I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
2
9
9
~
3
3
0
8
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
2
9
9
-
3
3
0
8
3299
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Enha
ncing
pedest
ria
n det
e
ction in
a
dv
erse cond
ition
s usin
g
YO
L
O
v
5
s with
a
da
ptive weig
ht
ed
fusio
n ef
ficien
t
ch
a
nnel
a
tt
ention
mo
dule
O
um
a
y
m
a
Ra
chidi
,
B
a
dr
B
o
uo
uli
d I
dris
s
i,
Cha
f
ik
E
d
-
Da
hm
a
ni
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
me
c
h
a
n
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
,
S
c
h
o
o
l
o
f
A
r
t
s
a
n
d
C
r
a
f
t
s
,
M
o
u
l
a
y
I
smai
l
U
n
i
v
e
r
si
t
y
,
M
e
k
n
e
s,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
6
,
2
0
2
5
R
ev
is
ed
Ma
y
1
9
,
2
0
2
6
Acc
ep
ted
J
u
n
1
9
,
2
0
2
6
P
e
d
e
strian
d
e
tec
ti
o
n
c
o
n
sti
tu
tes
a
c
rit
ica
l
tas
k
with
in
a
d
v
a
n
c
e
d
d
riv
e
r
a
ss
istan
c
e
sy
ste
m
s
(AD
AS),
wh
e
re
re
li
a
b
le
id
e
n
ti
fica
ti
o
n
o
f
p
e
d
e
strian
s
is
e
ss
e
n
ti
a
l
fo
r
e
n
su
rin
g
v
e
h
ic
u
l
a
r
sa
fe
ty
.
Alth
o
u
g
h
d
e
e
p
lea
rn
in
g
h
a
s
su
b
sta
n
ti
a
ll
y
imp
r
o
v
e
d
d
e
tec
ti
o
n
p
e
rfo
rm
a
n
c
e
,
e
x
isti
n
g
sta
te
-
of
-
th
e
-
a
rt
m
o
d
e
ls
c
o
n
ti
n
u
e
t
o
e
x
h
i
b
it
n
o
tab
le
d
e
g
ra
d
a
ti
o
n
in
a
d
v
e
rse
we
a
th
e
r
c
o
n
d
i
ti
o
n
s
a
n
d
lo
w
-
li
g
h
t.
To
m
it
i
g
a
te
th
e
se
c
h
a
ll
e
n
g
e
s,
t
h
is
st
u
d
y
i
n
tro
d
u
c
e
s
a
n
e
n
h
a
n
c
e
d
p
e
d
e
strian
d
e
tec
ti
o
n
fra
m
e
wo
rk
b
a
se
d
o
n
y
o
u
o
n
ly
lo
o
k
o
n
e
v
er
s
io
n
5
(
YO
LOv
5
s
)
,
re
train
e
d
o
n
a
n
a
u
g
m
e
n
ted
c
o
m
m
o
n
o
b
jec
t
in
c
o
n
tex
t
(
COCO
)
d
a
tas
e
t
fo
c
u
se
d
o
n
t
h
e
p
e
rso
n
c
las
s.
Ad
d
i
ti
o
n
a
ll
y
,
a
n
o
v
e
l,
li
g
h
twe
i
g
h
t
,
a
n
d
a
d
a
p
ti
v
e
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
c
a
ll
e
d
:
t
h
e
we
ig
h
ted
fu
si
o
n
e
fficie
n
t
c
h
a
n
n
e
l
a
tt
e
n
t
io
n
(W
F
-
ECA)
m
o
d
u
le
is
in
c
o
rp
o
ra
ted
in
t
o
th
e
d
e
tec
ti
o
n
a
rc
h
i
tec
tu
re
.
T
h
e
W
F
-
ECA
m
o
d
u
le
se
lec
ti
v
e
ly
f
o
c
u
ss
e
s
on
imp
o
rtan
t
fe
a
tu
re
s
with
o
u
t
c
o
m
p
ro
m
isin
g
c
o
m
p
u
tatio
n
a
l
e
ffi
c
ien
c
y
o
r
in
fe
re
n
c
e
sp
e
e
d
.
Co
m
p
a
ra
ti
v
e
e
x
p
e
rime
n
ts
d
e
m
o
n
stra
te
a
5
%
i
n
c
re
a
se
in
m
e
a
n
a
v
e
ra
g
e
p
re
c
isio
n
(m
AP)
in
c
o
m
p
a
riso
n
to
th
e
b
a
se
li
n
e
m
o
d
e
l,
th
e
re
b
y
d
e
m
o
n
stra
ti
n
g
t
h
e
e
ffica
c
y
o
f
th
e
p
ro
p
o
se
d
a
tt
e
n
ti
o
n
m
o
d
u
le
in
i
m
p
ro
v
i
n
g
d
e
tec
ti
o
n
ro
b
u
st
n
e
ss
u
n
d
e
r
c
h
a
l
len
g
i
n
g
e
n
v
ir
o
n
m
e
n
tal
c
o
n
d
it
i
o
n
s.
T
h
e
se
fin
d
i
n
g
s
h
i
g
h
li
g
h
t
th
e
p
o
ten
ti
a
l
o
f
a
tt
e
n
ti
o
n
-
b
a
se
d
m
e
c
h
a
n
ism
s
t
o
e
n
h
a
n
c
e
p
e
d
e
strian
d
e
tec
ti
o
n
p
e
rfo
rm
a
n
c
e
in
re
a
l
-
wo
rl
d
AD
AS
a
p
p
li
c
a
ti
o
n
s.
K
ey
w
o
r
d
s
:
A
d
v
a
n
c
e
d
d
r
i
v
e
r
a
s
s
i
s
t
a
n
c
e
s
y
s
t
e
m
C
o
m
p
u
ter
v
is
io
n
E
f
f
icien
t c
h
an
n
el
atten
tio
n
Ped
estrian
d
etec
tio
n
Yo
u
o
n
ly
lo
o
k
o
n
c
e
v
er
s
io
n
5
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
:
Ou
m
ay
m
a
R
ac
h
id
i
Dep
ar
tm
en
t o
f
E
lectr
o
m
ec
h
a
n
ical
E
n
g
in
ee
r
in
g
,
Sch
o
o
l o
f
Ar
ts
an
d
C
r
af
ts
,
Mo
u
lay
I
s
m
ail
Un
iv
er
s
ity
Ma
r
jan
e
2
,
B
P 1
5
2
9
0
,
Al
-
Ma
n
s
o
u
r
,
Me
k
n
es,
M
o
r
o
cc
o
E
m
ail: o
u
m
.
r
ac
h
id
i@
ed
u
.
u
m
i.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
last
f
ew
y
ea
r
s
,
ad
v
an
ce
d
d
r
iv
er
ass
is
tan
ce
s
y
s
tem
s
(
ADAS)
[
1
]
h
av
e
g
ai
n
ed
s
ig
n
if
ican
t
atten
tio
n
d
u
e
to
th
eir
cr
u
cia
l
r
o
le
in
e
n
h
an
cin
g
v
e
h
icle
s
af
ety
.
T
h
ese
s
y
s
tem
s
u
tili
ze
a
wid
e
r
an
g
e
o
f
f
u
n
ctio
n
alities
in
o
r
d
e
r
to
ad
d
r
ess
v
ar
io
u
s
ch
allen
g
e
s
en
co
u
n
ter
ed
d
u
r
i
n
g
d
r
iv
i
n
g
.
Am
o
n
g
th
ese
f
u
n
ctio
n
alities
,
p
ed
estrian
d
etec
tio
n
em
er
g
es
as
a
v
ital
co
m
p
o
n
e
n
t,
d
u
e
to
th
e
r
is
in
g
in
cid
en
ce
o
f
ac
cid
en
ts
in
v
o
lv
in
g
p
e
d
estrian
s
.
C
o
n
s
e
q
u
en
tly
,
im
p
o
r
tan
t
r
esear
ch
e
f
f
o
r
ts
h
a
v
e
b
ee
n
d
ir
ec
ted
to
war
d
en
h
an
cin
g
th
e
ac
cu
r
ac
y
an
d
s
p
ee
d
o
f
p
e
d
estrian
d
etec
tio
n
s
y
s
tem
s
.
I
n
itially
r
ely
in
g
o
n
tr
ad
itio
n
al
ap
p
r
o
ac
h
es
s
u
ch
as
h
is
to
g
r
am
o
f
o
r
ie
n
ted
g
r
ad
ien
ts
(
HOG)
[
2
]
co
m
b
in
ed
with
s
u
p
p
o
r
t v
ec
to
r
m
ac
h
in
es
(
SVM)
[
3
]
,
p
ed
estrian
d
etec
tio
n
h
as
s
ee
n
n
o
tab
le
ad
v
an
ce
m
e
n
ts
with
th
e
ad
v
e
n
t
o
f
d
ee
p
lear
n
in
g
[
4
]
.
Desp
ite
th
ese
ad
v
a
n
ce
m
en
ts
,
p
ed
estrian
d
etec
tio
n
co
n
tin
u
es
to
p
r
esen
t
co
n
s
id
er
ab
le
ch
allen
g
es,
p
ar
ti
cu
lar
ly
in
lo
w
-
lig
h
t e
n
v
ir
o
n
m
e
n
ts
an
d
u
n
d
er
ad
v
er
s
e
wea
th
er
co
n
d
itio
n
s
[
5
]
.
Fo
r
ex
am
p
le,
n
i
g
h
t
-
tim
e
d
r
iv
in
g
s
ig
n
if
ican
tly
r
ed
u
ce
s
im
ag
e
co
n
tr
ast,
m
ak
i
n
g
p
e
d
estrian
s
ilh
o
u
ettes
less
d
is
tin
g
u
is
h
ab
le
f
r
o
m
th
e
b
ac
k
g
r
o
u
n
d
.
Hea
v
y
r
ain
ca
n
in
tr
o
d
u
ce
m
o
tio
n
b
l
u
r
an
d
r
ef
lectiv
e
r
o
ad
s
u
r
f
ac
es
th
at
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
9
9
-
3
3
0
8
3300
d
is
to
r
t
v
is
u
al
f
ea
tu
r
es.
Similar
ly
,
d
en
s
e
f
o
g
r
ed
u
ce
s
v
is
ib
ilit
y
r
an
g
e
an
d
s
u
p
p
r
ess
es
ed
g
e
in
f
o
r
m
atio
n
,
wh
ile
s
tr
o
n
g
b
ac
k
lig
h
tin
g
d
u
r
i
n
g
s
u
n
r
is
e
o
r
s
u
n
s
et
ca
n
ca
u
s
e
p
ed
estrian
s
to
ap
p
ea
r
as
d
ar
k
s
ilh
o
u
ettes.
T
h
ese
s
ce
n
ar
io
s
o
f
ten
lead
to
m
is
s
e
d
d
etec
tio
n
s
(
f
alse
n
eg
ativ
es)
o
r
r
ed
u
ce
d
d
etec
tio
n
co
n
f
id
e
n
ce
in
co
n
v
en
tio
n
al
p
ed
estrian
d
etec
tio
n
m
o
d
els.
Nu
m
er
o
u
s
s
tu
d
ies
h
av
e
tr
ied
t
o
ad
d
r
ess
th
ese
lim
itatio
n
s
b
y
p
r
o
p
o
s
in
g
m
eth
o
d
s
th
at
en
h
an
ce
r
o
b
u
s
tn
ess
an
d
a
cc
u
r
ac
y
in
s
u
c
h
co
m
p
lex
s
ce
n
ar
io
s
[
6
]
–
[
9
]
.
T
o
in
cr
ea
s
e
t
h
e
m
o
d
el’
s
ac
c
u
r
ac
y
i
n
lo
w
-
lig
h
t
co
n
d
itio
n
s
,
Nata
p
r
awir
a
et
a
l.
[
1
0
]
p
r
o
p
o
s
ed
a
m
u
lti
-
s
p
ec
tr
al
p
ed
estrian
d
et
ec
tio
n
m
eth
o
d
b
ased
o
n
th
e
y
o
u
o
n
ly
lo
o
k
o
n
e
(
YO
L
O
)
v
3
ar
ch
itectu
r
e,
lev
er
ag
in
g
i
n
p
u
ts
f
r
o
m
r
e
d
,
g
r
ee
n
,
b
lu
e
(
R
GB
)
,
th
er
m
al,
an
d
co
m
b
in
e
d
m
u
lti
-
s
p
ec
tr
al
im
a
g
in
g
m
o
d
alities
.
I
n
th
is
ap
p
r
o
ac
h
,
th
e
th
r
ee
ch
an
n
els
o
f
th
e
R
GB
im
ag
e
wer
e
in
teg
r
ated
with
th
e
s
in
g
le
c
h
an
n
el
f
r
o
m
th
e
th
er
m
al
im
ag
e
to
f
o
r
m
a
u
n
if
ied
f
o
u
r
-
ch
a
n
n
el
in
p
u
t
r
ep
r
esen
tati
o
n
.
R
o
s
zy
k
et
a
l.
[
1
1
]
em
p
l
o
y
ed
th
e
YOL
Ov
4
ar
ch
itectu
r
e
to
d
ev
elo
p
a
r
ap
id
,
lo
w
-
laten
c
y
m
u
lti
-
s
p
ec
tr
al
p
ed
estrian
d
etec
tio
n
s
y
s
tem
tailo
r
ed
f
o
r
au
to
n
o
m
o
u
s
d
r
i
v
in
g
a
p
p
licatio
n
s
.
T
h
e
s
tu
d
y
ex
p
lo
r
ed
v
ar
io
u
s
m
o
d
el
ar
c
h
itectu
r
es
an
d
f
u
s
io
n
s
tr
ateg
ies,
with
f
ea
tu
r
e
-
lev
el
f
u
s
io
n
—
r
ef
er
r
ed
to
as
YOL
Ov
4
-
m
i
d
d
l
e
—
em
e
r
g
in
g
as
t
h
e
m
o
s
t
ef
f
ec
tiv
e
in
b
alan
cin
g
d
etec
tio
n
ac
cu
r
ac
y
a
n
d
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
.
Fo
llo
win
g
a
s
im
ilar
o
b
jectiv
e
b
u
t
a
d
o
p
tin
g
a
tr
an
s
f
er
lear
n
in
g
ap
p
r
o
ac
h
,
Hu
et
a
l.
[
1
2
]
em
p
lo
y
ed
C
y
cleG
AN
[
1
3
]
to
p
r
o
d
u
ce
s
y
n
th
etic
in
f
r
ar
e
d
im
ag
e
s
f
r
o
m
v
is
ib
le
-
lig
h
t
in
p
u
ts
,
th
er
eb
y
au
g
m
en
tin
g
th
e
co
m
p
u
ter
v
is
io
n
ce
n
te
r
(
C
VC
)
-
0
9
d
ataset.
T
h
is
au
g
m
en
te
d
d
ataset
was
th
en
u
s
ed
to
tr
ain
an
d
ev
alu
ate
YOL
Ov
3
[
1
4
]
an
d
f
aste
r
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
(
R
-
C
NN
)
[
1
5
]
m
o
d
els,
aim
in
g
to
en
h
a
n
ce
p
e
d
estrian
d
etec
tio
n
ac
cu
r
ac
y
in
lo
w
-
lig
h
t e
n
v
ir
o
n
m
e
n
ts
.
T
o
ad
d
r
ess
th
e
ch
allen
g
es
ca
u
s
ed
b
y
ad
v
er
s
e
wea
th
er
c
o
n
d
it
io
n
s
,
W
an
g
et
a
l.
[
1
6
]
p
r
o
p
o
s
ed
a
h
y
b
r
i
d
ap
p
r
o
ac
h
co
m
b
in
in
g
p
h
y
s
ical
m
o
d
elin
g
with
d
ee
p
lea
r
n
in
g
tec
h
n
iq
u
es.
T
h
e
r
esu
l
tin
g
m
o
d
el,
ca
lled
p
h
y
s
ics
-
awa
r
e
an
d
c
o
n
d
itio
n
a
l
(
PC
)
-
YOL
O,
u
s
es
two
n
o
v
el
co
m
p
o
n
en
ts
:
th
e
en
h
an
ce
d
p
h
y
s
ics
-
awa
r
e
b
lo
ck
(
E
PB
)
an
d
th
e
co
n
d
itio
n
al
wid
th
d
iv
er
s
if
icatio
n
b
r
an
ch
b
lo
ck
(
C
2
f
_
W
DB
B
)
.
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
is
ar
ch
itectu
r
e
was
v
alid
ated
th
r
o
u
g
h
ev
alu
atio
n
s
o
n
th
e
R
ain
y
C
ity
s
ca
p
e
an
d
Haz
y
C
ity
s
ca
p
e
d
ataset
s
.
L
iu
an
d
L
in
[
1
7
]
in
c
o
r
p
o
r
ate
d
th
e
d
ar
k
ch
an
n
el
d
eh
az
in
g
alg
o
r
ith
m
in
to
th
e
YOL
Ov
7
f
r
am
ewo
r
k
[
1
8
]
,
an
d
d
em
o
n
s
tr
ated
a
n
o
tab
le
en
h
a
n
ce
m
en
t
in
d
eh
az
i
n
g
e
f
f
icien
cy
.
T
h
is
in
te
g
r
atio
n
s
ig
n
if
ica
n
tly
im
p
r
o
v
ed
th
e
p
er
f
o
r
m
an
ce
o
f
YOL
Ov
7
in
f
o
g
g
y
co
n
d
itio
n
s
b
y
en
h
an
cin
g
th
e
m
o
d
el'
s
ab
ilit
y
to
d
etec
t
o
b
jects
in
lo
w
-
v
is
ib
ilit
y
en
v
ir
o
n
m
en
ts
.
Hn
ewa
et
a
l.
[
1
9
]
in
tr
o
d
u
ce
d
a
n
o
v
el
f
r
am
ewo
r
k
em
p
lo
y
in
g
cr
o
s
s
-
m
o
d
ality
k
n
o
wled
g
e
d
is
till
atio
n
(
C
M
KD)
to
en
h
an
ce
th
e
ef
f
ec
tiv
en
ess
o
f
R
G
B
-
b
ased
p
ed
estrian
d
etec
tio
n
u
n
d
e
r
lo
w
-
lig
h
t
an
d
ad
v
er
s
e
wea
th
er
co
n
d
itio
n
s
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
was
ev
alu
ated
o
n
th
e
"
s
ee
in
g
th
r
o
u
g
h
f
o
g
"
d
ataset,
th
u
s
,
co
n
f
ir
m
in
g
its
im
p
r
o
v
e
d
p
e
r
f
o
r
m
an
ce
in
c
h
allen
g
in
g
v
is
u
al
en
v
ir
o
n
m
en
ts
.
T
h
e
m
ajo
r
ity
o
f
r
esear
ch
ad
d
r
ess
es
p
ed
estrian
d
etec
tio
n
u
n
d
er
eith
er
lo
w
-
lig
h
t
co
n
d
itio
n
s
o
r
ad
v
er
s
e
wea
th
er
,
wh
ile
f
ew
s
tu
d
ies
co
n
cu
r
r
e
n
tly
co
n
s
id
er
b
o
t
h
ch
all
en
g
es.
R
ec
en
t
im
p
r
o
v
em
en
ts
i
n
co
m
p
u
ter
v
is
io
n
h
av
e
s
h
o
wn
th
at
ad
d
in
g
atten
tio
n
m
ec
h
an
is
m
s
ca
n
en
h
an
ce
f
ea
tu
r
e
r
ep
r
esen
tatio
n
b
y
en
ab
lin
g
m
o
d
els
to
f
o
cu
s
o
n
th
e
m
o
s
t
in
f
o
r
m
ati
v
e
r
eg
io
n
s
with
in
a
n
im
ag
e
.
D
esp
ite
th
ese
ad
v
an
ce
m
en
ts
,
th
er
e
r
em
ain
s
lim
ited
r
esear
ch
s
p
ec
if
ically
ex
am
in
i
n
g
th
e
ef
f
icac
y
o
f
atten
tio
n
m
o
d
u
les
f
o
r
p
ed
estrian
d
etec
t
io
n
in
en
v
i
r
o
n
m
e
n
ts
ch
ar
ac
ter
ized
b
y
lo
w
v
is
ib
ilit
y
an
d
p
o
o
r
wea
th
er
co
n
d
i
tio
n
s
.
Mo
s
t
s
tu
d
ies
em
p
lo
y
in
g
atten
tio
n
-
b
ased
ap
p
r
o
ac
h
es
h
av
e
p
r
im
ar
ily
c
o
n
ce
n
tr
ated
o
n
im
p
r
o
v
in
g
d
e
tectio
n
p
er
f
o
r
m
a
n
ce
in
d
e
n
s
ely
cr
o
wd
e
d
s
ce
n
es:
J
u
b
air
an
d
Naf
ea
[
2
0
]
p
r
o
p
o
s
ed
co
n
v
o
lu
tio
n
al
b
lo
c
k
atte
n
tio
n
m
o
d
u
les
(
C
B
AM
)
-
YO
L
Ov
8
,
i
n
co
r
p
o
r
atin
g
th
r
ee
C
B
AM
in
to
YOL
O
v
8
’
b
ac
k
b
o
n
e
.
E
x
p
e
r
im
en
tal
ev
alu
a
tio
n
co
n
d
u
cted
o
n
a
co
m
p
o
s
ite
d
ataset
co
m
p
r
is
in
g
1
,
5
2
0
im
ag
es
f
r
o
m
I
NR
I
A
an
d
E
T
H
d
atasets
d
em
o
n
s
tr
ated
th
at
th
e
in
teg
r
atio
n
o
f
C
B
AM
lead
s
to
im
p
o
r
ta
n
t
im
p
r
o
v
em
e
n
ts
in
b
o
th
r
ec
all
a
n
d
th
e
m
ea
n
av
e
r
ag
e
p
r
ec
is
io
n
(
m
AP)
ac
r
o
s
s
th
e
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
I
o
U
)
th
r
esh
o
ld
r
an
g
e
o
f
0
.
5
to
0
.
9
5
(
m
AP_
0
.
5
:0
.
9
5
)
.
W
an
g
et
a
l.
[
2
1
]
in
tr
o
d
u
ce
d
a
C
C
W
-
YOL
O
alg
o
r
ith
m
,
wh
ich
in
co
r
p
o
r
ated
a
lig
h
tweig
h
t
c
o
n
v
o
lu
ti
o
n
al
lay
er
u
tili
zin
g
Gh
o
s
tC
o
n
v
,
alo
n
g
s
id
e
an
im
p
r
o
v
e
d
C
2
f
m
o
d
u
le
aim
ed
at
b
o
o
s
tin
g
th
e
d
etec
tio
n
ca
p
ab
ilit
ies
o
f
th
e
n
etwo
r
k
.
Fu
r
th
er
m
o
r
e
,
th
e
m
o
d
el
in
teg
r
ated
th
e
co
o
r
d
in
ate
atten
tio
n
m
ec
h
an
is
m
to
m
o
r
e
ef
f
ec
tiv
ely
ca
p
tu
r
e
c
r
itical
tar
g
et
f
ea
tu
r
es,
t
h
er
eb
y
in
c
r
ea
s
in
g
th
e
m
o
d
el’
s
ad
ap
tab
ilit
y
ac
r
o
s
s
d
iv
er
s
e
d
etec
tio
n
s
ce
n
ar
io
s
an
d
f
u
r
th
e
r
en
h
an
cin
g
o
v
er
all
ac
cu
r
ac
y
.
Z
an
g
et
a
l.
[
2
2
]
in
tr
o
d
u
ce
d
m
u
lti
-
r
ec
ep
tiv
e
f
ield
an
d
atten
tio
n
m
ec
h
an
is
m
f
o
r
m
u
ltis
p
ec
tr
al
p
e
d
estrian
d
etec
tio
n
(
MA
PD
)
wh
ich
lev
er
ag
es
b
o
th
m
u
lti
-
r
ec
ep
tiv
e
f
ield
p
r
o
ce
s
s
in
g
an
d
atten
tio
n
m
ec
h
an
is
m
s
.
T
h
is
d
esig
n
ef
f
ec
tiv
ely
in
teg
r
ates
a
m
u
lti
-
r
ec
ep
tiv
e
f
ield
m
o
d
u
le
with
th
e
C
B
AM
to
f
o
r
m
a
co
m
b
in
ed
m
u
lti
-
r
ec
ep
tiv
e
f
ield
an
d
atten
tio
n
(
MA
)
m
o
d
u
le.
No
ta
b
ly
,
th
is
m
o
d
u
le
was
b
u
ilt
to
b
e
m
o
d
u
lar
,
f
ac
ilit
atin
g
ea
s
y
in
teg
r
atio
n
with
d
iv
er
s
e
p
r
e
-
e
x
is
tin
g
n
etwo
r
k
a
r
ch
itectu
r
es.
I
n
r
esp
o
n
s
e
to
lim
itatio
n
s
in
p
r
io
r
p
ed
estrian
d
etec
tio
n
r
esea
r
ch
u
n
d
er
ch
alle
n
g
in
g
v
is
u
al
c
o
n
d
itio
n
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
an
atten
ti
o
n
-
au
g
m
en
ted
YOL
Ov
5
s
ar
c
h
itectu
r
e
tr
ain
ed
o
n
a
n
en
h
a
n
c
ed
co
m
m
o
n
o
b
jects
in
co
n
tex
t
(
C
OC
O
)
d
ataset
[
2
3
]
,
wh
ic
h
is
au
g
m
e
n
ted
wit
h
s
y
n
th
etic
tr
an
s
f
o
r
m
atio
n
s
s
im
u
latin
g
d
ar
k
n
ess
,
n
o
is
e,
an
d
b
lu
r
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
is
to
ev
alu
ate
wh
eth
er
th
e
in
co
r
p
o
r
atio
n
o
f
a
n
o
v
e
l
atten
tio
n
m
o
d
u
le,
s
p
ec
if
ically
,
a
weig
h
ted
f
u
s
io
n
ef
f
icien
t
ch
an
n
el
atten
tio
n
(
W
F
-
E
C
A)
m
ec
h
an
is
m
th
at
ad
ap
tiv
ely
co
m
b
in
es
m
ax
an
d
av
er
ag
e
p
o
o
lin
g
,
ca
n
im
p
r
o
v
e
d
etec
tio
n
r
o
b
u
s
tn
e
s
s
an
d
ac
cu
r
ac
y
f
o
r
p
ed
estria
n
d
etec
tio
n
u
n
d
er
v
is
u
ally
d
eg
r
a
d
ed
co
n
d
itio
n
s
,
r
elativ
e
to
th
e
b
aselin
e
YOL
Ov
5
s
m
o
d
el.
T
h
e
m
ain
c
o
n
tr
i
b
u
tio
n
s
o
f
th
is
wo
r
k
ar
e:
i
)
th
e
j
u
s
tific
atio
n
o
f
YOL
Ov
5
s
as
a
b
aselin
e
th
r
o
u
g
h
co
m
p
ar
ativ
e
e
v
alu
atio
n
with
s
tate
-
of
-
th
e
-
ar
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
p
e
d
estr
ia
n
d
etec
tio
n
in
a
d
ve
r
s
e
co
n
d
itio
n
s
u
s
in
g
YOLOv5
s
w
ith
…
(
Ou
ma
yma
R
a
ch
id
i
)
3301
d
etec
to
r
s
;
ii
)
th
e
d
esig
n
o
f
W
F
-
E
C
A
to
d
y
n
am
ically
b
alan
c
e
p
o
o
lin
g
f
ea
tu
r
es
f
o
r
im
p
r
o
v
ed
r
o
b
u
s
tn
ess
;
an
d
iii
)
a
co
m
p
r
e
h
en
s
iv
e
ass
ess
m
en
t
o
f
W
F
-
E
C
A
ag
ain
s
t
ex
is
tin
g
atten
tio
n
m
o
d
u
les,
d
em
o
n
s
tr
atin
g
its
ef
f
ec
tiv
en
ess
in
en
h
a
n
cin
g
p
e
d
estrian
d
etec
tio
n
u
n
d
er
d
eg
r
a
d
ed
v
is
u
al
co
n
d
itio
n
s
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
p
ap
e
r
is
a
s
:
s
ec
ti
o
n
2
d
etails
th
e
m
eth
o
d
o
lo
g
y
,
a
n
d
p
r
o
v
id
es
an
o
v
e
r
v
iew
o
f
th
e
YOL
Ov
5
s
ar
ch
itectu
r
e
an
d
a
d
escr
ip
tio
n
o
f
t
h
e
p
r
o
p
o
s
ed
atten
tio
n
m
o
d
u
le.
Sectio
n
3
p
r
esen
ts
th
e
ex
p
er
im
en
tal
r
esu
lts
an
d
d
is
cu
s
s
io
n
s
.
S
ec
ti
o
n
4
co
n
clu
d
es th
e
s
tu
d
y
.
2.
M
E
T
H
O
D
T
o
im
p
r
o
v
e
p
ed
estrian
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
u
n
d
er
ad
v
er
s
e
v
is
u
al
co
n
d
itio
n
s
,
th
e
YOL
Ov
5
s
ar
ch
itectu
r
e
was
tr
ain
ed
o
n
a
C
OC
O
d
ataset
en
h
an
ce
d
wit
h
s
y
n
th
etic
tr
an
s
f
o
r
m
atio
n
s
s
i
m
u
latin
g
d
ar
k
n
ess
,
n
o
is
e,
an
d
b
lu
r
,
s
p
ec
if
ically
tar
g
etin
g
th
e
p
er
s
o
n
class
.
T
h
e
s
elec
tio
n
o
f
YOL
Ov
5
s
is
m
o
tiv
ated
b
y
its
ef
f
ec
tiv
e
co
m
p
r
o
m
is
e
b
etwe
en
d
etec
tio
n
ac
cu
r
ac
y
a
n
d
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
o
ju
s
tify
th
is
ch
o
ice,
tr
an
s
f
er
lear
n
in
g
ex
p
er
im
en
ts
wer
e
co
n
d
u
cted
u
s
in
g
YOL
Ov
5
s
alo
n
g
s
id
e
o
t
h
er
s
tate
-
of
-
th
e
-
ar
t
d
etec
to
r
s
,
in
clu
d
in
g
YOL
Ov
8
s
[
2
4
]
,
s
in
g
le
s
h
o
t
d
etec
to
r
(
SSD
)
[
2
5
]
,
a
n
d
E
f
f
icien
tDet
[
2
6
]
,
ev
al
u
ated
o
n
th
e
au
g
m
en
ted
d
ataset.
Star
tin
g
f
r
o
m
th
is
b
as
elin
e,
we
p
r
o
p
o
s
e
a
n
o
v
el,
lig
h
tweig
h
t a
tten
tio
n
m
o
d
u
le
in
s
p
i
r
ed
b
y
th
e
ef
f
icien
t
ch
an
n
el
atten
tio
n
(
E
C
A)
m
ec
h
an
is
m
.
T
h
e
m
o
d
u
le’
s
p
er
f
o
r
m
an
ce
is
ev
alu
ated
th
r
o
u
g
h
c
o
m
p
ar
ativ
e
an
aly
s
is
ag
ain
s
t
th
e
YOL
Ov
5
s
b
aselin
e
an
d
s
ev
er
al
estab
lis
h
ed
atten
tio
n
m
o
d
u
les
to
d
eter
m
in
e
its
im
p
ac
t
o
n
d
etec
tio
n
ac
cu
r
ac
y
in
c
h
allen
g
in
g
en
v
ir
o
n
m
en
ts
.
2
.
1
.
Da
t
a
s
et
A
s
p
ec
ialized
d
ataset
f
o
r
p
ed
estrian
d
etec
tio
n
was
cr
ea
ted
u
s
in
g
th
e
C
OC
O
d
ata
s
et.
Fro
m
th
is
d
ataset,
ar
o
u
n
d
4
0
,
0
0
0
im
ag
es
wer
e
ch
o
s
en
f
o
r
tr
ai
n
in
g
an
d
2
,
7
5
8
f
o
r
v
alid
atio
n
,
f
o
cu
s
in
g
o
n
im
ag
es
co
n
tain
in
g
th
e
p
e
r
s
o
n
ca
teg
o
r
y
.
T
o
s
tr
en
g
t
h
en
th
e
d
ete
cto
r
’
s
p
er
f
o
r
m
an
ce
an
d
h
an
d
le
co
m
m
o
n
is
s
u
es
en
co
u
n
ter
e
d
in
p
e
d
estrian
d
etec
tio
n
,
a
wid
e
r
an
g
e
o
f
d
ata
au
g
m
e
n
tatio
n
s
tr
ateg
ies
wer
e
ap
p
lied
.
R
o
b
o
f
lo
w
was
u
s
ed
b
o
th
to
c
o
n
s
tr
u
ct
an
d
a
u
g
m
en
t
t
h
e
d
ata
s
et.
T
h
e
au
g
m
en
tatio
n
s
p
e
r
f
o
r
m
ed
o
n
th
e
o
r
ig
in
al
im
ag
es in
clu
d
ed
:
−
B
r
ig
h
tn
ess
ad
ju
s
tm
en
t:
b
r
i
g
h
t
n
ess
was
r
ed
u
ce
d
b
y
2
0
%
to
i
m
p
r
o
v
e
th
e
m
o
d
el’
s
ab
ilit
y
to
o
p
e
r
ate
u
n
d
er
lo
w
-
lig
h
t c
o
n
d
itio
n
s
,
s
u
ch
as n
ig
h
ttime
s
ce
n
ar
io
s
.
−
B
lu
r
r
in
g
:
a
Gau
s
s
ian
b
l
u
r
wit
h
a
2
.
5
p
ix
el
was
ad
d
ed
to
i
m
p
r
o
v
e
th
e
m
o
d
el’
s
a
d
ap
tab
i
lity
to
ca
m
er
a
f
o
cu
s
is
s
u
es:
an
im
p
o
r
tan
t
c
o
n
s
id
er
atio
n
in
ADAS
ap
p
lica
tio
n
s
wh
er
e
b
o
t
h
th
e
ca
m
e
r
a
an
d
s
u
b
jects
(
e.
g
.
,
p
ed
estrian
s
o
r
c
y
clis
ts
)
m
ay
b
e
in
m
o
tio
n
.
−
No
is
e
ad
d
itio
n
:
r
an
d
o
m
n
o
is
e
was
in
tr
o
d
u
ce
d
to
u
p
to
2
%
o
f
th
e
p
ix
els,
s
im
u
latin
g
s
en
s
o
r
n
o
is
e.
T
h
is
en
co
u
r
a
g
es
th
e
m
o
d
el
to
g
en
e
r
alize
b
etter
b
y
lear
n
in
g
to
id
en
tify
p
ed
estrian
s
ev
e
n
wh
e
n
p
ar
ts
o
f
th
e
m
ar
e
o
b
s
cu
r
e
d
.
Fo
llo
win
g
th
ese
au
g
m
en
tatio
n
s
,
th
e
f
in
al
d
ataset
c
o
m
p
r
i
s
ed
1
1
8
,
4
2
5
tr
ain
i
n
g
im
a
g
es,
2
,
7
5
8
v
alid
atio
n
im
ag
es,
an
d
1
8
6
im
a
g
es r
eser
v
ed
f
o
r
test
in
g
.
T
h
e
co
m
p
lete
d
ataset
is
h
o
s
ted
o
n
R
o
b
o
f
l
o
w.
2
.
2
.
YO
L
O
v
5
s
ba
s
eline
a
rc
h
it
ec
t
ure
YOL
Ov
5
s
[
2
7
]
is
a
p
o
p
u
lar
ch
o
ice
f
o
r
o
b
ject
d
etec
tio
n
t
ask
s
b
ec
au
s
e
it
s
tr
ik
es
a
g
o
o
d
b
alan
ce
b
etwe
en
ac
cu
r
ac
y
an
d
co
m
p
u
tatio
n
al
s
p
ee
d
.
I
ts
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
th
r
ee
m
ain
p
ar
ts
:
th
e
b
ac
k
b
o
n
e
,
th
e
n
ec
k
,
an
d
th
e
h
ea
d
.
T
h
e
b
ac
k
b
o
n
e
u
s
es
C
3
an
d
co
n
v
o
l
u
tio
n
a
l
m
o
d
u
les,
ap
p
ly
in
g
th
e
cr
o
s
s
-
s
tag
e
p
ar
tial
(
C
SP
)
tech
n
iq
u
e
t
o
im
p
r
o
v
e
g
r
ad
ie
n
t
f
lo
w
an
d
en
a
b
le
ef
f
ec
tiv
e
r
eu
s
e
o
f
f
ea
tu
r
es
wh
ile
m
i
n
im
izin
g
r
e
d
u
n
d
an
t
ca
lcu
latio
n
s
.
T
h
is
s
etu
p
en
s
u
r
es
ef
f
icien
t
f
ea
tu
r
e
ex
tr
ac
tio
n
,
wh
ich
is
cr
u
cial
f
o
r
p
r
ec
is
e
d
etec
tio
n
.
T
h
e
n
ec
k
in
clu
d
es
a
s
p
atial
p
y
r
am
id
p
o
o
lin
g
f
ast
(
SP
PF
)
lay
er
alo
n
g
with
ad
d
itio
n
al
co
n
v
o
lu
tio
n
lay
er
s
,
wh
ich
h
elp
s
co
m
b
in
e
f
ea
tu
r
es
at
d
if
f
er
e
n
t
s
ca
les.
T
h
is
n
o
t
o
n
ly
in
cr
ea
s
es
th
e
r
ec
ep
tiv
e
f
ield
b
u
t
also
b
o
o
s
ts
th
e
m
o
d
el’
s
ab
ilit
y
to
u
n
d
er
s
tan
d
co
n
tex
t
ac
r
o
s
s
v
ar
io
u
s
r
eso
lu
tio
n
s
.
Fin
ally
,
th
e
h
ea
d
c
o
m
p
o
n
en
t
is
r
esp
o
n
s
ib
le
f
o
r
p
r
ed
ictin
g
b
o
u
n
d
in
g
b
o
x
c
o
o
r
d
in
ates,
o
b
jectn
ess
s
co
r
es,
a
n
d
class
p
r
o
b
ab
ilit
ies,
th
u
s
p
r
o
d
u
cin
g
th
e
f
in
al
d
etec
tio
n
o
u
tp
u
ts
.
An
o
v
er
v
ie
w
o
f
th
e
YOL
Ov
5
s
ar
ch
itectu
r
e
is
illu
s
tr
ated
in
F
ig
u
r
e
1
.
Alth
o
u
g
h
alter
n
ativ
e
m
o
d
els
s
u
ch
as
YOL
Ov
8
s
,
SS
D,
an
d
E
f
f
icien
tDet
d
em
o
n
s
tr
ate
im
p
r
o
v
em
e
n
ts
in
ce
r
tain
p
er
f
o
r
m
a
n
ce
m
etr
ics,
o
u
r
co
m
p
a
r
ativ
e
an
aly
s
is
(
s
e
ctio
n
3
)
co
n
f
ir
m
s
th
at
YOL
Ov
5
s
,
wh
en
r
etr
ain
ed
o
n
o
u
r
au
g
m
en
ted
d
ataset,
r
em
ain
s
th
e
m
o
s
t su
itab
le
ch
o
i
ce
f
o
r
r
ea
l
-
tim
e
p
ed
estrian
d
etec
tio
n
.
2
.
3
.
O
v
er
v
iew
o
f
ex
is
t
ing
a
t
t
ent
io
n m
o
du
les
Atten
tio
n
m
ec
h
an
is
m
s
h
a
v
e
b
ee
n
in
teg
r
ated
in
to
n
eu
r
al
n
etwo
r
k
ar
c
h
itectu
r
es
to
im
p
r
o
v
e
f
ea
tu
r
e
r
ep
r
esen
tatio
n
b
y
d
ir
ec
tin
g
f
o
cu
s
to
war
d
th
e
m
o
s
t
in
f
o
r
m
ativ
e
elem
e
n
ts
,
th
er
eb
y
im
p
r
o
v
in
g
o
v
e
r
all
p
er
f
o
r
m
an
ce
in
v
ar
io
u
s
v
is
io
n
task
s
.
A
r
an
g
e
o
f
atten
tio
n
m
o
d
u
les
h
as
b
ee
n
p
r
o
p
o
s
ed
in
th
e
liter
atu
r
e,
ea
ch
em
p
lo
y
in
g
d
if
f
er
en
t
s
tr
ateg
ies
to
r
ef
in
e
f
ea
tu
r
e
m
a
p
s
.
T
h
e
s
q
u
ee
ze
-
a
n
d
-
e
x
citatio
n
(
SE)
m
o
d
u
le
[
2
8
]
in
tr
o
d
u
ce
s
ch
a
n
n
el
-
wis
e
atten
tio
n
b
y
f
ir
s
t
ag
g
r
eg
atin
g
g
l
o
b
al
s
p
atial
in
f
o
r
m
atio
n
th
r
o
u
g
h
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
to
p
r
o
d
u
ce
a
ch
a
n
n
el
d
escr
ip
to
r
,
wh
ich
is
th
en
p
ass
ed
th
r
o
u
g
h
a
s
m
all
f
u
lly
co
n
n
ec
ted
n
etwo
r
k
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
9
9
-
3
3
0
8
3302
g
en
er
ate
ch
a
n
n
el
-
wis
e
weig
h
ts
f
o
r
f
ea
t
u
r
e
r
ec
alib
r
atio
n
.
T
h
e
C
B
AM
[
2
9
]
e
x
ten
d
s
th
is
ap
p
r
o
ac
h
b
y
s
eq
u
en
tially
ap
p
ly
i
n
g
b
o
th
ch
an
n
el
an
d
s
p
atial
atten
tio
n
.
I
t
in
teg
r
ates
b
o
t
h
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
an
d
m
a
x
p
o
o
lin
g
to
p
r
o
d
u
ce
ch
a
n
n
el
a
tten
tio
n
,
f
o
llo
we
d
b
y
co
n
v
o
l
u
tio
n
al
o
p
er
atio
n
s
o
n
th
e
a
g
g
r
eg
ated
f
ea
t
u
r
es
to
p
r
o
d
u
ce
a
s
p
atial
atten
tio
n
m
ap
.
T
h
e
E
C
A
m
o
d
u
le
[
3
0
]
o
f
f
er
s
a
m
o
r
e
lig
h
tweig
h
t
alter
n
ativ
e
b
y
av
o
i
d
in
g
d
im
en
s
io
n
ality
r
ed
u
ctio
n
an
d
in
s
tead
ca
p
tu
r
in
g
c
r
o
s
s
-
ch
an
n
el
in
ter
ac
tio
n
s
u
s
in
g
a
f
ast
1
D
co
n
v
o
l
u
tio
n
with
a
ca
r
ef
u
lly
s
elec
ted
k
er
n
el
s
ize.
Desp
ite
its
s
im
p
licity
,
E
C
A
ac
h
iev
es
co
m
p
etitiv
e
p
er
f
o
r
m
a
n
ce
co
m
p
a
r
ed
t
o
SE
an
d
C
B
AM
,
wh
ile
o
f
f
er
in
g
en
h
an
ce
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
,
wh
ich
m
ak
es
it
well
s
u
ite
d
f
o
r
r
ea
l
-
tim
e
an
d
r
eso
u
r
ce
-
lim
ited
ap
p
licatio
n
s
.
Fig
u
r
e
1
.
YOL
Ov
5
s
m
o
d
el
ar
c
h
itectu
r
e
2
.
4
.
I
m
pro
v
ed
YO
L
O
V5
s
mo
del w
it
h
weig
hte
d f
us
io
n e
f
f
icient
cha
nn
el
a
t
t
ent
io
n
A
p
r
im
ar
y
o
b
jectiv
e
o
f
th
is
s
tu
d
y
is
to
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
YOL
Ov
5
s
d
etec
to
r
u
n
d
er
ch
allen
g
in
g
c
o
n
d
itio
n
s
th
r
o
u
g
h
th
e
in
teg
r
atio
n
o
f
a
n
ef
f
ec
ti
v
e
y
et
lig
h
tweig
h
t
atten
tio
n
m
ec
h
an
is
m
.
B
u
ild
in
g
o
n
th
e
co
m
p
u
tatio
n
al
ef
f
icie
n
cy
an
d
ef
f
ec
tiv
en
ess
o
f
th
e
E
C
A
m
o
d
u
le,
we
p
r
o
p
o
s
e
a
n
o
v
el
atten
tio
n
m
ec
h
an
is
m
ter
m
ed
W
F
-
E
C
A.
W
h
ile
th
e
o
r
ig
in
al
E
C
A
m
o
d
u
le
u
tili
ze
s
o
n
ly
g
lo
b
al
av
er
ag
e
p
o
o
lin
g
to
e
n
co
d
e
ch
an
n
el
-
wis
e
in
f
o
r
m
atio
n
,
th
i
s
ap
p
r
o
ac
h
m
a
y
o
v
er
lo
o
k
co
m
p
lem
en
tar
y
f
ea
tu
r
es
th
at
ca
n
b
e
ca
p
tu
r
e
d
th
r
o
u
g
h
m
ax
p
o
o
li
n
g
.
I
t
is
well
estab
li
s
h
ed
th
at
av
er
ag
e
p
o
o
lin
g
a
n
d
m
ax
p
o
o
lin
g
em
p
h
asize
d
if
f
e
r
en
t
ch
ar
ac
ter
is
tics
o
f
f
ea
tu
r
e
m
a
p
s
:
av
er
ag
e
p
o
o
l
in
g
r
ef
lects
th
e
g
en
er
al
f
ea
t
u
r
e
d
is
tr
ib
u
tio
n
,
wh
er
ea
s
m
ax
p
o
o
lin
g
id
en
tifie
s
th
e
m
o
s
t
d
o
m
i
n
an
t r
esp
o
n
s
es.
R
ely
in
g
e
x
clu
s
iv
ely
o
n
o
n
e
p
o
o
li
n
g
s
tr
ateg
y
ca
n
co
n
s
tr
ain
th
e
m
o
d
u
le’
s
ca
p
ac
ity
to
ad
ap
tiv
ely
f
o
cu
s
o
n
t
h
e
m
o
s
t
i
n
f
o
r
m
ativ
e
ch
an
n
el
f
ea
tu
r
es.
T
o
ad
d
r
ess
th
is
lim
itatio
n
,
th
e
p
r
o
p
o
s
ed
W
F
-
E
C
A
m
o
d
u
le
ad
a
p
tiv
ely
in
teg
r
ates
b
o
th
av
er
a
g
e
an
d
m
ax
p
o
o
lin
g
o
p
er
atio
n
s
,
en
ab
lin
g
th
e
n
e
two
r
k
to
lear
n
a
n
o
p
tim
al
b
alan
ce
b
etwe
en
th
e
m
.
T
h
is
f
u
s
io
n
e
n
r
ich
es
ch
an
n
el
-
wis
e
co
n
tex
t
m
o
d
elin
g
wi
th
o
u
t
co
m
p
r
o
m
is
in
g
th
e
lig
h
tweig
h
t
n
atu
r
e
o
f
th
e
E
C
A
s
tr
u
ctu
r
e.
T
h
e
W
F
-
E
C
A
m
o
d
u
le
is
in
co
r
p
o
r
ated
at
th
e
en
d
o
f
th
e
b
ac
k
b
o
n
e
s
tag
e
o
f
th
e
YOL
Ov
5
s
ar
c
h
itectu
r
e,
as
illu
s
tr
ated
in
Fig
u
r
e
2
,
t
o
en
h
an
ce
f
ea
tu
r
e
ex
tr
ac
tio
n
ef
f
icien
c
y
in
a
co
m
p
u
tatio
n
ally
ec
o
n
o
m
ical
m
an
n
er
.
L
et
X
∈
.
.
r
ep
r
esen
t
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
,
wh
er
e
th
e
d
im
e
n
s
io
n
s
C
,
H,
an
d
W
r
ef
er
to
th
e
n
u
m
b
er
o
f
ch
a
n
n
els,
h
eig
h
t,
a
n
d
wid
th
,
r
esp
ec
tiv
ely
.
T
o
en
h
an
ce
ch
an
n
el
-
wis
e
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
wh
ile
m
ain
tain
in
g
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
,
th
e
p
r
o
p
o
s
ed
atten
tio
n
m
ec
h
a
n
is
m
f
ir
s
t
p
er
f
o
r
m
s
b
o
th
g
lo
b
al
av
e
r
ag
e
p
o
o
lin
g
an
d
g
l
o
b
al
m
ax
p
o
o
li
n
g
ac
r
o
s
s
th
e
s
p
atial
d
im
en
s
io
n
s
.
T
h
ese
o
p
er
atio
n
s
y
ield
t
wo
1
D
d
escr
ip
to
r
s
,
an
d
∈
wh
er
e
ea
ch
elem
e
n
t is co
m
p
u
ted
as
(
1
)
an
d
(
2
)
.
(
)
=
1
.
∑
∑
(
,
,
)
=
1
=
1
(
1
)
(
)
=
,
(
,
,
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
p
e
d
estr
ia
n
d
etec
tio
n
in
a
d
ve
r
s
e
co
n
d
itio
n
s
u
s
in
g
YOLOv5
s
w
ith
…
(
Ou
ma
yma
R
a
ch
id
i
)
3303
T
h
e
two
v
ec
to
r
s
ar
e
co
n
ca
te
n
a
ted
to
f
o
r
m
a
u
n
if
ied
d
escr
ip
t
o
r
f
∈
2
.
T
o
ad
a
p
tiv
ely
weig
h
th
e
r
el
ativ
e
im
p
o
r
tan
ce
o
f
av
er
a
g
e
an
d
m
ax
p
o
o
led
f
ea
tu
r
es,
a
lear
n
ab
le
p
ar
am
eter
α
∈
[
0
,
1
]
is
in
tr
o
d
u
ce
d
.
T
h
is
weig
h
tin
g
v
ec
to
r
is
g
en
er
a
ted
b
y
ap
p
l
y
in
g
a
p
a
r
am
eter
iz
ed
f
u
n
ctio
n
σ
ty
p
ically
im
p
lem
en
ted
as a
s
ig
m
o
id
to
th
e
co
n
ca
ten
ate
d
d
esc
r
ip
to
r
as sh
o
wn
in
(
3
)
.
=
(
.
)
(
3
)
W
h
er
e
∈
.
2
is
a
lear
n
ab
le
p
ar
am
eter
.
T
h
e
ad
ap
tiv
e
atten
tio
n
v
ec
to
r
is
th
e
n
c
o
m
p
u
te
d
as
a
co
n
v
ex
co
m
b
in
atio
n
o
f
th
e
two
p
o
o
led
v
ec
to
r
s
in
(
4
)
.
=
ʘ
+
(
1
−
)
ʘ
(
4
)
T
o
m
o
d
el
lo
ca
l
ch
an
n
el
d
e
p
en
d
en
cies
with
o
u
t
in
tr
o
d
u
cin
g
s
ig
n
if
ican
t
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
,
a
lig
h
tweig
h
t
o
n
e
-
d
im
en
s
io
n
al
c
o
n
v
o
lu
ti
o
n
is
s
u
b
s
eq
u
en
tly
ap
p
lied
to
th
e
ad
ap
tiv
e
v
ec
to
r
.
T
h
is
o
p
er
atio
n
ca
p
tu
r
es
in
ter
ac
tio
n
s
am
o
n
g
n
eig
h
b
o
r
in
g
c
h
an
n
els
ef
f
icie
n
tly
,
r
esu
ltin
g
in
a
n
en
h
a
n
c
ed
atten
tio
n
v
ec
to
r
∈
.
Fin
ally
,
th
is
v
ec
to
r
is
em
p
lo
y
ed
to
r
ec
alib
r
ate
th
e
o
r
ig
in
al
f
ea
tu
r
e
m
a
p
v
ia
c
h
an
n
el
-
wis
e
m
u
ltip
licatio
n
,
f
o
r
m
u
lated
as
(
5
)
.
′
(
,
,
)
=
(
,
,
)
,
∀
∈
{
1
,
…
,
}
,
∈
{
1
,
…
,
}
,
∈
{
1
,
…
,
}
(
5
)
Fig
u
r
e
2
.
T
h
e
im
p
r
o
v
ed
YOL
Ov
5
s
with
W
F
-
E
C
A
ar
ch
itect
u
r
e
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
itially
,
th
e
YOL
Ov
5
s
m
o
d
el
was
f
in
e
-
tu
n
ed
u
s
in
g
o
u
r
au
g
m
en
ted
d
ataset.
T
h
e
m
o
d
el
tr
ain
in
g
was
p
er
f
o
r
m
ed
u
s
in
g
th
e
Go
o
g
le
C
o
lab
p
latf
o
r
m
.
Su
b
s
eq
u
en
t
ly
,
we
ca
r
r
ied
o
u
t
a
co
m
p
ar
ativ
e
an
aly
s
is
with
s
ev
er
al
well
-
estab
lis
h
ed
o
b
je
ct
d
etec
tio
n
m
o
d
els.
B
ased
o
n
th
e
o
b
s
er
v
e
d
p
er
f
o
r
m
an
ce
u
n
d
e
r
ch
allen
g
in
g
co
n
d
itio
n
s
,
a
n
o
v
el
atten
tio
n
m
o
d
u
le
was in
co
r
p
o
r
ate
d
to
f
u
r
th
er
en
h
an
ce
th
e
m
o
d
el'
s
d
etec
tio
n
ca
p
ab
ilit
ies.
3
.
1
.
E
x
perim
ent
a
l
s
et
up
T
h
e
s
elec
ted
m
o
d
els:
YOL
Ov
5
s
,
YOL
Ov
8
s
,
SS
D,
an
d
E
f
f
icien
tDet,
wer
e
r
etr
ain
ed
u
s
in
g
th
e
Go
o
g
le
C
o
lab
p
latf
o
r
m
.
Prio
r
to
tr
ain
i
n
g
,
th
e
a
u
g
m
en
te
d
d
ataset
was
co
n
v
er
ted
in
t
o
th
e
ap
p
r
o
p
r
ia
te
f
o
r
m
at
f
o
r
ea
ch
m
o
d
el
u
s
in
g
th
e
R
o
b
o
f
lo
w
p
la
tf
o
r
m
.
T
h
e
r
etr
ain
in
g
p
r
o
ce
s
s
was
p
er
f
o
r
m
ed
f
o
r
3
0
0
ep
o
ch
s
u
s
in
g
a
b
atch
s
ize
o
f
1
6
an
d
a
n
in
p
u
t im
ag
e
r
eso
lu
tio
n
o
f
6
4
0
×6
4
0
p
ix
els.
T
h
e
r
etr
ain
ed
m
o
d
els
wer
e
a
s
s
es
s
ed
u
s
in
g
th
e
v
alid
atio
n
s
u
b
s
et
o
f
th
e
d
ataset.
Fo
r
co
m
p
ar
ativ
e
an
aly
s
is
,
we
em
p
lo
y
e
d
two
wi
d
ely
r
ec
o
g
n
ized
ev
alu
atio
n
m
etr
ics:
m
AP
at
I
o
U
th
r
esh
o
ld
0
.
5
(
m
AP_
0
.
5
)
an
d
m
AP
ac
r
o
s
s
m
u
ltip
le
I
o
U
th
r
esh
o
ld
s
r
an
g
in
g
f
r
o
m
0
.
5
to
0
.
9
5
with
a
s
tep
s
ize
o
f
0
.
0
5
(
m
AP_
0
.
5
:0
.
9
5
)
.
T
h
e
m
AP_
0
.
5
m
ea
s
u
r
es
th
e
av
er
a
g
e
p
r
ec
is
io
n
wh
e
n
th
e
I
o
U
is
f
ix
ed
at
5
0
%,
wh
ile
m
AP_
0
.
5
:0
.
9
5
o
f
f
e
r
s
a
m
o
r
e
th
o
r
o
u
g
h
e
v
alu
atio
n
b
y
av
e
r
ag
in
g
p
r
ec
is
io
n
o
v
er
a
s
p
ec
tr
u
m
o
f
s
tr
icter
I
o
U
th
r
esh
o
ld
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
9
9
-
3
3
0
8
3304
T
h
e
co
m
p
ar
ativ
e
r
esu
lts
ar
e
illu
s
tr
ated
in
Fig
u
r
e
3
.
As
d
ep
icted
,
b
o
th
YOL
Ov
5
s
an
d
YOL
Ov
8
s
s
ig
n
if
ican
tly
o
u
t
p
er
f
o
r
m
SS
D
an
d
E
f
f
icien
tDet
in
ter
m
s
o
f
d
etec
tio
n
ac
cu
r
ac
y
a
n
d
co
n
v
er
g
en
ce
s
p
ee
d
.
No
tab
ly
,
YOL
Ov
8
s
ac
h
iev
e
s
th
e
h
ig
h
est
ac
cu
r
ac
y
am
o
n
g
all
test
ed
m
o
d
els;
h
o
w
ev
er
,
its
in
cr
ea
s
ed
co
m
p
u
tatio
n
al
co
m
p
lex
ity
a
n
d
in
f
er
en
ce
laten
cy
m
ak
e
it
less
s
u
itab
le
f
o
r
r
ea
l
-
ti
m
e
d
ep
l
o
y
m
en
t
o
n
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
d
ev
ices.
I
n
co
n
tr
ast,
YOL
Ov
5
s
o
f
f
er
s
a
f
av
o
r
ab
le
co
m
p
r
o
m
is
e
b
et
wee
n
ac
cu
r
ac
y
an
d
ef
f
icien
cy
,
m
ak
in
g
it
p
ar
ticu
l
ar
ly
well
-
s
u
ited
f
o
r
r
ea
l
-
tim
e
p
ed
estrian
d
etec
tio
n
task
s
.
T
h
e
f
in
al
m
AP
v
alu
es
ar
e
p
r
esen
ted
in
T
ab
le
1
.
Fu
r
th
er
m
o
r
e
,
T
ab
le
2
p
r
esen
ts
th
e
in
f
er
en
ce
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
els
wh
en
d
e
p
lo
y
ed
o
n
a
R
asp
b
er
r
y
Pi
4
.
T
h
e
r
esu
lts
co
n
f
ir
m
th
at
YOL
Ov
5
s
o
f
f
er
s
a
b
alan
ce
d
c
o
m
p
r
o
m
is
e
b
etwe
en
d
etec
tio
n
ac
cu
r
ac
y
a
n
d
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
ca
p
ab
ilit
ies
.
T
h
is
r
ein
f
o
r
cin
g
its
p
r
ac
ticality
f
o
r
em
b
ed
d
ed
s
y
s
tem
s
an
d
ed
g
e
co
m
p
u
tin
g
s
ce
n
ar
io
s
.
Fig
u
r
e
3
.
C
o
m
p
a
r
is
o
n
b
etwe
e
n
m
o
d
e
r
n
d
etec
to
r
s
o
n
th
e
au
g
m
en
ted
d
ataset
T
ab
le
1
.
m
AP v
alu
es
o
f
th
e
c
h
o
s
en
m
o
d
els
M
o
d
e
l
mA
P
_
0
.
5
mA
P
_
0
.
5
:
0
.
9
5
Y
O
LO
v
5
s
0
.
7
9
0
.
5
9
Y
O
LO
v
8
s
0
.
8
2
0
.
6
2
Ef
f
i
c
i
e
n
t
D
e
t
0
0
.
7
6
0
.
5
7
SSD
0
.
6
1
0
.
4
9
T
ab
le
2
.
Fra
m
es p
e
r
s
ec
o
n
d
(
FP
S)
o
f
m
o
d
els o
n
R
asp
b
er
r
y
P
i 4
M
o
d
e
l
s
Y
O
LO
v
5
s
Y
O
LO
v
8
s
Ef
f
i
c
i
e
n
t
D
e
t
0
SSD
FPS
5
3
.
5
2
1
.
8
3
.
2
.
P
er
f
o
r
m
a
nce
o
f
t
he
pro
po
s
ed
a
t
t
ent
io
n m
o
du
le
T
o
an
aly
ze
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
im
p
r
o
v
ed
YOL
Ov
5
s
m
o
d
el
in
co
r
p
o
r
atin
g
th
e
W
F
-
E
C
A
atten
tio
n
m
ec
h
a
n
is
m
,
th
is
s
tu
d
y
c
o
n
d
u
cted
r
etr
ai
n
in
g
e
x
p
er
im
en
ts
u
s
in
g
th
e
a
u
g
m
en
ted
d
ataset
.
Fo
r
co
m
p
ar
is
o
n
,
ad
d
itio
n
al
r
etr
ain
in
g
was
p
er
f
o
r
m
e
d
o
n
YOL
O
v
5
s
in
teg
r
ated
with
alter
n
ativ
e
atten
tio
n
m
o
d
u
les,
n
am
ely
SE,
E
C
A,
an
d
C
B
A
M.
T
h
e
co
m
p
ar
ativ
e
p
er
f
o
r
m
an
ce
o
u
tco
m
es
ar
e
s
u
m
m
ar
iz
ed
in
Fig
u
r
e
4
.
T
h
e
f
in
d
in
g
s
co
n
f
ir
m
t
h
at
th
e
i
m
p
r
o
v
e
d
YOL
Ov
5
s
m
o
d
el
with
th
e
W
F
-
E
C
A
m
o
d
u
le
ac
h
iev
es
a
n
o
tab
le
en
h
an
ce
m
e
n
t
o
v
er
th
e
b
aselin
e
YOL
Ov
5
s
,
with
an
ap
p
r
o
x
im
ately
5
%
in
cr
ea
s
e
in
m
AP.
Fu
r
th
er
m
o
r
e,
th
e
WF
-
E
C
A
v
ar
ian
t
d
em
o
n
s
tr
ate
d
m
o
r
e
r
a
p
id
c
o
n
v
e
r
g
en
ce
d
u
r
in
g
tr
ain
in
g
a
n
d
c
o
n
s
is
ten
tly
h
ig
h
er
m
AP
v
alu
es
r
elativ
e
to
m
o
d
els
em
p
lo
y
in
g
th
e
o
th
e
r
atten
tio
n
m
o
d
u
les.
T
h
ese
f
in
d
in
g
s
u
n
d
er
s
co
r
e
th
e
e
f
f
icac
y
o
f
th
e
W
F
-
E
C
A
atten
tio
n
m
ec
h
an
is
m
in
en
h
an
cin
g
th
e
d
etec
tio
n
ac
cu
r
ac
y
an
d
tr
ain
in
g
ef
f
icien
cy
o
f
YOL
Ov
5
s
with
in
th
e
co
n
tex
t
o
f
th
e
a
u
g
m
en
te
d
d
ataset.
T
h
e
f
in
al
m
AP v
alu
es
ar
e
p
r
esen
ted
in
T
ab
le
3
.
T
o
en
s
u
r
e
t
h
e
r
o
b
u
s
tn
ess
o
f
th
ese
r
esu
lts
,
we
p
er
f
o
r
m
e
d
m
u
l
tip
le
in
d
ep
en
d
en
t
tr
ain
i
n
g
r
u
n
s
f
o
r
b
o
th
th
e
b
aselin
e
YOL
Ov
5
s
an
d
th
e
W
F
-
E
C
A
v
ar
ian
t
(
3
r
u
n
s
ea
c
h
)
.
T
ab
le
4
r
e
p
o
r
ts
th
e
m
ea
n
±
s
tan
d
ar
d
d
ev
iatio
n
o
f
th
e
m
AP
v
alu
es
ac
r
o
s
s
th
ese
r
u
n
s
.
Statis
tical
s
ig
n
if
ican
ce
was
as
s
es
s
ed
u
s
in
g
p
air
ed
t
-
test
s
,
wi
th
co
r
r
esp
o
n
d
in
g
p
-
v
al
u
es
in
clu
d
ed
to
d
em
o
n
s
tr
ate
th
at
th
e
p
er
f
o
r
m
a
n
ce
im
p
r
o
v
em
e
n
ts
o
f
W
F
-
E
C
A
o
v
er
th
e
b
aselin
e
ar
e
co
n
s
is
ten
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
p
e
d
estr
ia
n
d
etec
tio
n
in
a
d
ve
r
s
e
co
n
d
itio
n
s
u
s
in
g
YOLOv5
s
w
ith
…
(
Ou
ma
yma
R
a
ch
id
i
)
3305
F
i
g
u
r
e
4
.
C
o
m
p
a
r
i
s
o
n
b
e
t
w
e
e
n
d
i
f
f
e
r
e
n
t
a
t
t
e
n
t
i
o
n
m
o
d
u
l
e
s
i
n
c
o
r
p
o
r
a
t
e
d
w
i
t
h
Y
O
L
O
v
5
s
o
n
t
h
e
a
u
g
m
e
n
t
e
d
d
a
t
a
s
e
t
T
ab
le
3
.
m
AP v
alu
es
o
f
th
e
Y
OL
Ov
5
s
with
d
if
f
er
en
t a
tten
ti
o
n
m
o
d
u
les
M
o
d
e
l
mA
P
_
0
.
5
mA
P
_
0
.
5
:
0
.
9
5
Y
O
LO
v
5
s
0
.
7
9
0
.
5
9
Y
O
LO
v
5
s
S
E
0
.
8
0
0
.
6
0
Y
O
LO
v
5
s
C
B
A
M
0
.
8
1
0
.
6
2
Y
O
LO
v
5
s
EC
A
0
.
8
2
0
.
6
3
O
u
r
s
0
.
8
4
0
.
6
4
T
a
b
l
e
4
.
m
A
P
p
e
r
f
o
r
m
a
n
c
e
o
f
Y
O
L
O
v
5
s
a
n
d
Y
O
L
O
v
5
s
w
it
h
W
F
-
E
C
A
(
m
e
a
n
±
s
t
d
)
wi
t
h
s
t
at
i
s
t
i
c
al
s
i
g
n
i
f
i
c
a
n
c
e
M
o
d
e
l
mA
P
_
0
.
5
(
mea
n
±
st
d
)
mA
P
_
0
.
5
:
0
.
9
5
(
m
e
a
n
±
st
d
)
p
-
v
a
l
u
e
v
s YO
LO
v
5
s
Y
O
LO
v
5
s
0
.
7
9
±
0
.
0
2
0
.
5
9
±
0
.
0
2
-
Y
O
LO
v
5
s
W
F
-
EC
A
0
.
8
2
±
0
.
0
3
0
.
6
3
±
0
.
0
3
<
0
.
0
1
T
o
f
u
r
th
e
r
ass
ess
th
e
r
o
b
u
s
tn
e
s
s
an
d
g
en
er
aliza
tio
n
o
f
th
e
p
r
o
p
o
s
ed
W
F
-
E
C
A
m
o
d
u
le,
we
ev
alu
ated
b
o
th
YOL
Ov
5
s
an
d
YOL
Ov
5
s
W
F
-
E
C
A
o
n
Fo
g
g
y
C
ity
s
ca
p
es
an
d
E
x
Dar
k
,
wh
ic
h
p
r
esen
t
ch
allen
g
in
g
co
n
d
itio
n
s
s
u
ch
as
f
o
g
g
y
u
r
b
an
s
ce
n
es
an
d
ex
tr
em
ely
lo
w
-
lig
h
t
en
v
ir
o
n
m
en
ts
.
T
h
e
r
esu
lts
,
s
u
m
m
ar
ized
in
T
ab
le
5
,
s
h
o
w
th
at
wh
ile
d
etec
tio
n
p
er
f
o
r
m
an
ce
n
atu
r
ally
d
ec
r
ea
s
es
u
n
d
er
th
ese
d
if
f
icu
lt
s
ce
n
ar
io
s
,
th
e
W
F
-
E
C
A
v
ar
ian
t c
o
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
s
th
e
b
aselin
e
,
d
em
o
n
s
tr
atin
g
its
ab
ilit
y
to
en
h
a
n
ce
f
e
atu
r
e
ex
tr
ac
tio
n
a
n
d
d
etec
tio
n
ac
cu
r
ac
y
i
n
ad
v
er
s
e
co
n
d
itio
n
s
.
An
o
th
e
r
im
p
o
r
tan
t
asp
ec
t
to
c
o
n
s
id
er
is
th
e
in
f
lu
en
ce
o
f
th
e
p
r
o
p
o
s
ed
atten
tio
n
m
o
d
u
le
o
n
th
e
b
aselin
e
m
o
d
el'
s
co
m
p
u
ta
tio
n
al
ef
f
icien
c
y
,
p
ar
ticu
lar
ly
r
eg
ar
d
in
g
i
n
f
er
en
ce
s
p
ee
d
.
Giv
en
th
at
o
u
r
m
o
d
el
is
b
ased
o
n
th
e
E
C
A
m
ec
h
an
is
m
,
it
r
em
ain
s
lig
h
tweig
h
t.
As
d
em
o
n
s
tr
ated
in
T
ab
le
6
,
th
e
g
ig
a
f
lo
atin
g
p
o
in
t o
p
er
atio
n
s
p
er
s
ec
o
n
d
(
GFL
OPs
)
m
etr
ic
ex
h
ib
its
n
eg
lig
ib
le
ch
an
g
e,
in
d
icatin
g
th
at
th
e
ad
d
itio
n
o
f
th
e
atten
t
io
n
m
o
d
u
le
d
o
es
n
o
t
m
ea
n
in
g
f
u
lly
r
aise
th
e
co
m
p
u
tatio
n
a
l
co
m
p
lex
ity
o
f
th
e
b
aselin
e
m
o
d
el.
T
ab
le
5
.
Per
f
o
r
m
an
ce
o
f
YOL
Ov
5
s
an
d
YOL
Ov
5
s
with
W
F
-
E
C
A
o
n
Fo
g
g
y
C
ity
s
ca
p
es a
n
d
E
x
Dar
k
d
atasets
M
o
d
e
l
F
o
g
g
y
C
i
t
y
s
c
a
p
e
s
Ex
D
a
r
k
mA
P
_
0
.
5
mA
P
_
0
.
5
:
0
.
9
5
mA
P
_
0
.
5
mA
P
_
0
.
5
:
0
.
9
5
Y
O
LO
v
5
s
0
.
7
0
0
.
4
7
0
.
7
3
0
.
5
Y
O
LO
v
5
s
W
F
-
EC
A
0
.
7
7
0
.
5
4
0
.
7
8
0
.
5
4
T
ab
le
6
.
C
o
m
p
a
r
is
o
n
b
etwe
en
YOL
Ov
5
s
an
d
th
e
im
p
r
o
v
e
d
o
n
e
o
n
t
h
e
au
g
m
en
ted
d
ataset
M
o
d
e
l
N
u
mb
e
r
o
f
l
a
y
e
r
s
N
u
mb
e
r
o
f
p
a
r
a
me
t
e
r
s
G
F
LO
P
s
Y
O
LO
v
5
s
1
5
7
7
,
0
1
2
,
8
2
2
1
5
.
8
O
u
r
s
1
6
0
7
,
5
3
7
,
1
1
3
1
5
.
8
T
o
f
u
r
t
h
e
r
e
v
a
l
u
a
t
e
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
i
m
p
r
o
v
e
d
Y
O
L
Ov
5
s
m
o
d
e
l
,
a
q
u
a
l
it
a
t
i
v
e
v
i
s
u
a
l
c
o
m
p
a
r
i
s
o
n
w
a
s
c
o
n
d
u
c
t
e
d
a
g
a
i
n
s
t
t
h
e
b
a
s
e
l
i
n
e
m
o
d
e
l
u
s
i
n
g
a
s
e
t
o
f
n
o
v
e
l
t
e
s
t
i
m
a
g
e
s
r
e
p
r
es
e
n
t
i
n
g
c
h
a
l
le
n
g
i
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
9
9
-
3
3
0
8
3306
e
n
v
i
r
o
n
m
e
n
t
a
l
c
o
n
d
i
ti
o
n
s
,
i
n
c
lu
d
i
n
g
n
i
g
h
t
-
t
i
m
e
,
f
o
g
,
a
n
d
r
a
i
n
s
c
e
n
a
r
i
o
s
.
A
s
d
e
p
i
c
te
d
i
n
F
i
g
u
r
e
5
,
t
h
e
i
m
p
r
o
v
e
d
Y
O
L
O
v
5
s
c
o
n
s
is
t
e
n
t
l
y
e
x
h
i
b
i
ts
s
u
p
e
r
i
o
r
d
et
e
c
ti
o
n
p
e
r
f
o
r
m
a
n
c
e
u
n
d
e
r
a
d
v
e
r
s
e
c
o
n
d
i
ti
o
n
s
.
N
o
t
a
b
l
y
,
w
h
i
l
e
t
h
e
s
t
a
n
d
a
r
d
Y
O
L
O
v
5
s
e
x
p
e
r
i
e
n
ce
s
a
d
e
c
li
n
e
i
n
p
e
r
f
o
r
m
a
n
c
e
in
s
u
c
h
s
c
e
n
a
r
i
o
s
,
m
a
n
i
f
es
t
e
d
p
r
i
m
a
r
i
l
y
t
h
r
o
u
g
h
a
n
i
n
c
r
e
a
s
e
i
n
f
a
ls
e
n
e
g
a
ti
v
e
d
et
e
c
t
i
o
n
s
,
t
h
e
e
n
h
a
n
c
e
d
v
e
r
s
i
o
n
e
f
f
e
c
t
i
v
e
l
y
a
d
d
r
es
s
e
s
t
h
i
s
is
s
u
e
,
a
s
s
h
o
w
n
i
n
F
i
g
u
r
e
5
(
a
)
.
T
h
e
i
m
p
r
o
v
e
d
r
o
b
u
s
t
n
e
s
s
o
f
t
h
e
p
r
o
p
o
s
e
d
m
o
d
el
i
n
c
h
a
l
le
n
g
i
n
g
v
i
s
u
al
e
n
v
i
r
o
n
m
e
n
t
s
u
n
d
e
r
s
c
o
r
e
s
i
ts
p
r
a
c
t
i
c
a
l
a
p
p
li
c
a
b
il
i
t
y
a
n
d
e
f
f
i
ca
c
y
i
n
r
e
a
l
-
w
o
r
l
d
s
e
t
ti
n
g
s
,
as
s
h
o
w
n
i
n
F
i
g
u
r
e
5
(
b
)
.
Fig
u
r
e
5
.
Sid
e
-
by
-
s
id
e
co
m
p
ar
is
o
n
o
f
YOL
Ov
5
s
an
d
th
e
im
p
r
o
v
ed
d
etec
to
r
with
(
a
)
YOL
Ov
5
s
r
esu
lts
an
d
(
b
)
im
p
r
o
v
e
d
m
o
d
el
o
u
tp
u
ts
4.
CO
NCLU
SI
O
N
Ma
in
tain
in
g
s
af
e
d
r
iv
in
g
co
n
d
itio
n
s
in
s
ce
n
ar
io
s
in
v
o
lv
in
g
p
o
o
r
illu
m
in
atio
n
a
n
d
ad
v
er
s
e
wea
th
er
i
s
a
cr
itical
co
n
ce
r
n
in
ADAS
s
y
s
tem
s
.
Ped
estrian
d
etec
tio
n
b
ec
o
m
es
p
ar
ticu
lar
ly
ch
alle
n
g
in
g
u
n
d
er
th
ese
cir
cu
m
s
tan
ce
s
,
with
ev
en
r
o
b
u
s
t
m
o
d
els
s
u
ch
as
YOL
O
v
5
ex
h
ib
itin
g
r
ed
u
ce
d
p
er
f
o
r
m
an
ce
.
T
h
is
p
a
p
er
ex
p
lo
r
ed
th
e
ap
p
licatio
n
o
f
an
atten
tio
n
m
ec
h
an
is
m
to
ad
d
r
e
s
s
th
ese
lim
itatio
n
s
.
An
en
h
a
n
ce
d
v
er
s
io
n
o
f
th
e
YOL
Ov
5
s
m
o
d
el
was
p
r
o
p
o
s
ed
,
in
co
r
p
o
r
atin
g
a
n
o
v
el
W
F
-
E
C
A
m
o
d
u
le.
T
h
e
m
o
d
el
was
r
etr
ain
ed
o
n
an
au
g
m
en
ted
d
ataset
s
p
ec
if
ically
d
esig
n
ed
t
o
s
im
u
late
ch
allen
g
in
g
c
o
n
d
itio
n
s
,
in
clu
d
in
g
lo
w
-
lig
h
t
an
d
b
lu
r
ef
f
ec
ts
.
YOL
Ov
5
s
was
ch
o
s
en
as
th
e
b
aselin
e
as
it
o
f
f
er
s
an
ef
f
icien
t
co
m
p
r
o
m
is
e
b
et
wee
n
ac
cu
r
ac
y
a
n
d
co
m
p
u
tatio
n
al
ef
f
icien
cy
,
as
d
em
o
n
s
tr
ated
th
r
o
u
g
h
c
o
m
p
ar
a
tiv
e
ev
alu
atio
n
with
o
th
er
d
et
ec
tio
n
f
r
am
ewo
r
k
s
.
T
h
e
n
o
v
el
m
o
d
el
ac
h
iev
e
d
a
5
%
in
c
r
ea
s
e
in
m
AP,
wh
ile
m
ain
tain
in
g
th
e
o
r
i
g
in
al
in
f
er
en
ce
s
p
ee
d
,
th
u
s
co
n
f
ir
m
in
g
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
p
r
o
p
o
s
ed
atten
tio
n
-
b
ased
en
h
a
n
ce
m
en
t
in
d
if
f
i
cu
lt
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
.
T
h
e
in
clu
s
io
n
o
f
th
e
W
F
-
E
C
A
m
o
d
u
le
en
ab
led
th
e
m
o
d
el
to
m
o
r
e
ef
f
ec
tiv
ely
ca
p
tu
r
e
cr
itical
f
ea
tu
r
es
in
d
eg
r
ad
e
d
v
is
u
al
in
p
u
ts
.
E
x
ten
s
iv
e
ex
p
er
im
e
n
ts
co
n
f
ir
m
ed
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
th
e
b
aselin
e
a
cr
o
s
s
m
u
ltip
le
ad
v
er
s
e
s
ce
n
ar
io
s
.
T
h
ese
r
esu
lts
em
p
h
asiz
e
th
e
p
o
ten
tial
o
f
atten
tio
n
m
ec
h
an
is
m
s
in
en
h
a
n
cin
g
o
b
ject
d
etec
tio
n
r
eliab
ilit
y
u
n
d
e
r
r
ea
l
-
wo
r
ld
o
p
er
atio
n
a
l c
h
allen
g
es.
ACK
NO
WL
E
DG
M
E
N
T
S
T
h
e
au
th
o
r
s
wis
h
to
th
an
k
t
h
e
Un
iv
er
s
ity
o
f
Mo
u
lay
I
s
m
ail
f
o
r
s
u
p
p
o
r
tin
g
t
h
is
r
esear
ch
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
E
n
h
a
n
ci
n
g
p
e
d
estr
ia
n
d
etec
tio
n
in
a
d
ve
r
s
e
co
n
d
itio
n
s
u
s
in
g
YOLOv5
s
w
ith
…
(
Ou
ma
yma
R
a
ch
id
i
)
3307
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
Ou
m
ay
m
a
R
ac
h
id
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
B
ad
r
B
o
u
o
u
lid
I
d
r
is
s
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
h
af
ik
E
d
-
Dah
m
a
n
i
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
th
e
c
o
r
r
esp
o
n
d
in
g
au
t
h
o
r
,
[
OR
]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
S
.
G
u
l
i
n
o
,
G
.
V
i
c
h
i
,
F
.
C
e
c
c
h
e
t
t
o
,
L.
D
i
Li
l
l
o
,
a
n
d
D
.
V
a
n
g
i
,
“
A
c
o
m
b
i
n
e
d
c
o
mf
o
r
t
a
n
d
safe
t
y
-
b
a
s
e
d
a
p
p
r
o
a
c
h
t
o
a
ssess
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
a
d
v
a
n
c
e
d
d
r
i
v
e
r
a
ssi
st
a
n
c
e
f
u
n
c
t
i
o
n
s
,
”
E
u
ro
p
e
a
n
T
r
a
n
s
p
o
rt
Re
s
e
a
rc
h
R
e
v
i
e
w
,
v
o
l
.
1
7
,
n
o
.
1
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
2
5
4
4
-
024
-
0
0
6
9
6
-
4.
[
2
]
N
.
D
a
l
a
l
a
n
d
B
.
Tr
i
g
g
s
,
“
H
i
st
o
g
r
a
ms
o
f
o
r
i
e
n
t
e
d
g
r
a
d
i
e
n
t
s
f
o
r
h
u
m
a
n
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
0
5
I
EE
E
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
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
C
VPR
2
0
0
5
,
2
0
0
5
,
p
p
.
8
8
6
–
8
9
3
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
0
5
.
1
7
7
.
[
3
]
B
.
S
e
b
a
s
t
i
a
n
a
n
d
P
.
B
.
-
Tz
v
i
,
“
S
u
p
p
o
r
t
v
e
c
t
o
r
ma
c
h
i
n
e
b
a
se
d
r
e
a
l
-
t
i
me
t
e
r
r
a
i
n
e
st
i
ma
t
i
o
n
f
o
r
t
r
a
c
k
e
d
r
o
b
o
t
s
,
”
M
e
c
h
a
t
r
o
n
i
c
s
,
v
o
l
.
6
2
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
c
h
a
t
r
o
n
i
c
s
.
2
0
1
9
.
1
0
2
2
6
0
.
[
4
]
K
.
I
.
N
a
a
n
d
B
.
P
a
r
k
,
“
R
e
a
l
-
t
i
me
3
D
mu
l
t
i
-
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
a
n
d
t
r
a
c
k
i
n
g
u
si
n
g
3
D
Li
D
A
R
p
o
i
n
t
c
l
o
u
d
f
o
r
m
o
b
i
l
e
r
o
b
o
t
,
”
ET
RI
J
o
u
rn
a
l
,
v
o
l
.
4
5
,
n
o
.
5
,
p
p
.
8
3
6
–
8
4
6
,
2
0
2
3
,
d
o
i
:
1
0
.
4
2
1
8
/
e
t
r
i
j
.
2
0
2
3
-
0
1
1
6
.
[
5
]
R
e
n
d
i
a
n
d
D
.
F
i
t
r
i
a
n
a
h
,
“
E
n
h
a
n
c
i
n
g
l
o
w
-
l
i
g
h
t
p
e
d
e
st
r
i
a
n
d
e
t
e
c
t
i
o
n
:
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
n
d
Y
O
LO
v
8
i
n
t
e
g
r
a
t
i
o
n
w
i
t
h
a
u
t
o
m
a
t
e
d
d
a
t
a
s
e
t
,
”
B
u
l
l
e
t
i
n
o
f
El
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
a
n
d
I
n
f
o
r
m
a
t
i
c
s
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
1
9
6
9
–
1
9
8
0
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
5
9
1
/
e
e
i
.
v
1
4
i
3
.
8
9
0
3
.
[
6
]
X
.
X
i
o
n
g
,
J
.
Y
a
n
g
,
Y
.
J
i
a
n
g
,
a
n
d
X
.
H
u
,
“
A
d
u
a
l
-
st
r
e
a
m
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
f
o
r
v
i
si
b
l
e
a
n
d
i
n
f
r
a
r
e
d
i
ma
g
e
f
u
si
o
n
i
n
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
2
4
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
A
u
t
o
n
o
m
o
u
s
T
r
a
n
s
p
o
rt
a
t
i
o
n
,
2
0
2
5
,
p
p
.
2
7
–
34
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
96
-
3
9
7
3
-
1
_
4
.
[
7
]
F
.
C
a
o
e
t
a
l
.
,
“
S
S
C
D
-
Y
O
LO
:
sem
i
-
s
u
p
e
r
v
i
se
d
c
r
o
ss
-
do
m
a
i
n
Y
O
LO
v
8
f
o
r
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
i
n
l
o
w
-
l
i
g
h
t
c
o
n
d
i
t
i
o
n
s,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
3
,
p
p
.
6
1
2
2
5
–
6
1
2
3
6
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
5
.
3
5
5
7
3
8
7
.
[
8
]
M
.
H
u
e
t
a
l
.
,
“
A
n
e
n
h
a
n
c
e
d
f
e
a
t
u
r
e
-
f
u
si
o
n
n
e
t
w
o
r
k
f
o
r
sma
l
l
-
s
c
a
l
e
p
e
d
e
st
r
i
a
n
d
e
t
e
c
t
i
o
n
o
n
e
d
g
e
d
e
v
i
c
e
s,”
S
e
n
s
o
rs
,
v
o
l
.
2
4
,
n
o
.
2
2
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
4
2
2
7
3
0
8
.
[
9
]
M
.
Ju
n
g
a
n
d
J.
C
h
o
,
“
E
n
h
a
n
c
i
n
g
d
e
t
e
c
t
i
o
n
o
f
p
e
d
e
s
t
r
i
a
n
s
i
n
l
o
w
-
l
i
g
h
t
c
o
n
d
i
t
i
o
n
s
b
y
a
c
c
e
n
t
u
a
t
i
n
g
G
a
u
ssi
a
n
–
S
o
b
e
l
e
d
g
e
f
e
a
t
u
r
e
s
f
r
o
m d
e
p
t
h
ma
p
s
,
”
Ap
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
4
,
n
o
.
1
8
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
4
1
8
8
3
2
6
.
[
1
0
]
J.
N
a
t
a
p
r
a
w
i
r
a
,
Y
.
G
u
,
I
.
G
o
n
c
h
a
r
e
n
k
o
,
a
n
d
S
.
K
a
mi
j
o
,
“
P
e
d
e
st
r
i
a
n
d
e
t
e
c
t
i
o
n
o
n
mu
l
t
i
sp
e
c
t
r
a
l
i
m
a
g
e
s
i
n
d
i
f
f
e
r
e
n
t
l
i
g
h
t
i
n
g
c
o
n
d
i
t
i
o
n
s
,
”
i
n
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
n
s
u
m
e
r
El
e
c
t
ro
n
i
c
s
,
2
0
2
1
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
E
5
0
6
8
5
.
2
0
2
1
.
9
4
2
7
6
2
7
.
[
1
1
]
K
.
R
o
s
z
y
k
,
M
.
R
.
N
o
w
i
c
k
i
,
a
n
d
P
.
S
k
r
z
y
p
c
z
y
ń
s
k
i
,
“
A
d
o
p
t
i
n
g
t
h
e
Y
O
LO
v
4
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
l
o
w
-
l
a
t
e
n
c
y
m
u
l
t
i
s
p
e
c
t
r
a
l
p
e
d
e
st
r
i
a
n
d
e
t
e
c
t
i
o
n
i
n
a
u
t
o
n
o
m
o
u
s
d
r
i
v
i
n
g
,
”
S
e
n
so
rs
,
v
o
l
.
2
2
,
n
o
.
3
,
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
2
0
3
1
0
8
2
.
[
1
2
]
J.
H
u
,
Y
.
Zh
a
o
,
a
n
d
X
.
Zh
a
n
g
,
“
A
p
p
l
i
c
a
t
i
o
n
o
f
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
i
n
i
n
f
r
a
r
e
d
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
2
0
I
EE
E
5
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
m
a
g
e
,
Vi
si
o
n
a
n
d
C
o
m
p
u
t
i
n
g
,
I
C
I
VC
2
0
2
0
,
2
0
2
0
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
V
C
5
0
8
5
7
.
2
0
2
0
.
9
1
7
7
4
3
8
.
[
1
3
]
J.
Y
.
Zh
u
,
T
.
P
a
r
k
,
P
.
I
so
l
a
,
a
n
d
A
.
A
.
Ef
r
o
s
,
“
U
n
p
a
i
r
e
d
i
ma
g
e
-
to
-
i
m
a
g
e
t
r
a
n
s
l
a
t
i
o
n
u
s
i
n
g
c
y
c
l
e
-
c
o
n
si
s
t
e
n
t
a
d
v
e
r
sari
a
l
n
e
t
w
o
r
k
s,”
i
n
2
0
1
7
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
(
I
C
C
V)
,
2
0
1
7
,
p
p
.
2
2
4
2
–
2
2
5
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
1
7
.
2
4
4
.
[
1
4
]
J.
R
e
d
m
o
n
a
n
d
A
.
F
a
r
h
a
d
i
,
“
Y
O
LO
v
3
:
a
n
i
n
c
r
e
m
e
n
t
a
l
i
m
p
r
o
v
e
me
n
t
,
”
2
0
1
8
,
a
rX
i
v
:
1
8
0
4
.
0
2
7
6
7
.
[
1
5
]
S
.
R
e
n
,
K
.
H
e
,
R
.
G
i
r
s
h
i
c
k
,
a
n
d
J
.
S
u
n
,
“
F
a
s
t
e
r
R
-
C
N
N
:
t
o
w
a
r
d
s
r
e
a
l
-
t
i
m
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
w
i
t
h
r
e
g
i
o
n
p
r
o
p
o
s
a
l
n
e
t
w
o
r
k
s
,
”
i
n
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
2
0
1
7
,
v
o
l
.
3
9
,
n
o
.
6
,
p
p
.
1
1
3
7
–
1
1
4
9
,
d
o
i
:
1
0
.
1
1
0
9
/
T
P
A
M
I
.
2
0
1
6
.
2
5
7
7
0
3
1
.
[
1
6
]
Y
.
W
a
n
g
,
H
.
K
e
,
a
n
d
H
.
C
a
i
,
“
P
C
-
Y
O
LO
:
e
n
h
a
n
c
i
n
g
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
i
n
a
d
v
e
r
se
w
e
a
t
h
e
r
t
h
r
o
u
g
h
p
h
y
si
c
s
-
a
w
a
r
e
a
n
d
d
y
n
a
m
i
c
n
e
t
w
o
r
k
st
r
u
c
t
u
r
e
s,”
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
o
n
i
c
I
m
a
g
i
n
g
,
v
o
l
.
3
4
,
n
o
.
0
2
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
1
7
/
1
.
j
e
i
.
3
4
.
2
.
0
2
3
0
4
9
.
[
1
7
]
X
.
L
i
u
a
n
d
Y
.
Li
n
,
“
Y
O
LO
-
G
W
:
q
u
i
c
k
l
y
a
n
d
a
c
c
u
r
a
t
e
l
y
d
e
t
e
c
t
i
n
g
p
e
d
e
st
r
i
a
n
s
i
n
a
f
o
g
g
y
t
r
a
f
f
i
c
e
n
v
i
r
o
n
me
n
t
,
”
S
e
n
so
rs
,
v
o
l
.
2
3
,
n
o
.
1
2
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
3
1
2
5
5
3
9
.
[
1
8
]
W
.
E.
M
a
n
o
n
g
g
a
,
R
.
C
.
C
h
e
n
,
a
n
d
X
.
Ji
a
n
g
,
“
E
n
h
a
n
c
i
n
g
r
o
a
d
m
a
r
k
i
n
g
s
i
g
n
d
e
t
e
c
t
i
o
n
i
n
l
o
w
-
l
i
g
h
t
c
o
n
d
i
t
i
o
n
s
w
i
t
h
Y
O
LO
v
7
a
n
d
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
me
n
t
t
e
c
h
n
i
q
u
e
s,”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
1
,
n
o
.
3
,
2
0
2
4
,
d
o
i
:
1
0
.
6
7
0
3
/
I
JA
S
E.
2
0
2
4
0
6
_
2
1
(
3
)
.
0
0
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
2
9
9
-
3
3
0
8
3308
[
1
9
]
M
.
H
n
e
w
a
,
A
.
R
a
h
i
m
p
o
u
r
,
J
.
M
i
l
l
e
r
,
D
.
U
p
a
d
h
y
a
y
,
a
n
d
H
.
R
a
d
h
a
,
“
C
r
o
ss
mo
d
a
l
i
t
y
k
n
o
w
l
e
d
g
e
d
i
s
t
i
l
l
a
t
i
o
n
f
o
r
r
o
b
u
s
t
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
i
n
l
o
w
l
i
g
h
t
a
n
d
a
d
v
e
r
se
w
e
a
t
h
e
r
c
o
n
d
i
t
i
o
n
s
,
”
i
n
I
C
AS
S
P
,
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
A
c
o
u
st
i
c
s,
S
p
e
e
c
h
a
n
d
S
i
g
n
a
l
Pr
o
c
e
ssi
n
g
-
Pr
o
c
e
e
d
i
n
g
s
,
2
0
2
3
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
S
S
P
4
9
3
5
7
.
2
0
2
3
.
1
0
0
9
5
3
5
3
.
[
2
0
]
M
.
A
.
J
u
b
a
i
r
a
n
d
M
.
M
.
N
a
f
e
a
,
“
A
t
t
e
n
t
i
o
n
-
e
n
h
a
n
c
e
d
Y
O
LO
v
8
f
o
r
a
c
c
u
r
a
t
e
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
a
n
d
c
o
u
n
t
e
s
t
i
m
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
S
o
f
t
C
o
m
p
u
t
i
n
g
a
n
d
D
a
t
a
M
i
n
i
n
g
,
v
o
l
.
5
,
n
o
.
2
,
p
p
.
1
7
5
–
1
8
7
,
2
0
2
4
,
d
o
i
:
1
0
.
3
0
8
8
0
/
j
s
c
d
m.
2
0
2
4
.
0
5
.
0
2
.
0
1
3
.
[
2
1
]
Z.
W
a
n
g
,
S
.
Y
a
n
g
,
H
.
Q
i
n
,
Y
.
Li
u
,
a
n
d
J.
D
i
n
g
,
“
C
C
W
-
Y
O
LO
:
a
m
o
d
i
f
i
e
d
Y
O
LO
v
5
s
n
e
t
w
o
r
k
f
o
r
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
i
n
c
o
m
p
l
e
x
t
r
a
f
f
i
c
sce
n
e
s
,
”
I
n
f
o
rm
a
t
i
o
n
,
v
o
l
.
1
5
,
n
o
.
1
2
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
i
n
f
o
1
5
1
2
0
7
6
2
.
[
2
2
]
Y
.
Za
n
g
,
R
.
C
a
o
,
H
.
L
i
,
W
.
H
u
,
a
n
d
Q
.
Li
u
,
“
M
A
P
D
:
m
u
l
t
i
-
r
e
c
e
p
t
i
v
e
f
i
e
l
d
a
n
d
a
t
t
e
n
t
i
o
n
me
c
h
a
n
i
sm
f
o
r
m
u
l
t
i
s
p
e
c
t
r
a
l
p
e
d
e
s
t
r
i
a
n
d
e
t
e
c
t
i
o
n
,
”
Vi
su
a
l
C
o
m
p
u
t
e
r
,
v
o
l
.
4
0
,
n
o
.
4
,
p
p
.
2
8
1
9
–
2
8
3
1
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
3
7
1
-
023
-
0
2
9
8
8
-
7.
[
2
3
]
T.
-
Y
.
L
i
n
e
t
a
l
.
,
“
M
i
c
r
o
s
o
f
t
C
O
C
O
:
c
o
mm
o
n
o
b
j
e
c
t
s i
n
c
o
n
t
e
x
t
,
”
i
n
C
o
m
p
u
t
e
r V
i
s
i
o
n
–
EC
C
V
2
0
1
4
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
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
.
7
4
0
–
7
5
5
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
1
0
6
0
2
-
1
_
4
8
.
[
2
4
]
J.
S
a
n
g
,
Z.
Z
h
a
n
g
,
C
.
X
i
a
o
,
H
.
Lu
o
,
a
n
d
J
.
Z
h
a
n
g
,
“
A
n
i
m
p
r
o
v
e
d
Y
O
L
O
v
8
s
me
t
h
o
d
a
n
d
i
t
s
a
p
p
l
i
c
a
t
i
o
n
i
n
r
o
a
d
t
r
a
f
f
i
c
t
a
r
g
e
t
d
e
t
e
c
t
i
o
n
,
”
H
o
n
g
w
a
i
y
u
J
i
g
u
a
n
g
G
o
n
g
c
h
e
n
g
/
I
n
f
r
a
re
d
a
n
d
L
a
ser
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
5
3
,
n
o
.
1
1
,
2
0
2
4
,
d
o
i
:
1
0
.
3
7
8
8
/
I
R
LA
2
0
2
4
0
2
5
6
.
[
2
5
]
W
.
Li
u
e
t
a
l
.
,
“
S
S
D
:
s
i
n
g
l
e
s
h
o
t
m
u
l
t
i
b
o
x
d
e
t
e
c
t
o
r
,
”
i
n
C
o
m
p
u
t
e
r
V
i
si
o
n
–
EC
C
V
2
0
1
6
,
2
0
1
6
,
p
p
.
2
1
–
37
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
319
-
4
6
4
4
8
-
0
_
2
.
[
2
6
]
A
.
R
.
G
.
-
Esc
a
l
a
n
t
e
,
R
.
Q
.
F
.
-
A
g
u
i
l
a
r
,
A
.
P
.
-
Z
u
b
i
a
,
a
n
d
E.
M
.
-
V
a
r
g
a
s
,
“
A
u
t
o
m
a
t
i
c
b
r
a
k
e
d
r
i
v
e
r
a
ssi
s
t
a
n
c
e
s
y
st
e
m
b
a
s
e
d
o
n
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
f
u
z
z
y
l
o
g
i
c
,
”
PL
o
S
O
N
E
,
v
o
l
.
1
9
,
n
o
.
1
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
3
7
1
/
j
o
u
r
n
a
l
.
p
o
n
e
.
0
3
0
8
8
5
8
.
[
2
7
]
W
.
C
i
,
L
.
T
i
a
n
,
Y
.
G
e
,
H
.
H
o
u
,
a
n
d
J
.
M
a
,
“
A
n
a
b
s
o
l
u
t
e
d
i
s
t
a
n
c
e
e
s
t
i
m
a
t
i
o
n
me
t
h
o
d
f
o
r
o
b
st
a
c
l
e
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
Y
O
LO
v
5
s
,
”
i
n
AC
M
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
Pr
o
c
e
e
d
i
n
g
S
e
r
i
e
s
,
2
0
2
4
,
p
p
.
6
–
14
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
9
0
9
3
1
.
3
6
9
0
9
3
3
.
[
2
8
]
R
.
T.
N
.
V
.
S
.
C
h
a
p
p
a
a
n
d
M
.
E
l
-
S
h
a
r
k
a
w
y
,
“
S
q
u
e
e
z
e
-
a
n
d
-
e
x
c
i
t
a
t
i
o
n
sq
u
e
e
z
e
n
e
x
t
:
a
n
e
f
f
i
c
i
e
n
t
D
N
N
f
o
r
h
a
r
d
w
a
r
e
d
e
p
l
o
y
me
n
t
,
”
i
n
2
0
2
0
1
0
t
h
A
n
n
u
a
l
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
W
o
rks
h
o
p
a
n
d
C
o
n
f
e
re
n
c
e
,
C
C
W
C
2
0
2
0
,
2
0
2
0
,
p
p
.
6
9
1
–
697
,
d
o
i
:
1
0
.
1
1
0
9
/
C
C
W
C
4
7
5
2
4
.
2
0
2
0
.
9
0
3
1
1
1
9
.
[
2
9
]
D
.
H
u
,
Z.
Z
h
a
n
g
,
a
n
d
G
.
N
i
u
,
“
La
n
e
l
i
n
e
d
e
t
e
c
t
i
o
n
i
n
c
o
r
p
o
r
a
t
i
n
g
C
B
A
M
me
c
h
a
n
i
sm
a
n
d
d
e
f
o
r
ma
b
l
e
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
,
”
Be
i
j
i
n
g
H
a
n
g
k
o
n
g
H
a
n
g
t
i
a
n
D
a
x
u
e
Xu
e
b
a
o
/
J
o
u
r
n
a
l
o
f
Be
i
j
i
n
g
U
n
i
v
e
rs
i
t
y
o
f
A
e
ro
n
a
u
t
i
c
s
a
n
d
Ast
r
o
n
a
u
t
i
c
s
,
v
o
l
.
5
0
,
n
o
.
7
,
p
p
.
2
1
5
0
–
2
1
6
0
,
2
0
2
4
,
d
o
i
:
1
0
.
1
3
7
0
0
/
j
.
b
h
.
1
0
0
1
-
5
9
6
5
.
2
0
2
2
.
0
6
0
1
.
[
3
0
]
B
.
Li
u
,
H
.
W
a
n
g
,
Y
.
W
a
n
g
,
C
.
Zh
o
u
,
a
n
d
L.
C
a
i
,
“
L
a
n
e
l
i
n
e
t
y
p
e
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
i
mp
r
o
v
e
d
Y
O
LO
v
5
,
”
Ap
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
3
,
n
o
.
1
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
3
1
8
1
0
5
3
7
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
O
u
m
a
y
m
a
R
a
c
h
id
i
is
a
t
h
i
rd
-
y
e
a
r
P
h
.
D.
st
u
d
e
n
t
a
n
d
re
c
e
iv
e
d
a
n
e
n
g
i
n
e
e
r’s
d
e
g
re
e
fro
m
th
e
Na
ti
o
n
a
l
G
ra
d
u
a
te
S
c
h
o
o
l
o
f
Arts
a
n
d
Cra
fts,
M
e
k
n
e
s,
M
o
ro
c
c
o
,
i
n
2
0
2
1
.
S
h
e
is
a
c
tu
a
ll
y
wo
rk
i
n
g
a
s
a
n
e
lec
tro
m
e
c
h
a
n
ica
l
e
n
g
in
e
e
r
in
t
h
e
in
d
u
str
y
field
.
He
r
o
n
g
o
i
n
g
d
isse
rtatio
n
st
u
d
ies
p
e
d
e
strian
d
e
tec
ti
o
n
i
n
AD
AS
sy
ste
m
s
a
n
d
e
x
p
lo
re
s
d
e
e
p
lea
rn
in
g
tec
h
n
iq
u
e
s
fo
r
o
b
jec
t
d
e
tec
ti
o
n
.
S
h
e
is
p
a
rt
icu
larly
in
tere
ste
d
i
n
in
d
u
strial
c
o
n
tr
o
l
sy
ste
m
s,
F
P
G
A
-
b
a
se
d
AD
AS
sy
ste
m
s
,
a
n
d
e
lec
tri
c
a
l
v
e
h
icle
s
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
o
u
m
.
ra
c
h
id
i@e
d
u
.
u
m
i.
a
c
.
m
a
.
Ba
d
r
Bo
u
o
u
li
d
Idriss
i
re
c
e
iv
e
d
th
e
P
h
.
D.
d
e
g
re
e
fro
m
F
a
c
u
lt
é
P
o
ly
tec
h
n
i
q
u
e
d
e
M
o
n
s,
M
o
n
s,
Be
lg
i
u
m
,
in
1
9
9
7
a
n
d
t
h
e
e
n
g
i
n
e
e
r’s
d
e
g
re
e
fro
m
Ec
o
le
Na
ti
o
n
a
le
d
e
l’I
n
d
u
strie
M
in
é
ra
le,
Ra
b
a
t,
M
o
r
o
c
c
o
,
i
n
1
9
9
2
.
S
i
n
c
e
1
9
9
9
,
h
e
h
a
s
b
e
e
n
wo
rk
i
n
g
a
t
Ec
o
le
Na
ti
o
n
a
le
S
u
p
é
rieu
re
d
’Arts et
M
é
ti
e
rs (E
NSAM
-
M
e
k
n
è
s),
M
o
u
lay
Ism
a
ïl
Un
iv
e
rsit
y
,
M
e
k
n
è
s,
M
o
r
o
c
c
o
,
wh
e
re
h
e
is
a
p
r
o
fe
ss
o
r
i
n
t
h
e
D
e
p
a
rtme
n
t
o
f
E
lec
tro
m
e
c
h
a
n
ica
l
En
g
i
n
e
e
rin
g
,
i
n
t
h
e
a
re
a
s
o
f
p
o
we
r
e
lec
tro
n
ics
a
n
d
e
lec
tri
c
a
l
m
a
c
h
in
e
s.
His
re
se
a
rc
h
in
tere
sts
a
re
m
a
in
ly
e
lec
tri
c
d
riv
e
s,
in
d
u
strial
c
o
n
tro
l
sy
ste
m
s,
DSP
/
F
P
G
A
-
b
a
se
d
AD
AS
sy
ste
m
s
a
n
d
e
lec
tri
c
a
l
v
e
h
icle
s
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
b
.
b
o
u
o
u
li
d
@u
m
i.
a
c
.
m
a
.
Cha
fik
Ed
-
Da
h
m
a
n
i
re
c
e
iv
e
d
h
is
e
n
g
in
e
e
rin
g
d
e
g
re
e
in
M
e
c
h
a
tro
n
ics
fro
m
t
h
e
Un
iv
e
rsity
o
f
Ab
d
e
lma
lek
Essa
a
d
i
i
n
2
0
1
4
a
n
d
h
is
P
h
.
D.
d
e
g
re
e
fro
m
th
e
Un
i
v
e
rsity
o
f
M
o
h
a
m
e
d
5
-
Ra
b
a
t,
M
o
r
o
c
c
o
.
Cu
rre
n
tl
y
,
h
e
is
a
n
a
ss
istan
t
p
r
o
fe
ss
o
r
a
t
t
h
e
Na
ti
o
n
a
l
G
ra
d
u
a
te
S
c
h
o
o
l
o
f
Arts
a
n
d
Cra
fts
-
M
e
k
n
è
s,
M
o
ro
c
c
o
.
His
re
se
a
rc
h
in
tere
st
s
in
c
l
u
d
e
re
n
e
wa
b
le
e
n
e
rg
y
sy
ste
m
c
o
n
v
e
rsi
o
n
,
c
o
n
tro
l
sy
ste
m
s,
a
n
d
m
icro
g
ri
d
s
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
c
.
e
d
d
a
h
m
a
n
i@u
m
i.
a
c
.
m
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.