I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
6
,
Decem
b
er
20
25
,
p
p
.
5
9
3
4
~
5
9
4
1
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
9
3
4
-
5
9
4
1
5934
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Co
mputa
tiona
lly
eff
icie
nt
pix
el
wise
deep
l
ea
rning
ar
chitec
ture
for a
ccurate
dept
h recons
tr
uction
f
o
r sing
le
-
pho
ton
LiDAR
Yu Zh
a
ng
,
Yim
ing
Z
he
ng
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
V
i
t
e
r
b
i
S
c
h
o
o
l
o
f
E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
y
o
f
S
o
u
t
h
e
r
n
C
a
l
i
f
o
r
n
i
a
,
Lo
s An
g
e
l
e
s
,
U
n
i
t
e
d
S
t
a
t
e
s
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 2
7
,
2
0
2
4
R
ev
is
ed
Au
g
6
,
2
0
2
5
Acc
ep
ted
Sep
1
6
,
2
0
2
5
Th
is
wo
rk
in
tr
o
d
u
c
e
s
a
c
o
m
p
a
c
t
d
e
e
p
lea
rn
in
g
a
rc
h
it
e
c
tu
re
fo
r
d
e
p
th
ima
g
e
re
c
o
n
stru
c
ti
o
n
fro
m
ti
m
e
-
re
so
lv
e
d
si
n
g
le
-
p
h
o
to
n
h
isto
g
ra
m
s.
Un
l
ik
e
m
o
s
t
d
e
e
p
lea
rn
i
n
g
a
p
p
r
o
a
c
h
e
s
th
a
t
m
a
in
ly
re
ly
o
n
3
D
c
o
n
v
o
lu
t
io
n
s,
o
u
r
n
e
two
r
k
is
imp
lem
e
n
ted
p
u
re
ly
wi
th
1
D
c
o
n
v
o
lu
ti
o
n
s
with
o
u
t
a
ss
istan
c
e
fro
m
o
t
h
e
r
se
n
so
rs
o
r
p
re
-
p
r
o
c
e
ss
in
g
.
Bo
t
h
sy
n
t
h
e
ti
c
a
n
d
re
a
l
d
a
tas
e
ts
we
re
u
se
d
t
o
e
v
a
lu
a
te
th
e
a
c
c
u
ra
c
y
o
f
o
u
r
m
o
d
e
l
fo
r
c
h
a
ll
e
n
g
in
g
sig
n
a
l
-
to
-
b
a
c
k
g
ro
u
n
d
ra
ti
o
s
(S
BRs
),
ra
n
g
i
n
g
fr
o
m
5
:1
to
1
:
1
.
Co
n
v
e
n
ti
o
n
a
l
m
a
x
imu
m
li
k
e
li
h
o
o
d
(M
L)
a
n
d
a
n
o
th
e
r
p
h
o
to
n
-
e
fficie
n
t
o
p
ti
m
iza
ti
o
n
-
b
a
se
d
a
lg
o
ri
th
m
we
re
a
d
o
p
te
d
fo
r
p
e
rf
o
rm
a
n
c
e
c
o
m
p
a
riso
n
s.
Re
su
lt
s
fro
m
sy
n
th
e
ti
c
d
a
ta sh
o
w
th
a
t
o
u
r
m
o
d
e
l
a
c
h
ie
v
e
s
lo
we
r
m
e
a
n
a
b
so
lu
te
e
rr
o
r
(M
AE
).
Ad
d
it
i
o
n
a
l
ly
,
re
su
lt
s
fro
m
re
a
l
d
a
ta
in
d
ica
te
th
a
t
o
u
r
m
o
d
e
l
e
x
h
i
b
it
s
b
e
tt
e
r
re
c
o
n
str
u
c
ti
o
n
fo
r
h
ig
h
-
a
m
b
ien
t
e
ffe
c
ts
a
n
d
p
r
o
v
i
d
e
s
b
e
tt
e
r
sp
a
ti
a
l
in
fo
rm
a
ti
o
n
.
Un
li
k
e
e
x
isti
n
g
3
D
d
e
e
p
lea
rn
in
g
m
o
d
e
ls,
we
p
r
o
c
e
ss
p
ix
e
l
-
wise
h
isto
g
ra
m
s
c
o
n
ti
n
u
o
u
sl
y
,
ra
th
e
r
th
a
n
s
p
li
tt
in
g
th
e
p
o
in
t
c
lo
u
d
a
n
d
stit
c
h
i
n
g
th
e
m
a
fterw
a
rd
,
wh
ic
h
sa
v
e
s
m
e
m
o
ry
a
n
d
c
o
m
p
u
tatio
n
a
l
re
so
u
rc
e
s,
th
e
re
b
y
lay
in
g
a
fo
u
n
d
a
ti
o
n
fo
r
re
a
l
-
wo
rld
e
m
b
e
d
d
e
d
a
p
p
li
c
a
ti
o
n
s
.
K
ey
w
o
r
d
s
:
C
o
m
p
u
tatio
n
al
im
ag
in
g
C
o
m
p
u
tatio
n
al
in
tellig
en
ce
Dee
p
lear
n
in
g
Dep
th
r
ec
o
n
s
tr
u
ctio
n
Sin
g
le
-
p
h
o
to
n
L
iDAR
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Yu
Z
h
an
g
Dep
ar
tm
en
t o
f
E
lectr
ical
E
n
g
i
n
ee
r
in
g
,
Viter
b
i Sch
o
o
l
o
f
E
n
g
in
ee
r
in
g
,
Un
iv
er
s
ity
o
f
So
u
th
er
n
C
alif
o
r
n
ia
L
o
s
An
g
eles,
C
A
9
0
0
8
9
E
m
ail: y
z3
2
4
@
u
s
c.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
Sin
g
le
-
p
h
o
to
n
av
alan
c
h
e
d
io
d
es
(
SP
AD
s
)
h
av
e
b
ee
n
em
er
g
in
g
f
o
r
v
a
r
io
u
s
ap
p
licatio
n
s
t
h
at
r
ely
o
n
s
in
g
le
-
p
h
o
to
n
s
en
s
itiv
ity
,
s
u
ch
as
s
in
g
le
-
p
h
o
to
n
lig
h
t
d
etec
tio
n
an
d
r
an
g
in
g
(
L
iDAR
)
u
s
i
n
g
an
o
p
tim
izatio
n
-
b
ased
r
ec
o
n
s
tr
u
ctio
n
m
et
h
o
d
[
1
]
,
d
ee
p
lear
n
in
g
m
eth
o
d
s
[
2
]
,
[
3
]
,
an
d
b
io
m
ed
ical
s
ig
n
al
p
r
o
ce
s
s
in
g
f
o
r
f
lu
o
r
escen
ce
life
tim
e
im
ag
i
n
g
(
FLI
M)
[
4
]
,
an
d
n
o
n
-
lin
e
-
of
-
s
ig
h
t
im
ag
in
g
[
5
]
,
[
6
]
a
n
d
cr
y
p
to
g
r
ap
h
y
[
7
]
,
[
8
]
.
R
esear
ch
er
s
p
r
o
v
ed
th
at
u
s
in
g
d
ata
-
d
r
iv
en
d
ee
p
lear
n
in
g
(
DL
)
m
o
d
els
ca
n
ac
cu
r
ately
r
e
co
n
s
tr
u
ct
d
e
p
th
a
n
d
r
ef
lectiv
ity
im
ag
es f
r
o
m
th
e
3
D
p
o
in
t
cl
o
u
d
cu
b
es
t
h
at
in
clu
d
e
p
h
o
to
n
s
’
tim
e
-
of
-
f
lig
h
t
i
n
f
o
r
m
atio
n
a
n
d
s
p
atial
in
f
o
r
m
atio
n
.
Fu
r
th
er
,
th
ese
DL
m
o
d
els ar
e
r
o
b
u
s
t f
o
r
ex
tr
e
m
ely
lo
w
s
ig
n
al
-
to
-
b
ac
k
g
r
o
u
n
d
r
atio
s
(
SB
R
s
)
,
ev
en
less
th
an
o
n
e.
R
ec
o
n
s
tr
u
ctin
g
d
ep
th
in
f
o
r
m
atio
n
is
cr
u
cial
in
au
to
n
o
m
o
u
s
v
e
h
icles
th
at
n
ee
d
f
ast
an
d
ac
c
u
r
ate
r
esp
o
n
s
e,
e
v
en
in
l
o
w
-
v
is
ib
ilit
y
en
v
ir
o
n
m
en
ts
.
Alth
o
u
g
h
d
at
a
-
d
r
iv
en
m
eth
o
d
s
r
e
co
n
s
tr
u
ct
d
ep
th
i
m
ag
es
b
ased
o
n
SP
AD
ar
e
em
er
g
in
g
,
th
e
r
e
ar
e
s
till
ch
allen
g
es
f
o
r
th
e
DL
m
o
d
els.
First,
m
o
s
t
DL
m
o
d
els
ar
e
co
m
p
o
s
ed
o
f
3
D,
co
n
s
u
m
in
g
en
o
r
m
o
u
s
co
m
p
u
tin
g
m
em
o
r
y
f
o
r
th
e
c
o
m
p
u
tin
g
p
latf
o
r
m
.
E
v
en
f
o
r
h
ig
h
-
p
er
f
o
r
m
an
ce
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
its
(
GPUs
)
,
th
e
b
ig
p
o
in
t
clo
u
d
wi
th
b
ig
s
p
atial
r
eso
l
u
tio
n
s
h
o
u
ld
b
e
d
iv
id
ed
in
to
s
ev
er
al
b
atch
es
f
o
r
p
r
o
ce
s
s
in
g
an
d
s
titch
ed
e
v
en
tu
ally
to
o
b
tain
a
h
ig
h
-
r
eso
lu
tio
n
d
ep
t
h
im
ag
e.
Seco
n
d
,
t
h
e
p
r
ep
ar
atio
n
o
f
tr
ain
in
g
d
atasets
is
co
m
p
lex
,
lev
er
ag
in
g
im
a
g
e
p
r
o
ce
s
s
in
g
t
o
o
ls
[
9
]
an
d
lar
g
e
o
p
en
-
s
o
u
r
ce
[
1
0
]
,
[
1
1
]
d
ep
th
tr
ain
in
g
d
atasets
,
also
co
n
s
u
m
in
g
a
lo
n
g
tim
e
(
s
ev
er
al
h
o
u
r
s
)
.
T
h
is
wo
r
k
aim
s
to
d
esig
n
a
co
m
p
u
tatio
n
ally
ef
f
icien
t,
p
ix
el
-
wis
e
DL
m
o
d
el
with
a
co
m
p
ac
t
ar
ch
itectu
r
e
an
d
tr
ain
in
g
p
ip
elin
e
to
ad
d
r
ess
th
ese
two
b
o
ttlen
ec
k
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
o
mp
u
ta
tio
n
a
lly
efficien
t p
ix
elw
is
e
d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
…
(
Yu
Zh
a
n
g
)
5935
T
h
e
co
n
tr
ib
u
tio
n
s
o
f
t
h
is
s
tu
d
y
ar
e
s
u
m
m
ar
ized
as
f
o
llo
ws:
i)
W
e
lev
er
ag
e
d
e
x
is
tin
g
o
p
e
n
-
s
o
u
r
ce
d
d
ep
th
im
ag
e
d
atasets
to
g
en
er
ate
p
ix
el
-
wis
e
h
is
to
g
r
am
s
u
s
in
g
an
an
aly
tical
m
ath
e
m
atica
l
m
o
d
el
an
d
tr
ain
ed
a
d
ee
p
n
eu
r
al
n
etwo
r
k
;
ii)
W
e
d
esig
n
ed
a
co
m
p
ac
t
1
D
U
-
N
E
T
d
ee
p
n
e
u
r
al
n
etwo
r
k
f
o
r
ac
cu
r
ate,
en
d
-
to
-
en
d
p
ix
el
-
wis
e
d
ep
th
in
co
n
d
itio
n
s
o
f
lo
w
SB
R
s
;
an
d
iii)
W
e
q
u
an
titativ
ely
c
o
m
p
ar
e
d
o
u
r
m
o
d
el
with
e
x
is
tin
g
p
h
o
to
n
-
ef
f
icien
t
p
ix
el
-
wis
e
alg
o
r
ith
m
s
an
d
ac
h
iev
e
d
b
etter
a
cc
u
r
ac
y
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
is
s
tu
d
y
i
s
as
f
o
llo
ws.
Sectio
n
2
r
e
v
ie
ws
an
d
s
u
m
m
a
r
izes
th
e
e
x
is
tin
g
wo
r
k
.
Sectio
n
3
illu
s
tr
ates
th
e
m
at
h
em
atica
l
m
o
d
el
o
f
th
e
s
in
g
le
-
p
h
o
to
n
L
iDAR
.
Sectio
n
4
p
r
esen
ts
th
e
d
ee
p
lear
n
in
g
a
r
ch
itectu
r
e,
t
r
ain
in
g
s
y
n
th
etic
d
ata
g
en
e
r
atio
n
,
an
d
tr
ai
n
in
g
d
etails.
Sectio
n
5
q
u
an
titativ
ely
ev
alu
ates
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
DL
m
o
d
el
a
n
d
c
o
m
p
ar
es
i
t
with
p
h
o
to
n
-
e
f
f
icien
t
p
ix
el
-
wis
e
alg
o
r
ith
m
s
an
d
o
p
tim
izatio
n
-
b
ased
m
eth
o
d
s
.
Sectio
n
6
co
n
cl
u
d
es th
e
s
tu
d
y
.
2.
P
RIOR
WO
RK
2
.
1
.
O
pti
m
iza
t
io
n
-
a
nd
s
t
a
t
i
s
t
ic
-
ba
s
ed
a
lg
o
rit
hm
s
Sh
in
et
a
l.
[
1
]
f
ir
s
t
r
ep
o
r
ted
a
p
h
o
to
n
-
ef
f
icien
t
o
p
tim
izat
io
n
-
b
ased
m
eth
o
d
to
r
ec
o
n
s
t
r
u
ct
d
ep
th
im
ag
es
f
r
o
m
h
is
to
g
r
a
m
s
with
ex
tr
em
ely
lo
w
SB
R
r
atio
s
.
A
s
ig
n
al
an
d
n
o
is
e
u
n
m
ix
in
g
o
p
t
im
izatio
n
alg
o
r
ith
m
was
p
r
o
p
o
s
ed
[
1
2
]
to
ac
cu
r
ately
s
p
lit
s
ig
n
als
f
r
o
m
n
o
is
y
h
is
to
g
r
am
s
with
s
tr
o
n
g
am
b
ien
t
lig
h
t
an
d
ac
cu
r
atel
y
r
ec
o
n
s
tr
u
ct
d
e
p
th
an
d
r
ef
lecti
v
ity
im
ag
es.
T
h
e
r
e
v
er
s
ib
le
j
u
m
p
Ma
r
k
o
v
c
h
ain
Mo
n
te
C
a
r
lo
(
R
J
-
MCMC
)
.
A
s
tatis
t
ical
alg
o
r
ith
m
[
1
3
]
was
u
s
ed
to
p
er
f
o
r
m
B
ay
esian
in
f
e
r
en
ce
f
o
r
d
ep
th
a
n
d
in
ten
s
ity
r
ec
o
n
s
tr
u
ctio
n
o
f
3
D
s
ce
n
es.
An
im
p
r
o
v
e
d
R
J
-
MC
MC e
n
h
an
ce
d
b
y
a
p
o
in
t c
lo
u
d
d
en
o
is
in
g
[
1
4
]
ap
p
r
o
ac
h
was p
r
o
p
o
s
ed
to
ac
h
ie
v
e
r
ea
l
-
tim
e
3
D
r
ec
o
n
s
tr
u
ctio
n
o
f
m
o
v
in
g
o
b
jects.
Ko
o
et
a
l.
[
1
5
]
co
m
b
i
n
ed
a
s
tatis
tical
B
ay
e
s
ian
alg
o
r
ith
m
with
a
d
ee
p
lea
r
n
in
g
ar
ch
itectu
r
e,
tak
in
g
ad
v
a
n
tag
e
o
f
b
o
th
a
cc
u
r
ate
in
f
e
r
en
ce
a
n
d
m
o
d
el
-
f
r
ee
p
r
o
p
e
r
ties
o
f
s
tatis
t
ics
an
d
d
ee
p
lear
n
in
g
.
A
co
m
p
u
tatio
n
ally
ef
f
icien
t
B
ay
esian
alg
o
r
ith
m
was
al
s
o
p
r
o
p
o
s
ed
[
1
6
]
f
o
r
a
lo
w
-
p
h
o
to
n
-
co
u
n
t m
u
ltis
p
ec
tr
al
L
iDAR
ap
p
licatio
n
.
2
.
2
.
Dee
p lea
rning
a
lg
o
rit
hm
s
Dee
p
lear
n
in
g
is
b
ec
o
m
in
g
p
r
ev
alen
t
in
f
ea
tu
r
e
e
x
tr
ac
tio
n
in
co
m
p
u
ter
v
is
io
n
[
1
7
]
,
[
1
8
]
.
D
ee
p
n
eu
r
al
n
etwo
r
k
s
h
av
e
b
ee
n
ex
ten
s
i
v
ely
lev
er
a
g
ed
in
d
ep
t
h
r
e
co
n
s
tr
u
ctio
n
f
o
r
SP
AD
ar
r
a
y
s
eq
u
ip
p
e
d
with
tim
e
-
co
r
r
elate
d
s
in
g
le
-
p
h
o
to
n
co
u
n
tin
g
(
T
C
SP
C
)
.
A
s
en
s
o
r
f
u
s
io
n
[
1
9
]
3
D
d
ee
p
n
e
u
r
al
ar
ch
itectu
r
e
was
f
ir
s
t
in
tr
o
d
u
ce
d
to
m
er
g
e
h
ig
h
-
r
eso
lu
tio
n
in
ten
s
ity
a
n
d
lo
w
-
r
eso
l
u
tio
n
d
e
p
th
im
a
g
es
to
en
h
an
c
e
th
e
s
p
atial
f
ea
tu
r
e
ex
tr
ac
tio
n
d
u
r
in
g
th
e
tr
ai
n
in
g
.
T
h
e
ca
p
tu
r
ed
r
aw
d
ata
in
th
is
wo
r
k
was
wid
ely
a
d
o
p
ted
f
o
r
s
u
b
s
eq
u
en
t
wo
r
k
in
th
is
f
ield
[
2
0
]
,
[
2
1
]
.
An
o
th
e
r
f
u
s
io
n
ar
ch
itectu
r
e
was
r
e
p
o
r
ted
to
m
er
g
e
m
o
n
o
c
u
lar
d
ep
th
im
ag
es
with
3
D
p
o
in
t
clo
u
d
c
o
n
v
o
lu
tio
n
m
o
d
u
les
to
en
h
an
ce
d
ep
th
im
a
g
e
r
ec
o
n
s
tr
u
ctio
n
.
T
w
o
d
if
f
er
e
n
t
ar
ch
itectu
r
es
wer
e
in
v
esti
g
ated
f
o
r
n
o
n
-
f
u
s
io
n
a
r
ch
itectu
r
es
th
at
o
n
l
y
lev
er
a
g
e
th
e
p
o
in
t
clo
u
d
f
r
o
m
th
e
S
PAD
ar
r
ay
with
o
u
t
o
th
er
f
ea
tu
r
es
f
r
o
m
o
th
er
s
e
n
s
o
r
s
,
wh
er
e
r
esu
lts
in
d
icate
d
th
at
th
e
n
o
n
-
f
u
s
io
n
ar
ch
ite
ctu
r
e
co
u
ld
ac
h
iev
e
co
m
p
ar
ab
le
ac
cu
r
ac
y
to
f
u
s
io
n
-
b
ased
ar
ch
itectu
r
es.
A
3
D
co
n
v
o
lu
ti
o
n
al
ar
ch
itectu
r
e
with
p
ix
el
-
wis
e
r
esid
u
al
s
h
r
in
k
ag
e
[
3
]
was
r
ep
o
r
ted
to
r
ed
ef
in
e
th
e
o
p
tim
izatio
n
tar
g
et
as
a
cla
s
s
if
icatio
n
f
o
r
ea
ch
h
is
to
g
r
am
,
ac
h
iev
in
g
h
ig
h
r
ec
o
n
s
tr
u
ctio
n
ac
cu
r
ac
y
.
Stu
d
y
[
2
2
]
p
r
ese
n
ted
an
ed
g
e
-
en
h
a
n
ce
d
ar
c
h
itectu
r
e,
em
b
ed
d
in
g
atten
tio
n
m
o
d
u
les
in
th
eir
3
D
co
n
v
o
l
u
tio
n
al
ar
c
h
itectu
r
e,
to
im
p
r
o
v
e
th
e
e
d
g
e
r
ec
o
n
s
tr
u
ctio
n
.
Sp
a
r
s
ity
in
th
e
p
o
in
t
clo
u
d
was
in
v
esti
g
ated
[
2
3
]
t
o
ac
ce
ler
ate
th
e
i
n
f
er
en
c
e
o
f
a
3
D
ar
ch
itectu
r
e
,
ac
h
iev
in
g
r
ea
l
-
tim
e
h
ig
h
-
r
eso
lu
tio
n
d
e
p
th
r
ec
o
n
s
tr
u
ctio
n
.
SP
AD
is
also
u
s
ed
f
o
r
s
en
s
in
g
th
r
o
u
g
h
f
o
g
[
2
4
]
.
E
x
is
tin
g
s
tatis
tical
an
d
o
p
ti
m
izatio
n
m
eth
o
d
s
ar
e
h
ig
h
-
laten
cy
an
d
u
n
s
u
itab
le
f
o
r
em
b
ed
d
e
d
h
ar
d
war
e
in
s
in
g
le
-
p
h
o
to
n
L
i
DAR
s
y
s
tem
s
.
Mo
r
eo
v
er
,
d
esp
ite
th
e
f
ast
f
o
r
wa
r
d
p
r
o
p
ag
at
io
n
o
f
DL
m
o
d
els,
3
D
ten
s
o
r
p
r
o
ce
s
s
in
g
o
f
p
o
in
t
clo
u
d
s
r
em
ain
s
co
m
p
u
tatio
n
a
lly
in
ten
s
iv
e
o
n
h
ar
d
war
e.
T
h
i
s
wo
r
k
b
r
id
g
es
th
e
g
ap
b
etwe
en
DL
an
d
co
m
p
u
ta
tio
n
ally
ef
f
icien
t
m
eth
o
d
s
b
y
lev
er
ag
in
g
1
D
h
is
to
g
r
am
p
r
o
c
ess
in
g
,
r
esu
ltin
g
in
a
co
m
p
ac
t D
L
ar
c
h
itectu
r
e
an
d
s
im
p
lify
in
g
t
h
e
s
y
n
th
etic
d
at
a
g
en
er
atio
n
p
r
o
ce
s
s
.
3.
P
RO
B
L
E
M
D
E
F
I
N
I
T
I
O
N
T
h
e
ac
tiv
e
s
in
g
le
-
p
h
o
t
o
n
im
a
g
in
g
s
y
s
tem
s
h
av
e
b
ee
n
r
e
p
o
r
t
ed
in
e
x
is
tin
g
s
tu
d
ies
[
1
]
,
[
3
]
,
[
1
2
]
,
[
2
1
]
.
SP
AD
ar
r
ay
s
with
T
C
SP
C
-
b
ased
L
iDAR
s
y
s
tem
s
ca
n
b
e
well
ap
p
r
o
x
im
ated
a
n
d
m
o
d
elled
u
s
in
g
k
n
o
wn
o
p
tical
an
d
s
en
s
o
r
p
ar
am
eter
s
.
W
e
aim
to
r
ec
o
n
s
tr
u
ct
d
ep
th
in
f
o
r
m
atio
n
f
r
o
m
th
e
h
is
to
g
r
am
o
f
ea
ch
p
ix
el
,
wh
ich
is
s
u
b
ject
to
an
in
h
o
m
o
g
en
eo
u
s
Po
is
s
o
n
p
r
o
ce
s
s
[
2
2
]
.
T
h
er
ef
o
r
e,
th
e
h
is
to
g
r
am
ca
n
b
e
esti
m
ated
as (
1
)
.
(
)
=
·
(
(
)
+
)
+
,
(
1
)
wh
er
e
η
∈
(
0
,
1
)
,
in
d
icatin
g
th
e
q
u
a
n
tu
m
ef
f
icien
cy
o
f
th
e
s
en
s
o
r
.
an
d
ar
e
th
e
b
ac
k
g
r
o
u
n
d
n
o
is
e
an
d
d
ar
k
-
c
o
u
n
t
n
o
is
e
.
(
)
is
th
e
s
ig
n
al
f
lu
x
r
e
f
lecte
d
f
r
o
m
th
e
tar
g
et,
wh
ich
ca
n
b
e
m
o
d
elled
as
(
2
)
,
(
)
=
·
(
−
2
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
9
3
4
-
5
9
4
1
5936
wh
er
e
is
th
e
atten
u
atio
n
f
ac
t
o
r
,
is
th
e
d
is
tan
ce
f
r
o
m
th
e
s
en
s
o
r
to
th
e
tar
g
et,
a
n
d
is
th
e
s
p
ee
d
o
f
lig
h
t.
T
h
er
ef
o
r
e,
th
e
h
is
to
g
r
am
r
e
p
r
e
s
en
ts
th
at
th
e
r
ef
lecte
d
p
h
o
to
n
s
o
f
illu
m
in
atio
n
s
ca
n
b
e
m
o
d
elled
as
(
3
)
,
ℎ
(
)
~
(
)
(
)
(
3
)
B
y
f
o
llo
win
g
th
e
eq
u
atio
n
f
o
r
an
aly
tically
g
en
er
atin
g
s
y
n
th
etic
tr
ain
in
g
d
atasets
,
we
em
p
lo
y
th
e
d
atasets
to
tr
ain
a
d
ee
p
n
e
u
r
al
n
etwo
r
k
,
wh
ich
is
d
is
cu
s
s
ed
in
th
e
n
e
x
t
s
ec
tio
n
.
T
h
is
p
ix
el
-
wis
e
p
r
o
c
ess
in
g
DL
allev
iates
co
m
p
u
tatio
n
al
co
m
p
lex
ity
co
m
p
ar
ed
with
3
D
-
b
ased
DL
ar
ch
itectu
r
e,
s
im
p
lify
in
g
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
f
r
o
m
co
m
p
lex
3
D
laten
t sp
ac
e
to
1
D
laten
t sp
ac
e
wh
ile
m
ain
tain
in
g
ac
cu
r
ac
y
.
4.
DE
E
P
L
E
A
RNING
AR
CH
I
T
E
C
T
UR
E
I
n
s
p
ir
ed
b
y
p
r
e
v
io
u
s
U
-
NE
T
-
lik
e
3
D
ar
ch
itectu
r
es
[
2
]
,
[
2
0
]
,
[
2
1
]
,
we
p
r
o
p
o
s
ed
a
s
im
ilar
to
p
o
lo
g
y
,
b
u
t o
n
ly
1
D
co
n
v
o
lu
tio
n
was
u
s
ed
.
B
atch
es o
f
h
is
to
g
r
am
s
wer
e
f
ed
in
to
th
e
U
-
NE
T
f
o
r
tr
a
in
in
g
.
Key
m
o
d
u
les
in
th
e
n
etwo
r
k
(
d
o
wn
-
s
am
p
lin
g
,
co
n
ca
ten
atio
n
,
co
n
v
o
lu
ti
o
n
,
u
p
-
s
am
p
lin
g
)
ar
e
co
lo
r
e
d
in
Fig
u
r
e
1
an
d
in
d
icate
d
in
th
e
b
lack
d
ash
ed
b
o
x
.
B
atch
n
o
r
m
aliza
tio
n
m
o
d
u
les
wer
e
u
s
ed
to
im
p
r
o
v
e
th
e
tr
ain
in
g
s
tab
ilit
y
an
d
co
n
v
er
g
en
ce
s
p
ee
d
.
T
h
e
g
r
o
u
n
d
tr
u
th
d
ep
th
im
ag
es
in
th
e
tr
ain
in
g
d
atasets
ar
e
f
r
o
m
t
h
e
NYUV
2
d
atasets
[
1
3
]
.
A
tex
tu
r
e
f
ilter
in
g
[
1
2
]
a
lg
o
r
ith
m
was
u
s
ed
to
allev
iate
th
e
im
p
er
f
ec
ti
o
n
o
f
d
e
p
th
in
f
o
r
m
atio
n
d
u
e
to
th
e
Kin
ec
t
ca
m
er
a.
His
to
g
r
am
s
wer
e
g
en
er
ated
u
s
in
g
(
1
)
,
(
2
)
,
a
n
d
(
3
)
.
I
n
t
h
e
last
lay
er
,
th
e
m
u
lti
-
ch
an
n
el
f
ea
tu
r
e
is
p
r
o
ce
s
s
ed
in
to
a
s
in
g
le
-
c
h
a
n
n
el
f
ea
tu
r
e
a
n
d
p
r
o
ce
s
s
ed
b
y
an
ar
g
m
a
x
(
·
)
t
o
f
i
n
d
t
h
e
p
ea
k
in
d
e
x
,
t
h
er
eb
y
ca
lcu
latin
g
th
e
d
is
tan
ce
.
No
tab
ly
,
u
n
lik
e
p
r
e
v
io
u
s
tr
ain
in
g
d
atasets
o
f
3
D
ar
ch
itectu
r
e
th
at
g
en
er
ate
h
u
g
e
p
o
in
t
clo
u
d
s
(
ten
s
o
f
g
ig
a
b
y
t
es)
f
r
o
m
m
u
ltip
le
s
ce
n
es
in
th
e
d
atasets
,
o
u
r
m
o
d
el
o
n
ly
g
en
er
ates
h
is
to
g
r
am
s
f
r
o
m
o
n
e
s
ce
n
e,
wh
er
e
t
h
e
tr
a
in
in
g
d
atasets
ar
e
ju
s
t
1
5
.
1
M
B
.
Similar
ly
,
o
u
r
tr
ain
in
g
s
p
ee
d
is
ap
p
r
o
x
i
m
ately
ten
o
f
tim
es
f
aster
th
an
p
r
ev
io
u
s
3
D
U
-
NE
T
ar
c
h
itectu
r
es
[
2
]
,
[
2
0
]
,
an
d
[
2
1
]
.
W
h
ile
g
en
er
atin
g
h
is
to
g
r
am
s
in
th
e
tr
ain
in
g
d
atasets
,
we
d
ef
in
ed
o
p
tical
p
ar
a
m
eter
s
p
r
esen
te
d
in
T
ab
les 1
t
o
5
.
Fig
u
r
e
1
.
T
h
e
in
f
e
r
en
ce
p
i
p
elin
e
o
f
th
e
1
D
U
-
NE
T
ar
c
h
itectu
r
e
f
o
r
p
ix
el
-
wis
e
d
ep
t
h
im
ag
e
T
ab
le
1
.
Pre
-
d
ef
in
ed
p
ar
am
ete
r
s
to
g
en
er
ate
s
y
n
th
etic
h
is
to
g
r
am
s
P
a
r
a
me
t
e
r
V
a
l
u
e
N
u
mb
e
r
o
f
t
i
me
b
i
n
s
1
0
2
4
Te
mp
o
r
a
l
r
e
s
o
l
u
t
i
o
n
1
9
.
5
3
p
s
S
p
a
t
i
a
l
r
e
s
o
l
u
t
i
o
n
64
×
64
S
B
R
s
[
0
.
0
1
,
5
]
La
ser FW
H
M
1
1
7
.
1
8
p
s
La
ser FW
H
M
p
e
a
k
i
n
d
e
x
5
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
o
mp
u
ta
tio
n
a
lly
efficien
t p
ix
elw
is
e
d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
…
(
Yu
Zh
a
n
g
)
5937
T
ab
le
2
.
Do
wn
s
am
p
le
th
e
d
eta
ils
o
f
th
e
d
ee
p
n
eu
r
al
n
etwo
r
k
D
o
w
n
s
a
m
p
l
e
La
y
e
r
n
a
m
e
O
u
t
p
u
t
s
h
a
p
e
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
C
o
n
v
-
d
o
w
n
sam
p
l
e
,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
ma
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
,
5
1
2
)
R
e
LU
C
o
n
v
-
d
o
w
n
sam
p
l
e
,
K
(
7
)
,
S
(
2
)
,
P
(
3
)
+
B
a
t
c
h
n
o
r
ma
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
,
2
5
6
)
R
e
LU
C
o
n
v
-
d
o
w
n
sam
p
l
e
,
K
(
5
)
,
S
(
2
)
,
P
(
2
)
+
B
a
t
c
h
n
o
r
ma
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
,
1
2
8
)
R
e
LU
T
ab
le
3
.
C
o
n
v
o
lu
tio
n
al
la
y
er
s
d
etails o
f
th
e
d
ee
p
n
eu
r
al
n
etwo
r
k
C
o
n
v
-
same
sp
a
t
i
a
l
d
i
m
e
n
s
i
o
n
La
y
e
r
n
a
m
e
O
u
t
p
u
t
s
h
a
p
e
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
C
o
n
v
,
K
(
9
)
,
S
(
1
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
4
,
1
0
2
4
)
R
e
LU
C
o
n
v
,
K
(
9
)
,
S
(
1
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
4
,
1
0
2
4
)
R
e
LU
C
o
n
v
,
K
(
9
)
,
S
(
1
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
6
,
5
1
2
)
R
e
LU
C
o
n
v
,
K
(
9
)
,
S
(
1
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
6
,
5
1
2
)
R
e
LU
C
o
n
v
,
K
(
7
)
,
S
(
1
)
,
P
(
3
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
8
,
2
5
6
)
R
e
LU
C
o
n
v
,
K
(
7
)
,
S
(
1
)
,
P
(
3
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
8
,
2
5
6
)
R
e
LU
C
o
n
v
,
K
(
5
)
,
S
(
1
)
,
P
(
2
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
0
,
1
2
8
)
R
e
LU
C
o
n
v
,
K
(
5
)
,
S
(
1
)
,
P
(
2
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
0
,
1
2
8
)
R
e
LU
T
ab
le
4
.
De
-
c
o
n
v
o
lu
tio
n
al
lay
er
s
d
etails o
f
th
e
d
ee
p
n
eu
r
al
n
etwo
r
k
U
p
sam
p
l
e
La
y
e
r
n
a
m
e
O
u
t
p
u
t
s
h
a
p
e
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
C
o
n
v
Tr
a
n
s.,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
m
a
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
5
,
2
5
6
)
R
e
LU
C
o
n
v
Tr
a
n
s
,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
m
a
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
5
,
5
1
2
)
R
e
LU
C
o
n
v
Tr
a
n
s,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
m
a
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
5
,
1
0
2
4
)
R
e
LU
C
o
n
v
Tr
a
n
s,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
m
a
l
i
z
a
t
i
o
n
(
2
5
6
,
1
,
5
,
1
0
2
4
)
R
e
LU
T
ab
le
5
.
De
-
c
o
n
v
o
lu
tio
n
al
lay
er
s
d
etails o
f
th
e
d
ee
p
n
eu
r
al
n
etwo
r
k
R
e
f
i
n
e
La
y
e
r
n
a
m
e
O
u
t
p
u
t
s
h
a
p
e
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
C
o
n
v
,
K
(
9
)
,
S
(
2
)
,
P
(
4
)
+
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
(
2
5
6
,
1
,
1
,
1
0
2
4
)
-
T
h
e
d
ee
p
lear
n
in
g
m
o
d
el
is
im
p
lem
en
ted
u
s
in
g
Py
T
o
r
ch
an
d
r
u
n
s
o
n
an
NVI
DI
A
R
T
X
A1
0
0
0
GPU.
T
h
e
lear
n
in
g
r
ate
is
s
et
to
1
0
-
5
,
an
d
R
MSp
r
o
p
is
th
e
o
p
tim
ize
r
.
Ku
llb
ac
k
-
L
ei
b
ler
(
KL
)
d
i
v
er
g
en
ce
is
em
p
lo
y
e
d
as
th
e
lo
s
s
f
u
n
ctio
n
to
ev
alu
at
e
th
e
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
.
A
n
ea
r
ly
s
to
p
p
in
g
m
ec
h
an
is
m
is
in
co
r
p
o
r
ated
with
a
p
atien
ce
o
f
2
0
ep
o
c
h
s
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
T
h
e
d
ataset
c
o
m
p
r
is
es
5
0
,
0
0
0
h
is
to
g
r
am
s
f
o
r
tr
ain
in
g
,
with
a
n
ad
d
itio
n
al
5
,
0
0
0
f
o
r
v
alid
atio
n
d
u
r
i
n
g
tr
ain
i
n
g
.
Sig
n
al
-
to
-
b
ac
k
g
r
o
u
n
d
r
atio
s
(
SB
R
s
)
f
o
r
th
e
tr
ain
in
g
d
atasets
ar
e
s
et
to
5
,
2
.
5
,
1
,
0
.
5
,
0
.
2
,
0
.
1
,
0
.
0
5
,
an
d
0
.
0
1
,
co
n
s
is
ten
t
with
o
th
er
p
h
o
to
n
-
ef
f
icien
t
ar
ch
itectu
r
es
[
2
]
,
[
4
]
,
[
2
2
]
.
T
h
e
t
r
ain
in
g
an
d
v
alid
atio
n
lo
s
s
es
ar
e
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
is
s
etu
p
en
s
u
r
es
r
o
b
u
s
t
tr
ain
in
g
an
d
ev
alu
atio
n
o
f
th
e
d
ee
p
lear
n
i
n
g
m
o
d
el,
lev
er
a
g
in
g
Py
T
o
r
c
h
'
s
ca
p
ab
ilit
ie
s
an
d
h
ar
n
ess
in
g
th
e
co
m
p
u
tatio
n
al
p
o
wer
o
f
th
e
NVI
DI
A
R
T
X
A1
0
0
0
GPU
f
o
r
ef
f
icien
t
p
r
o
ce
s
s
in
g
.
Fig
u
r
e
2
.
KL
d
iv
er
g
e
n
ce
lo
s
s
cu
r
v
e
o
f
tr
ain
in
g
an
d
v
alid
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
9
3
4
-
5
9
4
1
5938
5.
Q
UANTI
T
A
T
I
V
E
E
VAL
U
AT
I
O
N
T
h
i
s
s
e
ct
i
o
n
a
s
s
e
s
s
es
t
h
e
p
r
e
c
i
s
i
o
n
o
f
d
e
p
t
h
r
e
c
o
n
s
t
r
u
c
t
i
o
n
a
c
h
i
e
v
e
d
b
y
o
u
r
d
e
e
p
l
e
a
r
n
i
n
g
a
r
c
h
i
t
e
c
t
u
r
e
.
I
t
j
u
x
t
a
p
o
s
es
i
t
w
i
t
h
c
o
n
v
e
n
t
io
n
a
l
m
a
x
i
m
u
m
li
k
e
l
i
h
o
o
d
(
ML
)
m
e
t
h
o
d
s
a
n
d
S
h
i
n
et
a
l
.
o
p
t
i
m
i
z
a
ti
o
n
-
b
a
s
e
d
a
l
g
o
r
i
t
h
m
.
S
y
n
t
h
e
ti
c
d
a
t
as
e
ts
ar
e
m
e
t
i
c
u
l
o
u
s
l
y
s
i
m
u
l
a
t
e
d
,
e
n
s
u
r
i
n
g
a
c
o
m
p
r
e
h
e
n
s
i
v
e
e
v
a
l
u
a
tio
n
f
r
a
m
e
w
o
r
k
.
5
.
1
.
Sy
nthet
ic
da
t
a
s
et
s
ev
a
lua
t
io
n
W
e
u
s
ed
d
ep
th
im
ag
es
in
th
e
Mid
d
leb
u
r
y
d
atasets
[
1
1
]
as
th
e
g
r
o
u
n
d
tr
u
th
(
GT
)
d
ep
t
h
im
ag
e
an
d
g
en
er
ated
s
y
n
th
etic
h
is
to
g
r
am
s
u
s
in
g
k
n
o
w
n
o
p
tical
p
ar
am
eter
s
f
o
r
o
u
r
n
etwo
r
k
’
s
e
v
alu
atio
n
.
T
h
e
d
atasets
wer
e
also
lev
er
ag
ed
b
y
o
t
h
er
SP
AD
-
b
ased
d
ep
th
im
ag
e
r
ec
o
n
s
tr
u
ctio
n
u
s
in
g
d
e
ep
lear
n
in
g
[
2
]
,
[
3
]
,
[
2
1
]
.
As
s
h
o
wn
in
Fig
u
r
e
3
,
o
u
r
n
etwo
r
k
is
r
o
b
u
s
t
f
o
r
lo
w
SB
R
s
.
W
e
also
co
m
p
ar
e
d
o
u
r
n
etwo
r
k
with
ML
an
d
Sh
in
et
a
l.
m
eth
o
d
s
,
wh
ich
ar
e
also
p
ix
el
-
wis
e
.
Sh
in
et
a
l
.
al
g
o
r
ith
m
also
u
s
ed
a
p
ix
el
-
av
er
ag
in
g
m
et
h
o
d
to
en
h
an
ce
th
e
ac
c
u
r
ac
y
o
f
s
p
ati
al
d
im
en
s
io
n
s
.
T
h
e
co
m
p
ar
is
o
n
ac
r
o
s
s
s
ev
en
d
if
f
e
r
en
t
SB
R
s
f
r
o
m
s
ev
en
s
ce
n
es.
T
h
e
r
esu
lts
ar
e
s
h
o
wn
in
T
ab
l
e
6
.
Fig
u
r
e
3
.
S
y
n
t
h
e
tic
d
at
asets
.
T
h
e
p
o
in
t
cl
o
u
d
is
s
im
u
l
ate
d
u
s
i
n
g
p
r
e
-
d
ef
in
e
d
o
p
ti
ca
l
p
a
r
a
m
et
er
s
.
D
if
f
e
r
en
t
SB
R
s
ar
e
1
,
2
.
5
,
a
n
d
2
.
M
AE
s
o
f
e
ac
h
r
ec
o
n
s
t
r
u
cte
d
i
m
a
g
e
v
er
s
u
s
th
e
GT
i
m
a
g
es
a
r
e
i
n
d
ic
a
ted
i
n
ea
c
h
i
m
ag
e
T
ab
le
6
.
Acc
u
r
ac
y
co
m
p
ar
is
o
n
s
am
o
n
g
th
r
ee
p
ix
el
-
wis
e
r
ec
o
n
s
tr
u
ctio
n
alg
o
r
ith
m
s
in
s
y
n
th
etic
tr
ain
in
g
d
atasets
A
l
g
o
r
i
t
h
m
S
B
R
B
o
o
k
Art
B
o
w
l
i
n
g
D
o
l
l
M
o
e
b
i
u
s
R
e
i
n
d
e
e
r
ML
5
2
.
5
9
2
.
5
2
2
.
1
3
2
.
5
5
2
.
4
8
2
.
5
6
2
.
5
4
.
7
8
4
.
6
6
4
.
3
2
4
.
7
0
4
.
5
3
4
.
5
7
1
6
.
8
7
6
.
7
9
6
.
3
2
6
.
7
3
6
.
5
4
6
.
6
9
S
h
i
n
e
t
a
l
.
5
2
.
2
5
2
.
1
4
2
.
0
7
2
.
2
1
2
.
1
1
2
.
5
4
2
.
5
4
.
5
3
4
.
4
4
4
.
0
3
4
.
5
8
4
.
4
6
4
.
5
3
1
6
.
4
4
6
.
2
3
6
.
0
2
6
.
5
7
6
.
5
2
6
.
4
5
1
D
U
N
ET
5
0
.
0
3
0
.
0
3
0
.
0
2
0
.
0
1
0
.
0
2
0
.
1
2
2
.
5
0
.
2
6
0
.
3
4
0
.
2
1
0
.
2
1
0
.
1
8
0
.
4
4
1
1
.
1
2
1
.
0
3
0
.
3
2
1
.
2
3
1
.
2
2
1
.
3
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
o
mp
u
ta
tio
n
a
lly
efficien
t p
ix
elw
is
e
d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
…
(
Yu
Zh
a
n
g
)
5939
5
.
2
.
Ca
pture
d
da
t
a
s
et
s
ev
a
lua
t
io
n
Ap
ar
t
f
r
o
m
th
e
ev
alu
atio
n
o
f
s
y
n
th
etic
d
atasets
,
we
a
ls
o
in
v
esti
g
ated
th
e
p
er
f
o
r
m
a
n
ce
o
f
ca
p
tu
r
ed
d
atasets
[
2
0
]
.
W
e
also
co
m
p
a
r
ed
o
u
r
d
ee
p
n
eu
r
al
n
etwo
r
k
with
ML
an
d
Sh
in
et
a
l.
m
et
h
o
d
.
As
s
h
o
wn
in
Fig
u
r
e
4
,
we
p
r
esen
t
th
e
r
ef
l
ec
tiv
ity
im
ag
es
to
r
ef
er
en
ce
s
p
atial
in
f
o
r
m
atio
n
.
T
h
e
in
te
n
s
ity
im
ag
es
wer
e
r
etr
iev
ed
f
r
o
m
t
h
e
s
ca
n
n
ed
p
o
in
t
clo
u
d
,
an
d
all
h
is
to
g
r
am
s
wer
e
tak
en
at
th
e
tem
p
o
r
al
d
i
m
en
s
io
n
.
R
eg
ar
d
i
n
g
ML
’
s
p
er
f
o
r
m
an
ce
,
th
e
r
ec
o
n
s
tr
u
ctio
n
d
e
p
th
im
ag
es
co
n
tain
n
u
m
er
o
u
s
NaN
v
alu
es
r
e
p
r
esen
ted
b
y
wh
it
e
p
ix
els
in
th
e
d
e
p
th
im
a
g
es
d
u
e
to
th
e
lo
w
p
h
o
t
o
n
c
o
u
n
ts
.
Ou
r
ap
p
r
o
ac
h
ac
h
iev
e
d
a
co
m
p
ar
ab
le
r
ec
o
n
s
tr
u
ctio
n
to
Sh
in
’
s
m
eth
o
d
.
No
ta
b
ly
,
Sh
in
’
s
m
eth
o
d
is
s
o
m
etim
es su
s
c
ep
tib
le
to
in
ten
s
e
am
b
ien
t lig
h
t.
Fo
r
ex
am
p
le
,
th
e
b
u
lb
was
n
o
t
r
ec
o
n
s
tr
u
cted
r
o
b
u
s
tly
in
th
e
lam
p
s
ce
n
e
.
An
d
o
u
r
m
et
h
o
d
ac
h
iev
ed
b
etter
v
is
u
aliza
tio
n
o
f
th
e
b
u
lb
.
Als
o
,
as
Sh
in
’
s
alg
o
r
it
h
m
s
in
v
o
lv
e
a
s
p
atial
av
er
ag
in
g
p
r
o
ce
s
s
,
th
e
s
p
atial
d
ep
th
m
ig
h
t
b
e
wo
r
s
e
if
p
ix
el
-
wis
e
d
ep
th
in
f
o
r
m
atio
n
is
n
o
t
r
ec
o
v
er
ed
ac
cu
r
ately
.
Fu
tu
r
e
wo
r
k
ca
n
e
m
p
lo
y
m
o
r
e
ad
v
a
n
ce
d
n
eu
r
al
n
etwo
r
k
s
,
s
u
ch
as a
g
r
ap
h
n
eu
r
al
n
etwo
r
k
(
GNN)
[
2
5
]
f
o
r
p
o
in
t c
lo
u
d
a
n
aly
s
is
[
2
3
]
.
Fig
u
r
e
4
.
R
ec
o
n
s
tr
u
cte
d
d
ep
t
h
im
ag
es o
f
th
e
ca
p
tu
r
ed
p
o
in
t
clo
u
d
.
T
h
er
e
ar
e
f
iv
e
s
ce
n
es: a
n
elep
h
an
t
d
o
ll,
a
h
allway
,
a
lam
p
,
a
b
all
o
n
a
s
tair
ca
s
e,
an
d
s
tu
f
f
o
n
a
tab
le.
I
m
ag
es o
f
r
e
f
lectiv
ity
,
in
ten
s
ity
,
an
d
r
ec
o
n
s
tr
u
ctio
n
ar
e
d
ep
icted
6.
CO
NCLU
SI
O
N
T
h
is
wo
r
k
p
r
esen
ts
a
c
o
m
p
a
ct
an
d
ac
cu
r
ate
1
D
d
e
p
th
im
ag
e
r
ec
o
n
s
tr
u
ctio
n
f
r
o
m
h
is
t
o
g
r
am
s
o
f
SP
AD
ar
r
ay
s
with
lo
w
SR
B
s
.
C
o
m
p
ar
ed
with
th
e
p
r
ev
io
u
s
3
D
d
ee
p
n
eu
r
al
n
etwo
r
k
s
t
h
at
r
eq
u
ir
e
ten
s
o
f
g
ig
ab
y
tes
o
f
tr
ain
in
g
d
ataset
s
o
f
p
o
i
n
t
clo
u
d
s
,
o
u
r
n
etwo
r
k
o
n
ly
r
e
q
u
ir
es
a
1
5
.
1
MB
h
is
to
g
r
am
tr
ai
n
in
g
d
ataset.
Ad
d
itio
n
ally
,
t
h
e
3
D
n
etwo
r
k
s
co
n
s
u
m
e
h
o
u
r
s
to
f
i
n
is
h
tr
ain
in
g
,
wh
er
ea
s
o
u
r
1
D
ar
ch
itectu
r
e
o
n
ly
r
eq
u
ir
es
1
2
m
i
n
u
tes.
Similar
ly
,
f
o
r
in
f
e
r
en
ce
,
th
e
h
ig
h
s
p
atial
r
eso
lu
tio
n
p
o
in
t
clo
u
d
f
o
r
3
D
n
etwo
r
k
s
s
h
o
u
ld
b
e
d
iv
id
ed
in
to
s
m
all
p
o
r
tio
n
s
,
f
o
r
ex
am
p
le,
1
/8
s
p
atial
r
eso
lu
tio
n
,
to
in
f
er
p
ar
tial
d
ep
th
im
ag
es
in
m
u
ltip
le
b
atch
es
an
d
s
titch
th
e
d
ep
th
i
m
ag
es
af
ter
war
d
s
.
3
D
co
n
v
o
l
u
tio
n
s
co
n
s
u
m
e
h
u
g
e
GPU
m
em
o
r
y
an
d
ca
n
n
o
t
b
e
p
r
o
ce
s
s
ed
in
o
n
e
b
atch
.
Ho
we
v
er
,
o
u
r
1
D
p
i
x
el
-
wis
e
ar
ch
itectu
r
e
d
o
es
n
o
t
h
av
e
th
e
m
em
o
r
y
o
v
er
f
lo
w
is
s
u
e
d
u
e
to
lig
h
tweig
h
t
1
D
co
n
v
o
lu
tio
n
s
,
m
ak
in
g
it
ea
s
ier
to
i
m
p
lem
en
t
o
n
em
b
ed
d
e
d
h
a
r
d
war
e
in
v
e
h
icles
o
r
d
r
o
n
es
f
o
r
p
r
ac
tical
ap
p
licatio
n
s
.
C
o
m
p
ar
ed
with
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
an
d
o
th
er
p
h
o
to
n
-
ef
f
icien
t
alg
o
r
ith
m
s
,
o
u
r
m
eth
o
d
s
s
h
o
w
h
ig
h
er
ac
c
u
r
ac
y
f
o
r
s
y
n
th
etic
d
atasets
.
As
f
o
r
th
e
ev
a
lu
atio
n
o
f
ca
p
tu
r
e
d
d
atasets
,
o
u
r
n
etwo
r
k
is
m
o
r
e
r
o
b
u
s
t
ag
ain
s
t
am
b
ien
t
lig
h
t.
T
h
e
lim
itatio
n
o
f
th
is
wo
r
k
is
th
at
n
o
s
p
atial
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
9
3
4
-
5
9
4
1
5940
in
f
o
r
m
atio
n
is
ex
tr
ac
ted
d
u
r
in
g
DL
tr
ain
in
g
d
u
e
to
th
e
p
ix
el
-
wis
e
p
r
o
ce
s
s
in
g
n
atu
r
e.
A
p
o
ten
tial
ap
p
r
o
ac
h
to
ad
d
r
ess
th
is
co
u
ld
in
v
o
lv
e
in
co
r
p
o
r
ati
n
g
a
lo
w
-
co
s
t RGB
im
ag
e
to
p
r
o
v
id
e
2
D
s
p
atial
s
tr
u
ctu
r
al
d
etails d
u
r
in
g
tr
ain
in
g
,
co
m
p
en
s
atin
g
f
o
r
t
h
e
lack
o
f
s
p
atial
r
eso
lu
tio
n
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Yu
Z
h
an
g
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Yim
in
g
Z
h
en
g
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
d
ec
lar
es th
at
th
e
r
e
is
n
o
co
n
f
lict o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
s
u
p
p
o
r
tin
g
th
is
s
tu
d
y
'
s
f
in
d
in
g
s
ar
e
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
t
h
o
r
,
YZ
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
D
.
S
h
i
n
e
t
a
l
.
,
“
P
h
o
t
o
n
-
e
f
f
i
c
i
e
n
t
i
ma
g
i
n
g
w
i
t
h
a
s
i
n
g
l
e
-
p
h
o
t
o
n
c
a
mera
,
”
N
a
t
u
r
e
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
7
,
n
o
.
1
,
J
u
n
.
2
0
1
6
,
d
o
i
:
1
0
.
1
0
3
8
/
n
c
o
mm
s
1
2
0
4
6
.
[
2
]
Z.
Z
a
n
g
,
D
.
X
i
a
o
,
a
n
d
D
.
D
a
y
-
U
e
i
Li
,
“
N
o
n
-
f
u
s
i
o
n
t
i
me
-
r
e
s
o
l
v
e
d
d
e
p
t
h
i
ma
g
e
r
e
c
o
n
st
r
u
c
t
i
o
n
u
s
i
n
g
a
h
i
g
h
l
y
e
f
f
i
c
i
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
,
”
O
p
t
i
c
s
Ex
p
res
s
,
v
o
l
.
2
9
,
n
o
.
1
3
,
p
.
1
9
2
7
8
,
J
u
n
.
2
0
2
1
,
d
o
i
:
1
0
.
1
3
6
4
/
o
e
.
4
2
5
9
1
7
.
[
3
]
G
.
Y
a
o
,
Y
.
C
h
e
n
,
Y
.
Li
u
,
X
.
H
u
,
a
n
d
Y
.
P
a
n
,
“
R
o
b
u
s
t
p
h
o
t
o
n
-
e
f
f
i
c
i
e
n
t
i
m
a
g
i
n
g
u
s
i
n
g
a
p
i
x
e
l
-
w
i
s
e
r
e
si
d
u
a
l
s
h
r
i
n
k
a
g
e
n
e
t
w
o
r
k
,
”
O
p
t
i
c
s
Ex
p
ress
,
v
o
l
.
3
0
,
n
o
.
1
1
,
p
.
1
8
8
5
6
,
M
a
y
2
0
2
2
,
d
o
i
:
1
0
.
1
3
6
4
/
o
e
.
4
5
2
5
9
7
.
[
4
]
D
.
X
i
a
o
,
Y
.
C
h
e
n
,
a
n
d
D
.
D
.
U
.
L
i
,
“
O
n
e
-
d
i
m
e
n
si
o
n
a
l
d
e
e
p
l
e
a
r
n
i
n
g
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
f
a
s
t
f
l
u
o
r
e
sc
e
n
c
e
l
i
f
e
t
i
me
i
ma
g
i
n
g
,
”
I
EEE
J
o
u
rn
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
Q
u
a
n
t
u
m
El
e
c
t
r
o
n
i
c
s
,
v
o
l
.
2
7
,
n
o
.
4
,
p
p
.
1
–
1
0
,
Ju
l
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TQ
E.
2
0
2
1
.
3
0
4
9
3
4
9
.
[
5
]
M
.
O
’
T
o
o
l
e
,
D
.
B
.
Li
n
d
e
l
l
,
a
n
d
G
.
W
e
t
z
s
t
e
i
n
,
“
C
o
n
f
o
c
a
l
n
o
n
-
l
i
n
e
-
of
-
s
i
g
h
t
i
ma
g
i
n
g
b
a
s
e
d
o
n
t
h
e
l
i
g
h
t
-
c
o
n
e
t
r
a
n
sf
o
r
m,”
N
a
t
u
re
,
v
o
l
.
5
5
5
,
n
o
.
7
6
9
6
,
p
p
.
3
3
8
–
3
4
1
,
M
a
r
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
3
8
/
n
a
t
u
r
e
2
5
4
8
9
.
[
6
]
X
.
S
u
,
Y
.
H
o
n
g
,
J
.
T.
Y
e
,
F
.
X
u
,
a
n
d
X
.
Y
u
a
n
,
“
M
o
d
e
l
-
g
u
i
d
e
d
i
t
e
r
a
t
i
v
e
d
i
f
f
u
s
i
o
n
sam
p
l
i
n
g
f
o
r
N
LO
S
r
e
c
o
n
st
r
u
c
t
i
o
n
,
”
I
EE
E
J
o
u
rn
a
l
o
f
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
Q
u
a
n
t
u
m
El
e
c
t
r
o
n
i
c
s
,
v
o
l
.
3
0
,
n
o
.
1
,
p
p
.
1
–
1
1
,
Ja
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TQ
E.
2
0
2
4
.
3
3
5
0
9
7
1
.
[
7
]
P
.
K
e
s
h
a
v
a
r
z
i
a
n
e
t
a
l
.
,
“
A
3
.
3
-
G
b
/
s
S
P
A
D
-
b
a
se
d
q
u
a
n
t
u
m
r
a
n
d
o
m
n
u
m
b
e
r
g
e
n
e
r
a
t
o
r
,
”
I
EEE
J
o
u
r
n
a
l
o
f
S
o
l
i
d
-
S
t
a
t
e
C
i
rc
u
i
t
s
,
v
o
l
.
5
8
,
n
o
.
9
,
p
p
.
2
6
3
2
–
2
6
4
7
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
S
C
.
2
0
2
3
.
3
2
7
4
6
9
2
.
[
8
]
A
.
To
n
t
i
n
i
,
L.
G
a
sp
a
r
i
n
i
,
N
.
M
a
ssari
,
a
n
d
R
.
P
a
sser
o
n
e
,
“
S
P
A
D
-
b
a
se
d
q
u
a
n
t
u
m
r
a
n
d
o
m
n
u
mb
e
r
g
e
n
e
r
a
t
o
r
w
i
t
h
a
n
n
t
h
-
o
r
d
e
r
r
a
n
k
a
l
g
o
r
i
t
h
m
o
n
F
P
G
A
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
i
rc
u
i
t
s
a
n
d
S
y
st
e
m
s
I
I
:
Ex
p
r
e
ss
Bri
e
f
s
,
v
o
l
.
6
6
,
n
o
.
1
2
,
p
p
.
2
0
6
7
–
2
0
7
1
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
TC
S
I
I
.
2
0
1
9
.
2
9
0
9
0
1
3
.
[
9
]
J
.
J
e
o
n
,
S
.
C
h
o
,
X
.
T
o
n
g
,
a
n
d
S
.
L
e
e
,
“
I
n
t
r
i
n
s
i
c
i
m
a
g
e
d
e
c
o
m
p
o
s
i
t
i
o
n
u
s
i
n
g
s
t
r
u
c
t
u
r
e
-
t
e
x
t
u
r
e
s
e
p
a
r
a
t
i
o
n
a
n
d
s
u
r
f
a
c
e
n
o
r
m
a
l
s
,
”
i
n
L
e
c
t
u
r
e
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
s
u
b
s
e
r
i
e
s
L
e
c
t
u
r
e
N
o
t
e
s
i
n
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
r
e
N
o
t
e
s
i
n
B
i
o
i
n
f
o
r
m
a
t
i
c
s
)
,
v
o
l
.
8
6
9
5
L
N
C
S
,
n
o
.
P
A
R
T
7
,
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
s
h
i
n
g
,
2
0
1
4
,
p
p
.
2
1
8
–
2
3
3
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
1
0
5
8
4
-
0
_
1
5
.
[
1
0
]
N
.
S
i
l
b
e
r
m
a
n
,
D
.
H
o
i
e
m,
P
.
K
o
h
l
i
,
a
n
d
R
.
F
e
r
g
u
s,
“
I
n
d
o
o
r
s
e
g
m
e
n
t
a
t
i
o
n
a
n
d
su
p
p
o
r
t
i
n
f
e
r
e
n
c
e
f
r
o
m
R
G
B
D
i
m
a
g
e
s,”
i
n
L
e
c
t
u
r
e
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
s
u
b
ser
i
e
s
L
e
c
t
u
r
e
N
o
t
e
s
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
re
N
o
t
e
s
i
n
Bi
o
i
n
f
o
rm
a
t
i
c
s)
,
v
o
l
.
7
5
7
6
LN
C
S
,
n
o
.
P
A
R
T
5
,
S
p
r
i
n
g
e
r
B
e
r
l
i
n
H
e
i
d
e
l
b
e
r
g
,
2
0
1
2
,
p
p
.
7
4
6
–
7
6
0
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
642
-
3
3
7
1
5
-
4
_
5
4
.
[
1
1
]
D
.
S
c
h
a
r
st
e
i
n
a
n
d
R
.
S
z
e
l
i
sk
i
,
“
A
t
a
x
o
n
o
m
y
a
n
d
e
v
a
l
u
a
t
i
o
n
o
f
d
e
n
se
t
w
o
-
f
r
a
me
s
t
e
r
e
o
c
o
r
r
e
sp
o
n
d
e
n
c
e
a
l
g
o
r
i
t
h
ms,”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
e
r V
i
si
o
n
,
v
o
l
.
4
7
,
n
o
.
1
–
3
,
p
p
.
7
–
4
2
,
A
p
r
.
2
0
0
2
,
d
o
i
:
1
0
.
1
0
2
3
/
A
:
1
0
1
4
5
7
3
2
1
9
9
7
7
.
[
1
2
]
J.
R
a
p
p
a
n
d
V
.
K
.
G
o
y
a
l
,
“
A
f
e
w
p
h
o
t
o
n
s
a
m
o
n
g
m
a
n
y
:
U
n
m
i
x
i
n
g
s
i
g
n
a
l
a
n
d
n
o
i
se
f
o
r
p
h
o
t
o
n
-
e
f
f
i
c
i
e
n
t
a
c
t
i
v
e
i
ma
g
i
n
g
,
”
I
E
E
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
m
a
g
i
n
g
,
v
o
l
.
3
,
n
o
.
3
,
p
p
.
4
4
5
–
4
5
9
,
S
e
p
.
2
0
1
7
,
d
o
i
:
1
0
.
1
1
0
9
/
t
c
i
.
2
0
1
7
.
2
7
0
6
0
2
8
.
[
1
3
]
J.
Ta
c
h
e
l
l
a
e
t
a
l
.
,
“
B
a
y
e
si
a
n
3
D
r
e
c
o
n
s
t
r
u
c
t
i
o
n
o
f
c
o
mp
l
e
x
sc
e
n
e
s
f
r
o
m
si
n
g
l
e
-
p
h
o
t
o
n
l
i
d
a
r
d
a
t
a
,
”
S
I
AM
J
o
u
rn
a
l
o
n
I
m
a
g
i
n
g
S
c
i
e
n
c
e
s
,
v
o
l
.
1
2
,
n
o
.
1
,
p
p
.
5
2
1
–
5
5
0
,
Jan
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
3
7
/
1
8
M
1
1
8
3
9
7
2
.
[
1
4
]
J.
T
a
c
h
e
l
l
a
e
t
a
l
.
,
“
R
e
a
l
-
t
i
me
3
D
r
e
c
o
n
st
r
u
c
t
i
o
n
f
r
o
m
s
i
n
g
l
e
-
p
h
o
t
o
n
l
i
d
a
r
d
a
t
a
u
si
n
g
p
l
u
g
-
a
n
d
-
p
l
a
y
p
o
i
n
t
c
l
o
u
d
d
e
n
o
i
s
e
r
s,”
N
a
t
u
r
e
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
1
0
,
n
o
.
1
,
N
o
v
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
4
6
7
-
0
1
9
-
1
2
9
4
3
-
7.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
C
o
mp
u
ta
tio
n
a
lly
efficien
t p
ix
elw
is
e
d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
…
(
Yu
Zh
a
n
g
)
5941
[
1
5
]
J.
K
o
o
,
A
.
H
a
l
i
mi
,
a
n
d
S
.
M
c
l
a
u
g
h
l
i
n
,
“
A
B
a
y
e
s
i
a
n
b
a
se
d
d
e
e
p
u
n
r
o
l
l
i
n
g
a
l
g
o
r
i
t
h
m
f
o
r
si
n
g
l
e
-
p
h
o
t
o
n
l
i
d
a
r
sy
s
t
e
ms,”
I
EE
E
J
o
u
rn
a
l
o
n
S
e
l
e
c
t
e
d
T
o
p
i
c
s
i
n
S
i
g
n
a
l
Pro
c
e
ssi
n
g
,
v
o
l
.
1
6
,
n
o
.
4
,
p
p
.
7
6
2
–
7
7
4
,
J
u
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
JS
TSP
.
2
0
2
2
.
3
1
7
0
2
2
8
.
[
1
6
]
J.
Ta
c
h
e
l
l
a
,
Y
.
A
l
t
m
a
n
n
,
M
.
M
a
r
q
u
e
z
,
H
.
A
r
g
u
e
l
l
o
-
F
u
e
n
t
e
s,
J.
-
Y
.
To
u
r
n
e
r
e
t
,
a
n
d
S
.
M
c
La
u
g
h
l
i
n
,
“
B
a
y
e
si
a
n
3
D
r
e
c
o
n
st
r
u
c
t
i
o
n
o
f
su
b
s
a
m
p
l
e
d
m
u
l
t
i
s
p
e
c
t
r
a
l
si
n
g
l
e
-
p
h
o
t
o
n
l
i
d
a
r
si
g
n
a
l
s,”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
m
a
g
i
n
g
,
v
o
l
.
6
,
p
p
.
2
0
8
–
2
2
0
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
t
c
i
.
2
0
1
9
.
2
9
4
5
2
0
4
.
[
1
7
]
J.
L
i
,
B
.
W
a
n
g
,
H
.
M
a
,
L.
G
a
o
,
a
n
d
H
.
F
u
,
“
V
i
s
u
a
l
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
n
d
t
r
a
c
k
i
n
g
me
t
h
o
d
b
a
se
d
o
n
c
o
r
n
e
r
f
l
o
w
d
e
t
e
c
t
i
o
n
,
”
I
E
C
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
a
t
i
c
s
,
v
o
l
.
1
,
n
o
.
1
,
p
p
.
3
–
9
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
6
2
7
6
2
/
t
i
s
.
2
0
2
4
.
1
3
6
8
9
5
.
[
1
8
]
F
.
W
a
n
g
a
n
d
S
.
Y
i
,
“
S
p
a
t
i
o
-
t
e
m
p
o
r
a
l
f
e
a
t
u
r
e
s
o
f
t
c
o
r
r
e
l
a
t
i
o
n
c
o
n
c
a
t
e
n
a
t
i
o
n
a
g
g
r
e
g
a
t
i
o
n
s
t
r
u
c
t
u
r
e
f
o
r
v
i
d
e
o
a
c
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
n
e
t
w
o
r
k
s,”
I
E
C
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
e
n
si
n
g
,
C
o
m
m
u
n
i
c
a
t
i
o
n
,
a
n
d
C
o
n
t
r
o
l
,
v
o
l
.
1
,
n
o
.
1
,
p
p
.
6
0
–
7
1
,
O
c
t
.
2
0
2
4
,
d
o
i
:
1
0
.
6
2
7
6
2
/
t
sc
c
.
2
0
2
4
.
2
1
2
7
5
1
.
[
1
9
]
Z.
S
u
n
,
D
.
B
.
Li
n
d
e
l
l
,
O
.
S
o
l
g
a
a
r
d
,
a
n
d
G
.
W
e
t
z
st
e
i
n
,
“
S
P
A
D
n
e
t
:
d
e
e
p
R
G
B
-
S
P
A
D
sen
s
o
r
f
u
s
i
o
n
a
ss
i
st
e
d
b
y
mo
n
o
c
u
l
a
r
d
e
p
t
h
e
st
i
mat
i
o
n
,
”
O
p
t
i
c
s
Ex
p
ress
,
v
o
l
.
2
8
,
n
o
.
1
0
,
p
.
1
4
9
4
8
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
1
3
6
4
/
o
e
.
3
9
2
3
8
6
.
[
2
0
]
X
.
Z
h
a
o
,
X
.
Ji
a
n
g
,
A
.
H
a
n
,
T.
M
a
o
,
W
.
H
e
,
a
n
d
Q
.
C
h
e
n
,
“
P
h
o
t
o
n
-
e
f
f
i
c
i
e
n
t
3
D
r
e
c
o
n
st
r
u
c
t
i
o
n
e
mp
l
o
y
i
n
g
a
e
d
g
e
e
n
h
a
n
c
e
m
e
n
t
met
h
o
d
,
”
O
p
t
i
c
s
Ex
p
ress
,
v
o
l
.
3
0
,
n
o
.
2
,
p
.
1
5
5
5
,
Ja
n
.
2
0
2
2
,
d
o
i
:
1
0
.
1
3
6
4
/
o
e
.
4
4
6
3
6
9
.
[
2
1
]
J.
P
e
n
g
,
Z.
X
i
o
n
g
,
X
.
H
u
a
n
g
,
Z
.
P
.
Li
,
D
.
Li
u
,
a
n
d
F
.
X
u
,
“
P
h
o
t
o
n
-
e
f
f
i
c
i
e
n
t
3
D
i
m
a
g
i
n
g
w
i
t
h
a
n
o
n
-
l
o
c
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
i
n
L
e
c
t
u
re
N
o
t
e
s
i
n
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
(
i
n
c
l
u
d
i
n
g
s
u
b
ser
i
e
s
L
e
c
t
u
r
e
N
o
t
e
s
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
L
e
c
t
u
re
N
o
t
e
s
i
n
Bi
o
i
n
f
o
rm
a
t
i
c
s)
,
v
o
l
.
1
2
3
5
1
LN
C
S
,
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
sh
i
n
g
,
2
0
2
0
,
p
p
.
2
2
5
–
2
4
1
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
030
-
5
8
5
3
9
-
6
_
1
4
.
[
2
2
]
D
.
L.
S
n
y
d
e
r
a
n
d
M
.
I
.
M
i
l
l
e
r
,
“
S
e
l
f
-
e
x
c
i
t
i
n
g
p
o
i
n
t
p
r
o
c
e
ss
e
s,”
i
n
R
a
n
d
o
m
Po
i
n
t
Pr
o
c
e
ss
e
s
i
n
T
i
m
e
a
n
d
S
p
a
c
e
,
S
p
r
i
n
g
e
r
N
e
w
Y
o
r
k
,
1
9
9
1
,
p
p
.
2
8
7
–
340
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
1
-
4
6
1
2
-
3
1
6
6
-
0
_
6
.
[
2
3
]
W
.
S
h
i
a
n
d
R
.
R
a
j
k
u
m
a
r
,
“
P
o
i
n
t
-
G
N
N
:
G
r
a
p
h
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
3
D
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
i
n
a
p
o
i
n
t
c
l
o
u
d
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
2
0
,
p
p
.
1
7
0
8
–
1
7
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
2
6
0
0
.
2
0
2
0
.
0
0
1
7
8
.
[
2
4
]
Z.
Za
n
g
a
n
d
D
.
D
.
U
e
i
Li
,
“
O
b
j
e
c
t
c
l
a
ssi
f
i
c
a
t
i
o
n
t
h
r
o
u
g
h
h
e
t
e
r
o
g
e
n
e
o
u
s
f
o
g
w
i
t
h
a
f
a
st
d
a
t
a
-
d
r
i
v
e
n
a
l
g
o
r
i
t
h
m
u
si
n
g
a
l
o
w
-
c
o
s
t
si
n
g
l
e
-
p
h
o
t
o
n
a
v
a
l
a
n
c
h
e
d
i
o
d
e
a
r
r
a
y
,
”
O
p
t
i
c
s
Ex
p
res
s
,
v
o
l
.
3
2
,
n
o
.
1
9
,
p
.
3
3
2
9
4
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
3
6
4
/
O
E.
5
2
7
2
4
4
.
[
2
5
]
W
.
Z
h
a
n
g
a
n
d
Q
.
H
o
n
g
,
“
M
o
d
e
l
i
n
g
b
r
a
i
n
f
u
n
c
t
i
o
n
a
l
n
e
t
w
o
r
k
s
u
s
i
n
g
g
r
a
p
h
n
e
u
r
a
l
n
e
t
w
o
r
k
s
:
A
r
e
v
i
e
w
a
n
d
c
l
i
n
i
c
a
l
a
p
p
l
i
c
a
t
i
o
n
,
”
I
EC
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
a
t
i
c
s
,
v
o
l
.
1
,
n
o
.
2
,
p
p
.
5
8
–
6
8
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
6
2
7
6
2
/
t
i
s
.
2
0
2
4
.
6
8
0
9
5
9
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Yu
Zh
a
n
g
re
c
e
iv
e
d
h
is
B.
E
n
g
.
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
ri
n
g
fro
m
th
e
U
n
iv
e
rsit
y
o
f
Ke
n
t,
U
n
it
e
d
Ki
n
g
d
o
m
,
in
2
0
2
4
.
He
is
c
u
rre
n
tl
y
p
u
rsu
i
n
g
th
e
M
.
S
.
d
e
g
re
e
in
e
lec
tri
c
a
l
e
n
g
in
e
e
rin
g
a
t
th
e
Viterb
i
S
c
h
o
o
l
o
f
E
n
g
i
n
e
e
rin
g
,
U
n
iv
e
rsit
y
o
f
S
o
u
t
h
e
r
n
Ca
li
fo
rn
ia,
L
o
s
An
g
e
les
,
U
n
it
e
d
S
tate
s
,
sin
c
e
Ja
n
u
a
ry
2
0
2
5
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
y
z
3
2
4
@u
sc
.
e
d
u
.
Yim
in
g
Z
h
e
n
g
re
c
e
iv
e
d
a
B.
En
g
.
d
e
g
re
e
in
ra
il
tran
sit
sig
n
a
l
a
n
d
c
o
n
tr
o
l
fro
m
Be
ij
in
g
Un
i
o
n
Un
iv
e
rsit
y
,
Be
ij
in
g
,
C
h
in
a
,
a
n
d
a
b
a
c
h
e
lo
r’s
d
e
g
re
e
in
m
a
n
a
g
e
m
e
n
t
in
tec
h
n
o
l
o
g
ica
l
sy
ste
m
s
fro
m
Ur
a
l
S
tate
Un
iv
e
rsit
y
o
f
Ra
il
wa
y
Tran
sp
o
r
t,
Ye
k
a
terin
b
u
r
g
,
Ru
ss
ia,
b
o
t
h
in
2
0
2
3
.
He
is
c
u
rre
n
tl
y
p
u
rs
u
in
g
th
e
M
.
S
.
d
e
g
re
e
in
e
lec
tri
c
a
l
e
n
g
in
e
e
rin
g
a
t
t
h
e
Viterb
i
S
c
h
o
o
l
o
f
E
n
g
i
n
e
e
rin
g
,
U
n
iv
e
rsit
y
o
f
S
o
u
th
e
r
n
Ca
li
f
o
rn
ia,
Lo
s
A
n
g
e
les
,
U
n
it
e
d
S
tate
s
,
sin
c
e
Ja
n
u
a
ry
2
0
2
5
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
y
z
h
e
n
g
5
2
@u
sc
.
e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.