
P
s
f
or
e
a
c
h a
r
c
hi
te
c
tu
r
e
i
s
pr
e
s
e
nt
e
d i
n T
a
bl
e
8.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
. 13, No. 4, D
e
c
e
m
be
r
2024:
4158
-
4168
4166
T
a
be
l
8.
C
om
pa
r
is
on of
f
il
e
s
iz
e
a
nd F
L
O
P
s
M
e
t
hods
F
i
l
e
S
i
z
e
F
L
O
P
s
m
A
l
e
xne
t
133K
b
21,329,778
C
a
r
N
e
t
257M
b
1,325,155,234
E
f
f
i
c
i
e
nt
P
a
r
k
i
ngN
e
t
90K
b
8,469,776
C
onc
i
s
e
C
a
r
N
e
t
2x
73K
b
6,084,818
C
onc
i
s
e
C
a
r
N
e
t
45K
b
3,963,916
B
a
s
e
d
on T
a
bl
e
8,
C
onc
is
e
C
a
r
N
e
t
ha
s
a
s
m
a
ll
f
il
e
s
iz
e
of
45
K
B
a
nd
C
onc
is
e
C
a
r
N
e
t2
x
onl
y
73
K
B
.
m
A
le
xne
t
ha
s
a
f
il
e
s
iz
e
of
133
K
B
,
a
lm
o
s
t
2
ti
m
e
s
th
e
s
iz
e
of
C
onc
is
e
C
a
r
N
e
t2
x.
B
a
s
e
d
on
th
e
num
be
r
of
F
L
O
P
s
,
C
onc
is
e
C
a
r
N
e
t2
x
is
a
ls
o
s
m
a
ll
e
r
th
a
n
m
A
le
xne
t
a
nd
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t,
w
hi
c
h
is
le
s
s
th
a
n
1/
3
th
e
num
be
r
of
F
L
O
P
s
of
m
A
le
xne
t
a
nd
F
L
O
P
s
of
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t.
T
he
f
e
w
e
r
num
be
r
o
f
pa
r
a
m
e
te
r
s
a
nd
F
L
O
P
s
w
il
l
s
pe
e
d
up
th
e
c
la
s
s
if
ic
a
ti
on
ti
m
e
,
w
hi
c
h
is
ne
e
de
d
in
r
e
a
l
-
ti
m
e
s
ys
te
m
s
.
S
pe
e
d
te
s
ti
ng
by
c
om
pa
r
in
g
ti
m
e
is
r
a
r
e
ly
don
e
by
r
e
s
e
a
r
c
he
r
s
be
c
a
us
e
th
e
c
e
nt
r
a
l
pr
oc
e
s
s
in
g
uni
t
(
C
P
U
)
ta
s
k
w
he
n
pe
r
f
or
m
in
g
c
om
put
a
ti
ons
is
unpr
e
di
c
ta
bl
e
.
T
he
r
e
f
or
e
,
th
e
te
s
t
s
w
e
r
e
a
lt
e
r
na
te
d
a
nd
r
e
pe
a
te
d
te
n
ti
m
e
s
te
s
ti
ng
f
or
e
a
c
h
m
e
th
od. T
he
r
e
s
ul
ts
of
t
he
c
om
p
a
r
is
on of
t
he
c
la
s
s
if
ic
a
ti
on s
pe
e
d of
a
pa
r
ki
ng s
pa
c
e
a
r
e
pr
e
s
e
nt
e
d i
n T
a
bl
e
9.
T
a
bl
e
9. P
a
r
ki
ng s
pa
c
e
c
la
s
s
if
ic
a
ti
on s
pe
e
d c
om
pa
r
is
on
M
e
t
hods
C
l
a
s
s
i
f
i
c
a
t
i
on T
i
m
e
(
S
e
c
ond)
m
A
l
e
xne
t
0.00948079
E
f
f
e
c
i
e
nt
P
a
r
ki
ngN
e
t
0.00859164
C
onc
i
s
e
C
a
r
N
e
t
2 X
0.00771057
C
onc
i
s
e
C
a
r
N
e
t
0.00661905
A
pa
r
ki
ng a
r
e
a
c
ont
a
in
s
m
a
ny pa
r
ki
ng
s
pa
c
e
s
t
ha
t
c
a
n be
c
la
s
s
i
f
ie
d s
im
ul
ta
ne
ous
ly
. C
a
m
e
r
a
A
i
n t
h
e
C
N
R
P
a
r
k
da
ta
s
e
t
is
c
a
pa
bl
e
of
m
oni
to
r
in
g
51
pa
r
ki
ng
s
pa
c
e
s
a
t
onc
e
.
T
he
f
a
s
te
r
th
e
ti
m
e
f
or
c
la
s
s
if
yi
ng
a
pa
r
ki
ng
a
r
e
a
,
th
e
f
a
s
te
r
th
e
in
f
or
m
a
ti
on
w
il
l
be
r
e
c
e
iv
e
d
by
th
e
dr
iv
e
r
.
E
f
f
e
c
ie
nt
C
a
r
N
e
t
is
th
e
f
a
s
te
s
t
a
r
c
hi
te
c
tu
r
e
f
or
c
la
s
s
if
yi
ng
pa
r
ki
ng s
pa
c
e
s
in
pa
r
ki
ng
a
r
e
a
s
. T
he
c
la
s
s
if
ic
a
ti
on
ti
m
e
s
f
or
c
a
m
e
r
a
A
in
th
e
C
N
R
P
a
r
k
da
ta
s
e
t
a
r
e
pr
e
s
e
nt
e
d i
n T
a
bl
e
10.
T
a
bl
e
10. P
a
r
ki
ng a
r
e
a
c
la
s
s
if
ic
a
ti
on t
im
e
M
e
t
hods
C
l
a
s
i
f
i
c
a
t
i
on T
i
m
e
(
S
e
c
ond)
F
r
a
m
e
s
/
S
e
c
ond
E
f
f
e
c
i
e
nt
P
a
r
ki
ngN
e
t
0.438174
2.282200
C
onc
i
s
e
C
a
r
N
e
t
2 X
0.393239
2.542982
C
onc
i
s
e
C
a
r
N
e
t
0.337572
2.962334
B
a
s
e
d
on T
a
bl
e
10, it
i
s
c
le
a
r
t
ha
t
th
e
c
l
a
s
s
if
ic
a
ti
on t
im
e
f
or
a
s
i
ngl
e
pa
r
ki
ng s
pa
c
e
us
in
g
m
A
le
xne
t
is
0.48
s
e
c
ond
s
,
w
hi
le
C
on
c
is
e
C
a
r
N
e
t
obt
a
in
s
a
f
a
s
t
e
r
ti
m
e
of
0.337
s
e
c
ond
s
.
I
n
te
r
m
s
of
pr
oc
e
s
s
in
g
s
p
e
e
d,
m
A
le
xne
t
c
a
n
c
a
t
e
gor
iz
e
P
K
L
ot
a
t
a
r
a
te
of
2
pi
c
tu
r
e
f
r
a
m
e
s
pe
r
s
e
c
ond,
but
C
on
c
is
e
C
a
r
N
e
t
out
pe
r
f
or
m
s
at
2.9
f
r
a
m
e
s
pe
r
s
e
c
ond.
O
ur
s
tu
dy
not
onl
y
de
m
ons
tr
a
te
s
th
e
im
pr
ov
e
d
a
c
c
ur
a
c
y
of
th
e
ne
w
a
r
c
hi
te
c
tu
r
e
but
a
ls
o
it
s
be
tt
e
r
s
pe
e
d c
om
pa
r
e
d t
o ot
he
r
e
xi
s
ti
ng a
r
c
hi
te
c
tu
r
e
s
.
4.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
f
ound
a
ne
w
C
N
N
a
r
c
hi
te
c
tu
r
e
dubbe
d
C
onc
is
e
C
a
r
N
e
t
f
or
id
e
nt
i
f
yi
ng
pa
r
ki
ng
s
pa
c
e
s
,
w
hi
c
h
a
ddr
e
s
s
e
s
th
e
di
f
f
ic
ul
ty
of
pa
r
ki
ng
s
pa
c
e
a
va
il
a
bi
li
ty
in
ur
ba
n
a
r
e
a
s
.
C
onc
i
s
e
C
a
r
N
e
t
us
e
s
s
m
a
ll
f
il
te
r
s
,
na
m
e
ly
3
3
a
nd
1
1
w
it
h
a
s
m
a
ll
num
be
r
of
f
il
te
r
s
,
th
is
c
a
us
e
s
th
e
num
b
e
r
of
pa
r
a
m
e
te
r
s
to
be
le
s
s
.
C
onc
is
e
C
a
r
N
e
t
ha
s
br
a
nc
he
s
w
it
h
di
f
f
e
r
e
nt
s
tr
uc
tu
r
e
s
,
a
f
te
r
th
e
f
ir
s
t
c
onvolut
io
n
la
ye
r
.
T
hi
s
br
a
nc
h
is
in
te
nde
d
to
ge
ne
r
a
te
m
or
e
f
e
a
tu
r
e
va
r
ia
nc
e
,
th
e
r
e
by
c
ont
r
ib
ut
in
g
to
i
nc
r
e
a
s
e
d a
c
c
ur
a
c
y.
B
a
s
e
d
on
th
e
te
s
t,
th
e
a
c
c
ur
a
c
y
of
C
onc
is
e
C
a
r
N
e
t2
x
out
pe
r
f
or
m
s
th
e
a
c
c
ur
a
c
y
of
m
A
le
xne
t,
E
f
f
ic
ie
nt
P
a
r
ki
ngN
e
t,
a
nd
C
a
r
ne
t
a
r
c
hi
te
c
tu
r
e
s
.
T
he
a
c
c
ur
a
c
y
of
C
onc
is
e
C
a
r
N
e
t2
x
is
99.37%
on
th
e
C
N
R
P
a
r
kE
X
T
da
ta
s
e
t
a
nd
a
n
a
ve
r
a
ge
of
97.27%
f
or
th
e
P
K
L
ot
da
ta
s
e
t,
e
ve
n
r
e
a
c
hi
ng
99.93%
in
th
e
s
a
m
e
s
ubs
e
ts
on
t
he
P
K
L
ot
da
ta
s
e
t.
I
n
a
ddi
ti
on,
C
onc
is
e
C
a
r
N
e
t
a
ls
o ha
s
f
e
w
e
r
pa
r
a
m
e
te
r
s
, f
il
e
s
iz
e
, a
nd F
L
O
P
s
c
om
pa
r
e
d t
o ot
he
r
a
r
c
hi
te
c
tu
r
e
s
. T
he
r
e
f
or
e
, C
onc
is
e
C
a
r
N
e
t
is
f
a
s
te
r
a
nd
c
a
n
be
im
pl
e
m
e
nt
e
d
f
or
pa
r
ki
ng
s
pa
c
e
c
la
s
s
if
ic
a
ti
on
w
it
h
lo
w
-
c
om
pu
ti
ng
de
vi
c
e
s
,
s
uc
h
a
s
th
e
R
a
s
pbe
r
r
y P
i.
T
hi
s
w
il
l
s
a
v
e
c
os
t
s
i
n pur
c
ha
s
in
g c
om
put
in
g t
ool
s
. T
he
w
e
a
kne
s
s
of
t
hi
s
r
e
s
e
a
r
c
h i
s
t
he
m
a
nua
l
de
te
r
m
in
a
ti
on
of
th
e
c
oor
di
na
te
s
f
or
e
a
c
h
p
a
r
ki
ng
s
pa
c
e
.
T
hi
s
te
c
hnol
ogy
is
e
xpe
c
te
d
to
be
a
bl
e
to
de
t
e
c
t
pa
r
ki
ng s
pa
c
e
c
oor
di
na
te
s
i
nde
pe
nde
nt
ly
i
n t
he
f
ut
ur
e
, t
he
r
e
by
e
li
m
in
a
ti
ng t
he
ne
e
d f
or
m
a
nua
l
la
be
li
ng.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
onc
is
e
C
a
r
N
e
t:
c
on
v
ol
ut
io
nal
ne
ur
al
ne
tw
o
r
k
f
or
pa
r
k
in
g s
pac
e
c
la
s
s
if
ic
at
io
n (
M
ar
w
an R
am
li
)
4167
R
E
F
E
R
E
N
C
E
S
[
1]
I
.
S
uka
r
no,
H
.
M
a
t
s
um
ot
o,
a
nd
L
.
S
us
a
nt
i
,
“
T
r
a
ns
por
t
a
t
i
on
e
ne
r
gy
c
on
s
um
pt
i
on
a
nd
e
m
i
s
s
i
ons
-
a
vi
e
w
f
r
om
c
i
t
y
of
I
ndone
s
i
a
,”
F
ut
ur
e
C
i
t
i
e
s
and E
nv
i
r
onm
e
nt
, vol
. 2, no. 1, J
un. 2016, doi
:
10.1186/
s
40984
-
016
-
0019
-
x.
[
2]
S
.
U
t
t
a
r
a
,
N
.
B
huva
nda
s
,
a
nd
V
.
A
gg
a
r
w
a
l
,
“
I
m
pa
c
t
s
of
ur
ba
ni
z
a
t
i
on
on
e
n
vi
r
onm
e
nt
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
R
e
s
e
a
r
c
h
i
n
E
ngi
ne
e
r
i
ng and A
ppl
i
e
d Sc
i
e
nc
e
s
, vol
. 2, no. 2, pp. 1637
–
1645, 2012.
[
3]
T
.
L
i
n,
H
.
R
i
va
no,
a
nd
F
.
L
e
M
oue
l
,
“
A
s
ur
ve
y
of
s
m
a
r
t
pa
r
ki
ng
s
ol
u
t
i
ons
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
I
nt
e
l
l
i
ge
nt
T
r
ans
por
t
at
i
on
Sy
s
t
e
m
s
, vol
. 18, no. 12, pp. 3229
–
3253, D
e
c
. 2017, doi
:
10.1109/
T
I
T
S
.2017.2
685143.
[
4]
Y.
-
Y
.
C
hi
a
ng
a
nd C
.
A
. K
nobl
oc
k,
“
A
ge
ne
r
a
l
a
ppr
oa
c
h
f
or
e
xt
r
a
c
t
i
ng
r
oa
d
ve
c
t
or
da
t
a
f
r
om
r
a
s
t
e
r
m
a
ps
,”
I
nt
e
r
nat
i
onal
J
our
nal
o
n
D
oc
um
e
nt
A
nal
y
s
i
s
and R
e
c
ogni
t
i
on (
I
J
D
A
R
)
, vol
. 16, no. 1, pp. 55
–
81, M
a
r
. 2
013, doi
:
10.1007/
s
10032
-
011
-
0177
-
1.
[
5]
R
.
Y
us
ni
t
a
,
F
.
N
or
ba
ya
,
a
nd
N
. B
a
s
h
a
r
uddi
n,
“
I
nt
e
l
l
i
ge
nt
pa
r
ki
ng
s
pa
c
e
de
t
e
c
t
i
o
n
s
ys
t
e
m
ba
s
e
d
on
i
m
a
ge
pr
oc
e
s
s
i
ng,”
I
nt
e
r
nat
i
ona
l
J
our
nal
of
I
nnov
at
i
on, M
anage
m
e
nt
and T
e
c
hnol
ogy
, vol
. 3, no. 3, pp. 232
–
235, 2012.
[
6]
H
.
A
l
-
K
ha
r
us
i
a
nd
I
.
A
l
-
B
a
ha
dl
y,
“
I
nt
e
l
l
i
ge
nt
pa
r
ki
ng
m
a
na
ge
m
e
nt
s
ys
t
e
m
ba
s
e
d
on
i
m
a
ge
pr
oc
e
s
s
i
ng,”
W
or
l
d
J
our
nal
of
E
ngi
ne
e
r
i
ng and T
e
c
hnol
ogy
, vol
. 2, no. 2, pp. 55
–
67, 2014, doi
:
10.4236/
w
j
e
t
.2014.22006.
[
7]
I
.
M
a
s
m
oudi
a
nd
A
.
W
a
l
i
,
“
V
i
s
i
on
ba
s
e
d a
ppr
oa
c
h
f
or
a
da
pt
i
ve
pa
r
ki
ng
l
ot
s
oc
c
upa
nc
y
e
s
t
i
m
a
t
i
on,”
P
at
t
e
r
n
R
e
c
ogni
t
i
on
and
I
m
age
A
nal
y
s
i
s
, vol
. 29, no. 3, pp. 515
–
522, J
ul
. 2019, doi
:
10.1134/
S
1054661819030
143.
[
8]
G
.
A
m
a
t
o,
F
.
C
a
r
r
a
r
a
,
F
.
F
a
l
c
hi
,
C
.
G
e
nna
r
o,
C
.
M
e
ghi
ni
,
a
nd
C
.
V
a
i
r
o,
“
D
e
e
p
l
e
a
r
ni
ng
f
or
de
c
e
nt
r
a
l
i
z
e
d
pa
r
ki
ng
l
ot
oc
c
upa
nc
y
de
t
e
c
t
i
on,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 72, pp. 327
–
334, A
pr
. 2017,
doi
:
10.1016/
j
.e
s
w
a
.2016.10.055.
[
9]
S
.
N
ur
ul
l
a
ye
v
a
nd
S
.
-
W
.
L
e
e
,
“
G
e
ne
r
a
l
i
z
e
d
pa
r
ki
ng
oc
c
up
a
nc
y
a
na
l
y
s
i
s
ba
s
e
d
on
di
l
a
t
e
d
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
Se
ns
o
r
s
,
vol
. 19, no. 2, J
a
n. 2019, doi
:
10.3390/
s
19020277.
[
10]
A
. K
r
i
z
he
vs
ky,
I
. S
ut
s
ke
ve
r
,
a
nd G
.
E
. H
i
nt
on, “
I
m
a
ge
ne
t
c
l
a
s
s
i
f
i
c
a
t
i
on w
i
t
h d
e
e
p c
onvol
ut
i
ona
l
n
e
ur
a
l
ne
t
w
or
ks
,”
i
n
A
dv
anc
e
s
i
n
ne
ur
al
i
nf
or
m
at
i
on pr
oc
e
s
s
i
ng
s
y
s
t
e
m
s
, 2012, pp. 1097
–
1105.
[
11]
K
.
S
i
m
onya
n
a
nd
A
.
Z
i
s
s
e
r
m
a
n,
“
V
e
r
y
de
e
p
c
onvol
ut
i
ona
l
ne
t
w
or
ks
f
or
l
a
r
ge
-
s
c
a
l
e
i
m
a
ge
r
e
c
ogni
t
i
on,”
3
r
d
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on L
e
ar
ni
ng R
e
pr
e
s
e
nt
at
i
ons
(
I
C
L
R
2015)
, C
om
put
at
i
onal
and B
i
ol
ogi
c
al
L
e
ar
ni
ng Soc
i
e
t
y
,
2015, pp. 1
–
14.
[
12]
C
.
S
z
e
ge
dy
e
t
al
.
,
“
G
oi
ng
de
e
pe
r
w
i
t
h
c
onvol
ut
i
ons
,”
i
n
2015
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
, J
un. 2015, pp. 1
–
9, doi
:
10.1109
/
C
V
P
R
.2015.7298594.
[
13]
K
.
H
e
,
X
.
Z
ha
ng,
S
.
R
e
n,
a
nd
J
.
S
un,
“
D
e
e
p
r
e
s
i
dua
l
l
e
a
r
ni
ng
f
or
i
m
a
ge
r
e
c
ogni
t
i
on
,”
i
n
2016
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and P
at
t
e
r
n R
e
c
ogni
t
i
on (
C
V
P
R
)
,
J
un. 2016, pp. 770
–
778, doi
:
10.1109/
C
V
P
R
.2016.90.
[
14]
M
.
S
a
ndl
e
r
,
A
.
H
ow
a
r
d,
M
.
Z
hu,
A
.
Z
hm
ogi
nov,
a
nd
L
.
-
C
.
C
he
n,
“
M
obi
l
e
N
e
t
V
2:
i
nve
r
t
e
d
r
e
s
i
dua
l
s
a
nd
l
i
ne
a
r
bot
t
l
e
n
e
c
ks
,”
i
n
2018
I
E
E
E
/
C
V
F
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
J
un.
2018,
pp.
4510
–
4520,
doi
:
10.1109/
C
V
P
R
.2018.00474.
[
15]
N
.
M
a
,
X
.
Z
ha
ng,
H
.
-
T
.
Z
he
ng,
a
nd
J
.
S
un,
“
S
hu
f
f
l
e
N
e
t
V
2:
p
r
a
c
t
i
c
a
l
gui
de
l
i
ne
s
f
or
e
f
f
i
c
i
e
nt
C
N
N
a
r
c
hi
t
e
c
t
ur
e
de
s
i
g
n,”
i
n
P
r
oc
e
e
di
ngs
of
t
he
E
ur
ope
an c
onf
e
r
e
nc
e
on c
om
put
e
r
v
i
s
i
on (
E
C
C
V
)
, 2018, pp. 122
–
138.
[
16]
M
. K
l
a
i
be
r
, D
. S
a
ut
e
r
, H
.
B
a
um
ga
r
t
l
, a
nd R
.
B
ue
t
t
ne
r
, “
A
s
ys
t
e
m
a
t
i
c
l
i
t
e
r
a
t
ur
e
r
e
vi
e
w
on t
r
a
ns
f
e
r
l
e
a
r
ni
ng
f
or
3D
-
C
N
N
s
,”
i
n
202
1
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on
N
e
ur
al
N
e
t
w
or
k
s
(
I
J
C
N
N
)
, J
ul
. 2021, pp. 1
–
10, doi
:
10.1109/
I
J
C
N
N
52387.2021.9533302.
[
17]
Z
. Z
hou,
J
.
S
hi
n, L
.
Z
ha
ng, S
.
G
ur
udu, M
. G
ot
w
a
y, a
nd
J
.
L
i
a
ng, “
F
i
ne
-
t
uni
ng c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
f
or
bi
o
m
e
di
c
a
l
i
m
a
ge
a
na
l
ys
i
s
:
a
c
t
i
ve
l
y
a
nd
i
nc
r
e
m
e
nt
a
l
l
y,”
i
n
2017
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
,
J
u
l
.
2017,
pp. 4761
–
4772, doi
:
10.1109/
C
V
P
R
.2017.506.
[
18]
A
.
A
l
qa
ht
a
ni
,
X
.
X
i
e
,
a
nd
M
.
W
.
J
one
s
,
“
L
i
t
e
r
a
t
ur
e
r
e
vi
e
w
of
de
e
p
ne
t
w
or
k
c
om
pr
e
s
s
i
on,”
I
nf
or
m
at
i
c
s
,
vol
.
8,
no.
4,
N
ov.
2021,
doi
:
10.3390/
i
nf
or
m
a
t
i
c
s
8040077.
[
19]
Y
.
L
e
c
un,
L
.
B
ot
t
ou,
Y
.
B
e
ngi
o,
a
nd
P
.
H
a
f
f
ne
r
,
“
G
r
a
di
e
nt
-
ba
s
e
d
l
e
a
r
ni
ng
a
ppl
i
e
d
t
o
doc
um
e
nt
r
e
c
ogni
t
i
on,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
, vol
. 86, no. 11, pp. 2278
–
2324, 1998, doi
:
10.1109/
5.726791.
[
20]
S
.
R
a
hm
a
n,
M
.
R
a
m
l
i
,
F
.
A
r
ni
a
,
R
.
M
uha
r
a
r
,
a
nd
A
.
S
e
m
bi
r
i
ng,
“
P
e
r
f
or
m
a
nc
e
a
n
a
l
ys
i
s
of
m
A
l
e
xne
t
by
t
r
a
i
ni
ng
opt
i
on
a
nd
a
c
t
i
va
t
i
on
f
unc
t
i
on t
uni
ng on pa
r
ki
ng i
m
a
ge
s
,”
I
O
P
C
onf
e
r
e
n
c
e
Se
r
i
e
s
:
M
at
e
r
i
al
s
S
c
i
e
nc
e
and E
ngi
ne
e
r
i
ng
,
vol
. 1087, no.
1, F
e
b.
2021, doi
:
10.1088/
1757
-
899X
/
1087/
1/
012084.
[
21]
S
. R
a
hm
a
n, M
.
R
a
m
l
i
, F
. A
r
ni
a
, R
. M
uh
a
r
a
r
, M
. I
khw
a
n, a
nd S
.
M
unz
i
r
, “
E
nha
nc
e
m
e
nt
of
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k f
or
ur
ba
n
e
nvi
r
onm
e
nt
pa
r
ki
ng s
pa
c
e
c
l
a
s
s
i
f
i
c
a
t
i
on,”
G
l
obal
J
ou
r
nal
of
E
nv
i
r
onm
e
nt
al
Sc
i
e
nc
e
and M
anage
m
e
nt
, vol
. 8, no. 3, pp. 315
–
326
,
2022, doi
:
ht
t
ps
:
/
/
doi
.or
g/
10.22034/
gj
e
s
m
.2022.03.02.
[
22]
P
.
P
.
B
.
D.
G
us
m
a
o,
G
.
F
r
a
nc
i
ni
,
S
.
L
e
ps
øy,
a
nd
E
.
M
a
gl
i
,
“
F
a
s
t
t
r
a
i
ni
ng
o
f
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
v
i
a
ke
r
ne
l
r
e
s
c
a
l
i
n
g,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp. 1
-
13,
2016.
[
23]
M
.
J
.
J
.
P
.
V.
G
r
i
ns
ve
n,
B
.
V.
G
i
nne
ke
n,
C
.
B
.
H
oyng,
T
. T
he
e
l
e
n,
a
nd
C
.
I
. S
a
nc
he
z
,
“
F
a
s
t
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k
t
r
a
i
ni
ng
us
i
ng
s
e
l
e
c
t
i
ve
d
a
t
a
s
a
m
pl
i
ng:
a
ppl
i
c
a
t
i
on
t
o
he
m
or
r
ha
ge
de
t
e
c
t
i
on
i
n
c
ol
or
f
undus
i
m
a
ge
s
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
M
e
di
c
al
I
m
agi
ng
, vol
. 35, no. 5, pp. 1273
–
1284, M
a
y 2016, doi
:
10.1109/
T
M
I
.2016.2526689.
[
24]
J
.
Y
a
m
a
na
ka
,
S
.
K
uw
a
s
hi
m
a
,
a
nd
T
.
K
ur
i
t
a
,
“
F
a
s
t
a
nd
a
c
c
ur
a
t
e
i
m
a
ge
s
upe
r
r
e
s
ol
ut
i
on
by
de
e
p
C
N
N
w
i
t
h
s
ki
p
c
onne
c
t
i
on
a
nd
ne
t
w
or
k
i
n
ne
t
w
or
k
,”
i
n
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
N
e
ur
al
I
nf
or
m
at
i
on
P
r
oc
e
s
s
i
ng
,
S
pr
i
nge
r
,
2017,
pp.
217
–
225
,
doi
:
10.1007/
978
-
3
-
319
-
70096
-
0_23
.
[
25]
S
.
D
.
K
ha
n
a
nd
H
.
U
l
l
a
h,
“
A
s
ur
ve
y
of
a
dv
a
nc
e
s
i
n
vi
s
i
on
-
ba
s
e
d
ve
hi
c
l
e
r
e
-
i
de
nt
i
f
i
c
a
t
i
on,”
C
om
put
e
r
V
i
s
i
on
and
I
m
ag
e
U
nde
r
s
t
andi
ng
, vol
. 182, pp. 50
–
63, M
a
y 2019, doi
:
10.1016/
j
.c
vi
u.2019.03.001.
[
26]
N
. T
r
ue
, “
V
a
c
a
nt
pa
r
ki
ng s
pa
c
e
de
t
e
c
t
i
on i
n s
t
a
t
i
c
i
m
a
ge
s
,
”
U
ni
v
e
r
s
i
t
y
of
C
al
i
f
o
r
ni
a, San D
i
e
go
, vol
. 17, pp. 659
–
662, 2007.
[
27]
Q
.
W
u,
C
.
H
ua
ng,
S
.
W
a
ng,
W
.
C
hi
u,
a
nd
T
.
C
he
n,
“
R
obus
t
pa
r
ki
ng
s
pa
c
e
de
t
e
c
t
i
on
c
ons
i
de
r
i
ng
i
nt
e
r
-
s
pa
c
e
c
or
r
e
l
a
t
i
on,”
i
n
M
ul
t
i
m
e
di
a and E
x
po, 2007 I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
, J
ul
. 2007, pp. 659
–
662, doi
:
10.1109/
I
C
M
E
.2007.4284736.
[
28]
J
.
K
.
S
uhr
,
H
.
G
.
J
ung,
K
.
B
a
e
,
a
nd
J
.
K
i
m
,
“
A
ut
om
a
t
i
c
f
r
e
e
p
a
r
ki
ng
s
pa
c
e
de
t
e
c
t
i
on
by
us
i
ng
m
ot
i
on
s
t
e
r
e
o
-
ba
s
e
d
3
D
r
e
c
ons
t
r
uc
t
i
on,”
M
ac
hi
ne
V
i
s
i
on and A
ppl
i
c
at
i
ons
, vol
. 21, no. 2, pp. 163
–
176, F
e
b. 2010, doi
:
10.1007/
s
00138
-
008
-
0156
-
9.
[
29]
N
. B
i
bi
, M
. N
. M
a
j
i
d, H
. D
a
w
ood, a
nd
P
. G
uo, “
A
ut
om
a
t
i
c
pa
r
ki
ng s
pa
c
e
de
t
e
c
t
i
on s
y
s
t
e
m
,”
i
n
2017 2nd
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ul
t
i
m
e
di
a and I
m
age
P
r
oc
e
s
s
i
ng (
I
C
M
I
P
)
, M
a
r
. 2017, pp. 11
–
15, doi
:
10.1
109/
I
C
M
I
P
.2017.4.
[
30]
P
.
R
.
L
.
D.
A
l
m
e
i
da
,
L
.
S
.
O
l
i
ve
i
r
a
,
A
.
S
.
B
r
i
t
t
o,
E
.
J
.
S
i
l
va
,
a
nd
A
.
L
.
K
oe
r
i
c
h,
“
P
K
L
ot
–
A
r
obus
t
da
t
a
s
e
t
f
or
pa
r
ki
ng
l
ot
c
l
a
s
s
i
f
i
c
a
t
i
on,”
E
x
pe
r
t
S
y
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 42, no. 11, pp. 4937
–
4949, J
ul
. 2015, doi
:
10.1016/
j
.e
s
w
a
.2015.02.009.
[
31]
L.
-
C
.
C
he
n,
R
.
-
K
.
S
he
u,
W
.
-
Y
.
P
e
ng,
J
.
-
H
.
W
u,
a
nd
C
.
-
H
.
T
s
e
ng,
“
V
i
de
o
-
ba
s
e
d
pa
r
ki
ng
oc
c
upa
nc
y
d
e
t
e
c
t
i
on
f
or
s
m
a
r
t
c
ont
r
ol
s
ys
t
e
m
,”
A
ppl
i
e
d Sc
i
e
nc
e
s
, vol
. 10, no. 3,
F
e
b. 2020, doi
:
10.3390/
a
pp1003107
9.
[
32]
T
. C
onni
e
, M
.
K
.
O
. G
oh,
V
.
C
. K
oo,
K
. T
.
M
ur
a
t
a
,
a
nd S
. P
.
A
m
nua
i
s
uk,
“
I
m
pr
ove
d
pa
r
ki
ng s
pa
c
e
r
e
c
ogni
t
i
on vi
a
gr
a
s
s
m
a
nni
a
n
de
e
p
s
t
a
c
ki
ng
ne
t
w
or
k
w
i
t
h
i
l
l
um
i
na
t
i
on
c
or
r
e
c
t
i
on
,”
i
n
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
at
i
onal
I
nt
e
l
l
i
ge
nc
e
i
n
I
nf
or
m
at
i
on
Sy
s
t
e
m
, S
pr
i
nge
r
, 2021, pp. 150
–
159
, doi
:
10.1007/
978
-
3
-
030
-
68133
-
3_15
.
[
33]
S
. R
a
hm
a
n
e
t
al
.
, “
M
i
ni
s
huf
f
l
e
ne
t
f
or
e
f
f
i
c
i
e
nt
pa
r
ki
ng s
pa
c
e
c
l
a
s
s
i
f
i
c
a
t
i
on,”
i
n
A
I
P
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ngs
, 2023, vol
. 2480, no
.
1, doi
:
10.1063/
5.0103430.
[
34]
S
.
R
a
f
i
que
,
S
.
G
ul
,
K
.
J
a
n,
a
nd
G
.
M
.
K
ha
n,
“
O
pt
i
m
i
z
e
d
r
e
a
l
-
t
i
m
e
pa
r
ki
ng
m
a
na
ge
m
e
nt
f
r
a
m
e
w
or
k
us
i
ng
de
e
p
l
e
a
r
ni
ng,”
E
x
pe
r
t
Evaluation Warning : The document was created with Spire.PDF for Python.