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.
1
6
,
No
.
4
,
A
u
g
u
s
t
20
2
6
,
p
p
.
2
2
3
0
~
2
2
4
6
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
4
.
pp
2
2
3
0
-
2
2
4
6
2230
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Enha
ncement
of
YO
L
O
v
8
f
o
r
o
bje
ct
de
tect
io
n
in
a
d
v
erse
wea
ther
co
nditio
ns usi
ng
g
enerati
v
e adv
ersa
ria
l net
wo
rk
T
a
lifha
ni Ca
lv
in T
s
hi
po
t
a
,
Chun
l
ing
T
u,
M
uk
a
t
s
hu
ng
Cla
ud
e
Na
wej
,
Sem
pe
T
ho
m
L
eho
lo
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
y
st
e
m
E
n
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
I
C
T,
Ts
h
w
a
n
e
U
n
i
v
e
r
si
t
y
o
f
T
e
c
h
n
o
l
o
g
y
,
P
r
e
t
o
r
i
a
,
S
o
u
t
h
A
f
r
i
c
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
4
,
2
0
2
5
R
ev
is
ed
Ma
y
1
8
,
2
0
2
6
Acc
ep
ted
J
u
l 2
2
,
2
0
2
6
De
tec
ti
n
g
o
b
jec
ts
in
b
a
d
we
a
th
e
r
li
k
e
ra
in
,
f
o
g
,
sn
o
w,
o
r
l
o
w
li
g
h
t
is
stil
l
d
iffi
c
u
lt
b
e
c
a
u
se
v
isib
il
it
y
d
ro
p
s,
n
o
ise
i
n
c
re
a
se
s,
a
n
d
c
o
n
tras
t
g
e
ts
wo
rse
,
a
ll
o
f
wh
ic
h
h
u
rt
d
e
tec
ti
o
n
a
c
c
u
ra
c
y
.
M
o
st
c
u
rre
n
t
m
e
th
o
d
s
e
it
h
e
r
imp
ro
v
e
d
e
tec
to
r
d
e
sig
n
s
o
r
u
se
ima
g
e
p
re
p
ro
c
e
ss
in
g
o
n
t
h
e
ir
o
wn
.
Th
e
y
u
su
a
ll
y
fo
c
u
s
o
n
j
u
st
o
n
e
t
y
p
e
o
f
we
a
th
e
r
a
n
d
d
o
n
o
t
u
se
a
c
o
m
m
o
n
wa
y
t
o
e
v
a
lu
a
te
re
su
lt
s.
T
h
is
p
a
p
e
r
in
tr
o
d
u
c
e
s
a
YO
LOv
8
fra
m
e
wo
rk
imp
ro
v
e
d
wit
h
a
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
(G
AN
)
fo
r
ima
g
e
e
n
h
a
n
c
e
m
e
n
t
b
e
fo
re
d
e
tec
t
io
n
.
I
n
ste
a
d
o
f
j
u
st
m
a
k
i
n
g
ima
g
e
s
lo
o
k
b
e
tt
e
r,
t
h
e
G
AN
is
train
e
d
t
o
h
e
lp
t
h
e
o
b
jec
t
d
e
tec
to
r
w
o
rk
b
e
tt
e
r
in
t
o
u
g
h
c
o
n
d
it
io
n
s,
s
o
it
c
a
n
e
x
trac
t
fe
a
tu
re
s
m
o
re
e
ffe
c
ti
v
e
ly
wh
e
n
i
m
a
g
e
s
a
re
d
e
g
ra
d
e
d
.
Th
e
m
o
d
e
l
wa
s
tes
ted
o
n
d
a
tas
e
ts
with
d
iffere
n
t
we
a
th
e
r
c
o
n
d
i
ti
o
n
s
u
si
n
g
sta
n
d
a
rd
m
e
tri
c
s
li
k
e
P
re
c
isio
n
,
Re
c
a
ll
,
F
1
-
sc
o
re
,
a
n
d
m
e
a
n
a
v
e
ra
g
e
p
re
c
isio
n
(m
AP).
Re
su
lt
s
sh
o
w
th
a
t
t
h
is
m
e
th
o
d
c
o
n
siste
n
tl
y
imp
ro
v
e
s
p
e
rfo
rm
a
n
c
e
,
with
u
p
to
a
6
.
5
%
in
c
re
a
se
in
m
AP@0
.
5
o
v
e
r
Y
OLOv
8
-
S
T
E
a
n
d
9
.
2
%
o
v
e
r
I
A
-
YO
LO,
e
sp
e
c
ially
in
f
o
g
g
y
a
n
d
l
o
w
-
li
g
h
t
situ
a
ti
o
n
s.
T
h
e
se
re
su
lt
s
s
h
o
w
th
a
t
a
d
d
i
n
g
G
A
N
-
b
a
se
d
p
re
p
ro
c
e
ss
in
g
to
YO
LOv
8
m
a
k
e
s
d
e
tec
ti
o
n
m
o
re
r
e
li
a
b
le
a
n
d
stil
l
k
e
e
p
s
t
h
e
p
ro
c
e
ss
fa
st.
Th
is
fra
m
e
wo
rk
o
ffe
rs
a
p
ra
c
ti
c
a
l
a
n
d
sc
a
lab
l
e
so
lu
ti
o
n
f
o
r
re
a
l
-
wo
rl
d
u
se
s
li
k
e
se
lf
-
d
riv
i
n
g
c
a
rs,
sm
a
rt
tran
sp
o
rt
a
ti
o
n
,
a
n
d
su
rv
e
il
lan
c
e
.
K
ey
w
o
r
d
s
:
Ad
v
er
s
e
wea
th
er
co
n
d
itio
n
s
Au
to
n
o
m
o
u
s
v
eh
icles
Gen
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
s
Ob
ject
d
etec
tio
n
YOL
Ov
8
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
:
T
alif
h
an
i Calv
in
T
s
h
ip
o
ta
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
Facu
lty
o
f
I
C
T
,
T
s
h
wan
e
Un
iv
er
s
ity
o
f
T
ec
h
n
o
lo
g
y
B
lo
ck
K,
2
Au
b
r
e
y
Ma
tlak
ala
St,
So
s
h
an
g
u
v
e
-
K,
Pre
to
r
ia,
0
1
5
2
,
So
u
th
Af
r
ica
E
m
ail: tsh
ip
o
ta6
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
co
m
p
u
ter
v
is
io
n
,
o
b
ject
d
etec
tio
n
h
as
e
m
er
g
ed
as
a
f
u
n
d
am
en
tal
jo
b
th
at
is
ess
en
tial
f
o
r
ap
p
licatio
n
s
in
in
tellig
en
t
tr
an
s
p
o
r
tatio
n
s
y
s
tem
s
,
au
to
n
o
m
o
u
s
d
r
iv
in
g
,
a
n
d
s
u
r
v
e
illan
ce
.
R
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
is
n
o
w
p
o
s
s
ib
le
th
an
k
s
to
r
ec
en
t
d
ev
elo
p
m
en
t
s
in
s
in
g
le
-
s
tag
e
d
etec
to
r
s
,
p
ar
ticu
lar
ly
th
e
Yo
u
o
n
ly
lo
o
k
o
n
ce
(
YOL
O)
s
er
i
es,
wh
ich
h
a
v
e
p
r
o
d
u
ce
d
e
n
co
u
r
ag
in
g
s
p
ee
d
an
d
ac
cu
r
ac
y
r
esu
lts
[
1
]
,
[
2
]
.
YOL
Ov
8
is
a
p
o
p
u
lar
o
p
tio
n
f
o
r
s
m
ar
t
en
v
ir
o
n
m
en
t
d
e
p
lo
y
m
en
ts
d
u
e
t
o
its
r
o
b
u
s
t
tr
ad
e
-
o
f
f
b
etwe
en
d
etec
tio
n
p
r
ec
is
io
n
a
n
d
in
f
er
e
n
ce
tim
e
[
3
]
.
E
v
en
with
th
ese
a
d
v
an
ce
m
en
t
s
,
in
clem
en
t
wea
th
er
-
s
u
c
h
as
r
ain
,
f
o
g
,
s
n
o
w,
an
d
p
o
o
r
lig
h
t
-
co
n
tin
u
es
to
p
o
s
e
s
ig
n
if
ica
n
t
ch
allen
g
e
s
to
ac
cu
r
ate
d
etec
tio
n
.
T
h
r
o
u
g
h
d
ec
r
ea
s
ed
v
is
ib
ilit
y
,
b
lu
r
,
n
o
is
e,
an
d
p
o
o
r
co
n
tr
ast,
th
ese
f
ac
to
r
s
d
eter
io
r
ate
im
ag
e
q
u
ality
,
wh
ich
h
as
an
ad
v
er
s
e
ef
f
ec
t
o
n
d
etec
tio
n
m
ea
s
u
r
es
s
u
ch
as
m
ea
n
av
er
a
g
e
p
r
e
cisi
o
n
(
m
AP
)
,
r
ec
all,
an
d
p
r
ec
is
io
n
.
Fo
r
ex
am
p
le,
m
o
d
els
tr
ain
ed
in
clea
r
wea
th
er
f
r
eq
u
en
tly
p
er
f
o
r
m
p
o
o
r
ly
i
n
f
o
g
o
r
wet
c
o
n
d
itio
n
s
,
lead
in
g
to
s
ig
n
if
ica
n
t d
ec
lin
es in
d
etec
tio
n
p
er
f
o
r
m
an
ce
[
4
]
–
[
6
]
.
R
ec
en
t
s
tu
d
ies
h
av
e
lo
o
k
ed
in
two
p
r
im
ar
y
ap
p
r
o
ac
h
es
to
r
eso
lv
e
th
ese
is
s
u
es.
I
m
p
r
o
v
in
g
d
etec
tio
n
m
o
d
el
d
esig
n
s
is
th
e
in
itial
s
t
ep
.
T
o
in
cr
ea
s
e
r
o
b
u
s
tn
ess
in
th
e
f
ac
e
o
f
u
n
f
av
o
r
ab
le
wea
t
h
er
,
YOL
Ov
8
-
STE
,
f
o
r
in
s
tan
ce
,
co
m
b
in
es
a
s
p
atial
-
tem
p
o
r
al
f
ea
tu
r
e
e
x
tr
ac
t
io
n
m
o
d
u
le,
an
ef
f
ec
tiv
e
m
u
lti
-
s
ca
le
atten
tio
n
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
E
n
h
a
n
ce
men
t o
f YOLOv8
fo
r
o
b
ject
d
etec
tio
n
in
a
d
ve
r
s
e
w
ea
th
er c
o
n
d
itio
n
s
…
(
Ta
lifh
a
n
i
C
a
lvin
Ts
h
ip
o
ta
)
2231
m
ec
h
an
is
m
,
an
d
s
o
f
t
-
NM
S
[
7
]
.
Sp
ec
ialized
atten
tio
n
m
o
d
u
les
ar
e
in
t
r
o
d
u
ce
d
b
y
o
th
e
r
wo
r
k
s
lik
e
Sn
o
wYOL
O
an
d
W
ea
th
er
Net
to
co
m
b
at
f
o
g
a
n
d
s
n
o
w
o
cc
lu
s
io
n
[
8
]
,
[
9
]
.
Pre
p
r
o
ce
s
s
in
g
o
r
d
ata
au
g
m
e
n
tatio
n
/en
h
an
ce
m
en
t
tech
n
iq
u
es
ar
e
u
s
ed
in
th
e
s
ec
o
n
d
s
tr
ateg
y
.
Fo
r
in
s
tan
ce
,
“
Ob
ject
d
etec
tio
n
in
ad
v
er
s
e
wea
th
er
f
o
r
au
to
n
o
m
o
u
s
d
r
iv
in
g
th
r
o
u
g
h
d
ata
m
er
g
in
g
an
d
YOL
Ov
8
”
s
h
o
ws
th
at
YOL
Ov
8
p
er
f
o
r
m
s
s
ig
n
if
ican
tly
b
etter
in
wea
t
h
er
-
v
ar
y
in
g
co
n
d
itio
n
s
th
an
b
ase
weig
h
ts
wh
en
tr
ain
in
g
o
n
m
er
g
ed
d
atasets
(
f
o
r
i
n
s
tan
ce
,
co
m
b
in
in
g
A
C
DC
an
d
DAW
N)
o
r
u
s
in
g
au
g
m
en
tatio
n
[
1
]
.
Similar
ly
,
to
im
p
r
o
v
e
im
ag
es
b
ef
o
r
e
o
r
d
u
r
in
g
d
etec
tio
n
,
im
ag
e
-
ad
ap
tiv
e
YOL
O
(
I
A
-
Y
OL
O)
in
co
r
p
o
r
ates
d
if
f
er
en
tiab
le
im
ag
e
p
r
o
ce
s
s
in
g
m
o
d
u
les,
p
ar
ticu
lar
ly
in
lo
w
-
lig
h
t
an
d
f
o
g
g
y
s
itu
atio
n
s
[
4
]
.
Gen
er
ati
v
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs)
ar
e
u
s
ed
in
o
th
er
r
esear
ch
to
r
esto
r
e
im
ag
es
b
ef
o
r
e
d
etec
tio
n
,
wh
ich
g
r
ea
tly
en
h
an
ce
s
p
er
f
o
r
m
a
n
ce
in
h
az
e
an
d
r
ain
[
1
0
]
,
[
1
1
]
.
T
h
er
e
ar
e
s
till
a
n
u
m
b
e
r
o
f
g
a
p
s
,
th
o
u
g
h
:
m
o
s
t
s
tu
d
ies
o
n
ly
co
n
s
id
er
o
n
e
ty
p
e
o
f
wea
th
er
o
r
test
o
n
s
m
all
d
ataset
s
;
co
m
b
in
ed
en
h
an
ce
m
e
n
t
+
d
etec
tio
n
p
ip
elin
es
ar
e
f
r
eq
u
en
tly
lack
in
g
(
p
ar
ticu
lar
ly
f
o
r
YOL
Ov
8
)
ac
r
o
s
s
a
v
ar
iety
o
f
u
n
f
a
v
o
r
a
b
le
wea
th
er
c
o
n
d
itio
n
s
;
an
d
s
o
m
e
en
h
a
n
ce
m
en
t
te
ch
n
iq
u
es
p
r
io
r
itize
p
er
ce
p
tu
al
im
a
g
e
q
u
ality
o
v
e
r
ac
tu
al
d
o
wn
s
tr
ea
m
d
etec
tio
n
ac
cu
r
ac
y
.
T
h
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
in
tr
o
d
u
ce
s
a
u
n
if
ied
GAN
–
YOL
Ov
8
f
r
a
m
ewo
r
k
th
at
is
ev
alu
ated
ac
r
o
s
s
m
u
ltip
le
ad
v
er
s
e
wea
th
er
s
ce
n
ar
io
s
u
s
in
g
a
co
n
s
is
ten
t
ex
p
er
im
en
tal
p
r
o
to
co
l,
in
co
n
tr
ast
to
p
r
ev
io
u
s
G
AN
–
YOL
O
p
ip
elin
es
th
at
m
ai
n
ly
f
o
cu
s
o
n
s
in
g
le
wea
th
er
co
n
d
itio
n
s
o
r
ea
r
lier
YOL
O
v
er
s
io
n
s
(
e.
g
.
,
YOL
O
v
5
o
r
YOL
Ov
7
)
.
U
n
lik
e
p
e
r
ce
p
tu
al
-
q
u
ality
-
d
r
iv
en
en
h
an
ce
m
e
n
t te
ch
n
iq
u
es,
o
u
r
GAN
is
s
p
ec
if
ically
tu
n
ed
f
o
r
d
o
wn
s
tr
ea
m
d
etec
tio
n
p
er
f
o
r
m
an
ce
in
s
tead
o
f
ju
s
t
v
is
u
al
m
etr
ics.
Ad
d
itio
n
ally
,
th
e
in
teg
r
atio
n
is
m
ad
e
to
b
e
l
ig
h
tweig
h
t
an
d
m
o
d
el
-
ag
n
o
s
tic,
m
ain
tain
in
g
th
e
r
ea
l
-
tim
e
f
ea
tu
r
es
o
f
YOL
Ov
8
.
T
o
th
e
b
est
o
f
o
u
r
k
n
o
wled
g
e,
th
is
wo
r
k
o
f
f
er
s
th
e
f
ir
s
t
th
o
r
o
u
g
h
ass
ess
m
en
t
o
f
GAN
-
b
ased
p
r
ep
r
o
ce
s
s
in
g
th
at
is
ex
p
licitly
in
teg
r
ated
with
YOL
Ov
8
an
d
v
alid
ated
u
s
in
g
in
f
er
e
n
ce
-
tim
e
an
aly
s
is
,
ab
latio
n
s
tu
d
ies,
a
n
d
m
u
lti
-
r
u
n
s
tatis
tical
co
n
s
is
ten
cy
u
n
d
e
r
lo
w
-
lig
h
t,
r
ai
n
,
f
o
g
,
an
d
s
n
o
w
co
n
d
itio
n
s
.
T
h
is
wo
r
k
p
r
o
p
o
s
es
a
n
o
v
el
GAN
-
en
h
an
ce
d
YOL
Ov
8
f
r
a
m
ewo
r
k
th
at
in
c
o
r
p
o
r
ates
a
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
(
GAN)
a
s
a
p
r
ep
r
o
ce
s
s
in
g
m
o
d
u
le
to
i
m
p
r
o
v
e
im
ag
e
q
u
ality
b
e
f
o
r
e
o
b
ject
r
ec
o
g
n
itio
n
,
th
er
eb
y
o
v
er
co
m
i
n
g
th
ese
is
s
u
es.
T
h
e
s
u
g
g
ested
m
eth
o
d
p
r
o
v
id
es a
h
y
b
r
id
e
n
h
an
ce
m
e
n
t
–
d
etec
tio
n
p
ip
elin
e
in
wh
ich
th
e
GAN
is
tailo
r
ed
t
o
im
p
r
o
v
e
d
o
wn
s
tr
ea
m
d
etec
ti
o
n
p
e
r
f
o
r
m
an
ce
r
ath
er
th
a
n
p
er
ce
p
tu
al
q
u
ality
,
in
co
n
tr
ast
to
tr
ad
itio
n
al
ap
p
r
o
ac
h
es
th
at
f
o
cu
s
eith
er
o
n
d
etec
to
r
ar
c
h
itectu
r
e
o
r
in
d
ep
en
d
en
t
im
ag
e
en
h
an
ce
m
e
n
t.
T
h
e
m
ain
n
o
v
e
lty
is
th
e
clo
s
e
in
teg
r
atio
n
o
f
YOL
Ov
8
an
d
GAN
-
b
ased
r
e
s
to
r
atio
n
,
en
a
b
lin
g
b
etter
f
ea
tu
r
e
ex
tr
ac
tio
n
u
n
d
e
r
p
o
o
r
v
is
ib
ilit
y
co
n
d
itio
n
s
wh
ile
m
ain
tain
in
g
r
ea
l
-
tim
e
in
f
er
e
n
ce
.
T
h
is
ap
p
r
o
ac
h
im
p
r
o
v
es
r
o
b
u
s
tn
ess
in
a
v
ar
ie
ty
o
f
u
n
f
av
o
r
ab
le
wea
th
er
c
o
n
d
itio
n
s
b
y
o
f
f
er
in
g
a
lig
h
tweig
h
t,
m
o
d
el
-
a
g
n
o
s
tic
s
o
lu
tio
n
.
T
h
is
p
ap
er
in
tr
o
d
u
ce
s
a
GAN
-
en
h
an
ce
d
YOL
Ov
8
d
etec
tio
n
f
r
am
ewo
r
k
to
s
o
lv
e
th
e
d
if
f
i
cu
lties
o
f
v
eh
icle
r
ec
o
g
n
itio
n
in
b
ad
wea
th
er
.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
ar
e
s
u
m
m
a
r
ized
as
f
o
llo
ws:
a.
H
y
b
r
i
d
GAN
–
YO
L
O
v
8
F
r
a
m
e
w
o
r
k
:
T
h
i
s
s
t
u
d
y
i
n
t
r
o
d
u
c
e
s
a
n
o
v
e
l
h
y
b
r
i
d
s
y
s
t
e
m
t
h
a
t
in
t
e
g
r
a
t
e
s
GA
N
-
b
ased
im
ag
e
en
h
a
n
ce
m
en
t
w
ith
YOL
Ov
8
to
im
p
r
o
v
e
o
b
j
ec
t
d
etec
tio
n
p
er
f
o
r
m
an
ce
u
n
d
er
ch
allen
g
in
g
wea
th
er
co
n
d
itio
n
s
.
b.
Dete
ctio
n
-
o
r
ien
ted
im
ag
e
en
h
an
ce
m
en
t:
u
n
lik
e
co
n
v
en
tio
n
a
l
GAN
-
b
ased
ap
p
r
o
ac
h
es
th
at
f
o
cu
s
o
n
v
is
u
al
q
u
ality
m
etr
ics
s
u
ch
as
PS
NR
an
d
SS
I
M,
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
e
n
h
an
ce
s
im
ag
es
s
p
ec
if
ic
ally
to
im
p
r
o
v
e
d
o
wn
s
tr
ea
m
o
b
ject
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
.
c.
Mu
lti
-
wea
th
er
r
o
b
u
s
tn
ess
:
t
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
is
ev
alu
ated
u
n
d
er
m
u
ltip
le
ad
v
er
s
e
wea
th
er
co
n
d
itio
n
s
,
i
n
clu
d
in
g
r
ai
n
,
f
o
g
,
s
n
o
w,
an
d
lo
w
-
lig
h
t
en
v
ir
o
n
m
en
ts
,
ad
d
r
ess
in
g
th
e
lim
itatio
n
o
f
ex
is
tin
g
m
eth
o
d
s
th
at
ar
e
t
y
p
ically
d
esi
g
n
ed
f
o
r
a
s
in
g
le
wea
th
e
r
s
ce
n
ar
io
.
d.
C
o
m
p
r
eh
en
s
iv
e
ex
p
er
im
en
ta
l
v
alid
atio
n
:
ex
ten
s
iv
e
ex
p
er
im
en
ts
,
in
clu
d
in
g
ab
latio
n
s
tu
d
ies
an
d
co
m
p
ar
is
o
n
s
with
s
tate
-
of
-
th
e
-
ar
t
m
eth
o
d
s
,
d
em
o
n
s
tr
ate
c
o
n
s
is
ten
t
im
p
r
o
v
em
e
n
ts
in
d
et
ec
tio
n
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
.
e.
R
ea
l
-
tim
e
d
ep
lo
y
m
en
t
f
ea
s
ib
ilit
y
:
d
esp
ite
th
e
in
clu
s
io
n
o
f
GAN
-
b
ased
p
r
ep
r
o
ce
s
s
in
g
,
th
e
f
r
am
ewo
r
k
m
ain
tain
s
n
ea
r
r
ea
l
-
tim
e
p
e
r
f
o
r
m
a
n
ce
,
m
ak
in
g
it
s
u
itab
le
f
o
r
p
r
ac
tical
d
ep
l
o
y
m
en
t
in
i
n
tellig
en
t
tr
an
s
p
o
r
tatio
n
a
n
d
s
u
r
v
eillan
ce
s
y
s
tem
s
.
Pap
er
s
tr
u
ctu
r
e:
T
h
e
r
em
ain
d
er
o
f
th
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
ws.
R
e
lated
wo
r
k
is
r
ev
iewe
d
in
s
ec
tio
n
2
.
Ou
r
m
eth
o
d
o
lo
g
y
(
d
atasets
,
GAN
d
esig
n
,
Y
OL
Ov
8
in
teg
r
atio
n
)
is
ex
p
lain
ed
in
s
ec
tio
n
3
.
C
o
m
p
ar
ativ
e
an
aly
s
is
an
d
ex
p
er
im
en
tal
r
esu
lts
ar
e
p
r
esen
ted
in
s
ec
tio
n
4
.
Dis
cu
s
s
io
n
a
n
d
p
er
s
p
ec
tiv
es
ar
e
p
r
o
v
id
e
d
in
s
ec
tio
n
5
.
Sectio
n
6
wr
ap
s
u
p
a
n
d
o
f
f
er
s
id
ea
s
f
o
r
f
u
r
th
e
r
r
esear
ch
.
2.
RE
L
AT
E
D
W
O
R
K
In
r
ec
en
t
y
ea
r
s
,
a
lo
t
o
f
r
es
ea
r
ch
h
as
b
ee
n
d
o
n
e
on
r
eli
ab
le
o
b
ject
r
ec
o
g
n
itio
n
in
a
v
ar
iety
of
en
v
ir
o
n
m
en
tal
cir
c
u
m
s
tan
ce
s
,
with
a
p
ar
ticu
lar
e
m
p
h
asis
o
n
en
h
an
cin
g
r
esil
ien
ce
i
n
a
p
p
licatio
n
s
s
u
ch
as
s
u
r
v
eillan
ce
an
d
au
to
n
o
m
o
u
s
d
r
iv
in
g
.
T
h
e
YOL
O
f
am
ily
h
as
g
ain
ed
p
o
p
u
lar
ity
b
ec
au
s
e
it
ca
n
ac
co
m
p
lis
h
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
2
2
3
0
-
2246
2232
r
ea
l
-
tim
e
in
f
er
e
n
ce
wh
ile
k
ee
p
in
g
c
o
m
p
etitiv
e
ac
c
u
r
ac
y
,
ev
en
wh
en
cu
r
r
en
t
d
etec
to
r
s
lik
e
Fas
ter
R
-
C
NN
an
d
SSD
o
f
f
er
g
r
ea
t
p
er
f
o
r
m
an
ce
[
1
2
]
,
[
1
3
]
.
Alth
o
u
g
h
th
e
m
o
s
t
r
ec
en
t
v
er
s
io
n
,
YOL
Ov
8
,
h
a
s
s
h
o
wn
im
p
r
o
v
ed
s
p
ee
d
an
d
ac
cu
r
ac
y
,
it
s
till
h
as
tr
o
u
b
le
in
s
itu
atio
n
s
with
u
n
f
a
v
o
r
ab
le
wea
th
er
,
s
u
ch
as
r
ai
n
,
f
o
g
,
s
n
o
w,
an
d
lo
w
lig
h
t
lev
els
[
1
4
]
,
[
1
5
]
.
T
h
ese
d
r
awb
ac
k
s
r
esu
lt
f
r
o
m
p
r
e
v
io
u
s
m
o
d
els’
lim
ited
ca
p
ac
ity
to
g
en
er
alize
to
v
is
u
ally
im
p
air
ed
s
ettin
g
s
d
u
e
to
th
eir
p
r
ep
o
n
d
er
a
n
ce
o
f
t
r
ain
in
g
o
n
c
lear
-
wea
th
er
d
atasets
.
Nu
m
er
o
u
s
i
n
v
esti
g
atio
n
s
h
a
v
e
en
d
ea
v
o
r
e
d
to
en
h
an
ce
YOL
O
-
b
ased
m
o
d
els
f
o
r
u
n
f
av
o
r
ab
le
wea
th
er
co
n
d
itio
n
s
.
Fo
r
e
x
am
p
le
[
1
4
]
s
h
o
wn
th
at
in
cr
ea
s
in
g
th
e
d
iv
er
s
ity
o
f
tr
ai
n
in
g
d
ata
c
an
g
r
ea
tly
im
p
r
o
v
e
YOL
Ov
8
p
er
f
o
r
m
an
ce
b
y
p
r
o
p
o
s
in
g
a
d
ata
-
m
er
g
in
g
tech
n
i
q
u
e
em
p
lo
y
in
g
th
e
AC
DC
an
d
DAWN
d
atase
ts
.
Similar
to
th
is
,
C
h
en
[
1
5
]
as
s
ess
ed
YOL
O
m
o
d
els
u
s
in
g
t
h
e
Fo
g
g
y
C
ity
s
ca
p
es
d
ataset
a
n
d
f
o
u
n
d
th
at
th
e
y
p
er
f
o
r
m
ed
n
o
ticea
b
l
y
wo
r
s
e
in
f
o
g
,
esp
ec
ially
wh
en
it
ca
m
e
to
id
en
tify
in
g
s
m
all
o
r
p
ar
tially
o
b
s
cu
r
ed
v
eh
icles.
I
n
r
esp
o
n
s
e
[
1
5
]
p
r
esen
ted
YOL
Ov
8
-
STE
,
an
im
p
r
o
v
e
d
v
er
s
io
n
th
at
in
co
r
p
o
r
ates
m
u
lti
-
s
ca
le
atten
tio
n
p
r
o
ce
s
s
es,
s
o
f
t
n
o
n
-
m
ax
im
u
m
s
u
p
p
r
ess
io
n
(
s
o
f
t
-
NM
S),
an
d
s
p
ati
al
-
tem
p
o
r
al
f
ea
t
u
r
e
ex
tr
ac
tio
n
.
YOL
Ov
8
-
STE
r
e
g
u
lar
ly
b
ea
t
b
aselin
e
YOL
Ov
8
b
y
[
6
]
[
8
]
p
er
ce
n
t
in
m
AP@
0
.
5
in
f
o
g
a
n
d
s
n
o
w
co
n
d
itio
n
s
,
ac
co
r
d
in
g
to
th
eir
r
esu
lts
o
n
th
e
R
T
T
S
an
d
DA
W
N
d
atasets
.
I
t
was
al
s
o
s
u
g
g
ested
to
u
s
e
I
m
ag
e
-
Ad
ap
tiv
e
YOL
O
(
I
A
-
YOL
O)
,
wh
ich
in
teg
r
ates
d
if
f
er
e
n
tiab
le
im
ag
e
p
r
o
ce
s
s
in
g
lay
e
r
s
in
to
th
e
d
etec
tio
n
p
ip
elin
e
[
1
6
]
.
B
y
d
y
n
am
ically
b
o
o
s
tin
g
v
is
u
al
q
u
alities
b
ased
o
n
th
e
ty
p
e
o
f
d
eg
r
a
d
atio
n
,
t
h
is
u
p
d
ate
allo
wed
th
e
n
etwo
r
k
to
o
p
er
ate
b
etter
in
b
o
th
lo
w
-
lig
h
t
an
d
f
o
g
g
y
c
o
n
d
itio
n
s
.
T
h
ese
r
esu
lts
h
ig
h
li
g
h
t
th
e
im
p
o
r
ta
n
ce
o
f
ad
ap
tin
g
d
etec
tio
n
m
eth
o
d
s
to
th
e
u
n
iq
u
e
c
h
allen
g
es p
o
s
e
d
b
y
wea
th
e
r
-
d
e
g
r
ad
ed
en
v
ir
o
n
m
en
ts
.
R
esear
ch
er
s
ar
e
in
cr
ea
s
in
g
ly
u
s
in
g
g
en
er
ativ
e
a
d
v
er
s
ar
ia
l
n
etwo
r
k
s
(
GANs)
f
o
r
p
r
e
p
r
o
ce
s
s
in
g
an
d
p
ictu
r
e
r
esto
r
atio
n
in
ch
allen
g
in
g
s
ce
n
a
r
io
s
,
in
a
d
d
itio
n
to
d
etec
to
r
-
f
o
cu
s
ed
ad
v
an
ce
m
e
n
ts
.
Fo
r
im
ag
e
-
to
-
im
ag
e
tr
an
s
latio
n
p
r
o
b
lem
s
,
GAN
-
b
ased
m
o
d
els
lik
e
Pix
2
Pix
[
1
2
]
an
d
C
y
cleG
AN
[
1
7
]
h
av
e
g
ain
e
d
p
o
p
u
lar
ity
d
u
e
to
th
eir
a
b
ilit
y
to
tr
an
s
f
o
r
m
d
eter
io
r
ated
im
ag
es
in
to
clea
r
o
n
es
with
o
u
t
th
e
n
ee
d
f
o
r
p
air
ed
d
atasets
.
B
u
ild
in
g
o
n
th
ese
p
illar
s
,
it
h
as
b
ee
n
s
u
g
g
ested
th
at
Deh
az
eGA
N
[
1
8
]
an
d
its
v
ar
iatio
n
s
ca
n
p
r
ec
is
ely
r
esto
r
e
v
is
io
n
in
b
lu
r
r
y
im
ag
es,
wh
ich
will
r
esu
lt
in
s
ig
n
if
ican
t
ad
v
a
n
ce
m
en
ts
in
o
b
ject
d
etec
tio
n
jo
b
s
d
o
wn
th
e
r
o
ad
.
Similar
ly
,
it
h
as
b
ee
n
s
h
o
wn
th
at
GAN
-
b
ased
lo
w
-
lig
h
t
im
p
r
o
v
e
m
en
t
f
r
am
ewo
r
k
s
im
p
r
o
v
e
th
e
c
o
n
t
r
ast
an
d
clar
ity
o
f
f
ea
t
u
r
es,
e
n
ab
lin
g
d
etec
to
r
s
lik
e
Fas
ter
R
-
C
NN
an
d
YOL
O
to
p
er
f
o
r
m
b
etter
in
lo
w
-
lig
h
t
co
n
d
itio
n
s
[
1
9
]
.
R
ec
en
t
r
esear
ch
h
as
also
d
e
m
o
n
s
tr
ated
t
h
e
b
en
ef
its
o
f
p
ip
elin
e
in
teg
r
atio
n
,
w
h
ich
in
v
o
lv
es
a
d
ir
ec
t
r
elatio
n
s
h
ip
b
etwe
en
o
b
ject
d
etec
to
r
s
an
d
GAN
p
r
e
p
r
o
ce
s
s
in
g
.
Sev
er
al
s
tu
d
ies
em
p
lo
y
C
y
cleG
AN
to
p
r
o
d
u
ce
“
clea
r
e
r
”
im
ag
es
f
r
o
m
r
ain
y
o
r
s
n
o
wy
co
n
d
itio
n
s
b
ef
o
r
e
tr
an
s
f
er
r
in
g
th
em
in
to
YOL
Ov
5
/YOL
Ov
8
,
wh
ich
en
h
an
ce
s
p
r
ec
is
io
n
a
n
d
r
ec
all,
as
o
p
p
o
s
ed
to
u
s
in
g
th
e
d
etec
to
r
alo
n
e
[
2
0
]
,
[
1
3
]
.
Desp
ite
th
ese
ad
v
an
ce
m
en
ts
,
th
er
e
ar
e
s
till
a
n
u
m
b
er
o
f
r
e
s
ea
r
ch
g
ap
s
th
at
n
ee
d
to
b
e
ad
d
r
ess
ed
.
First,
m
o
s
t
s
tu
d
ies
co
n
ce
n
tr
ate
o
n
o
n
e
wea
th
er
co
n
d
itio
n
,
s
u
ch
as
f
o
g
o
r
lo
w
lig
h
t,
an
d
r
ar
ely
ev
alu
ate
in
m
an
y
ad
v
e
r
s
e
wea
th
er
co
n
d
itio
n
s
.
Seco
n
d
,
b
ec
au
s
e
m
an
y
GAN
-
b
ased
tech
n
iq
u
es
ar
e
d
esig
n
ed
f
o
r
p
er
ce
p
tu
a
l
q
u
ality
m
etr
ics
lik
e
SS
I
M
an
d
PS
NR
r
ath
er
th
an
o
b
ject
d
etec
tio
n
an
d
id
en
tific
atio
n
m
etr
ics
lik
e
m
AP
o
r
r
ec
all,
th
eir
ap
p
licab
ilit
y
in
r
ea
l
-
wo
r
ld
d
etec
tio
n
p
ip
elin
es
i
s
lim
ited
.
L
ast
b
u
t
n
o
t
leas
t,
t
h
e
co
m
b
in
atio
n
o
f
YOL
Ov
8
with
GAN
-
b
ased
p
r
e
-
p
r
o
ce
s
s
in
g
h
as
n
o
t
b
ee
n
f
u
lly
lo
o
k
ed
in
to
in
a
co
h
esiv
e
f
r
a
m
ewo
r
k
in
a
v
ar
iety
o
f
s
ce
n
ar
i
o
s
,
in
clu
d
in
g
lo
w
lig
h
t,
f
o
g
,
r
ain
,
an
d
s
n
o
w
.
T
o
ad
d
r
ess
th
ese
is
s
u
es,
we
d
ev
elo
p
an
d
ass
ess
a
GAN
-
en
h
an
ce
d
YOL
Ov
8
p
i
p
elin
e
th
at
in
cr
ea
s
es
th
e
r
o
b
u
s
tn
ess
o
f
o
b
ject
d
etec
tio
n
an
d
id
en
tific
atio
n
in
a
r
an
g
e
o
f
u
n
f
av
o
r
ab
le
wea
th
er
co
n
d
itio
n
s
.
3.
M
E
T
H
O
DO
L
O
G
Y
T
h
is
s
tu
d
y
u
s
es
a
s
y
s
tem
atic
s
t
r
ateg
y
to
ass
ess
an
d
im
p
r
o
v
e
t
h
e
ca
p
ac
ity
to
d
etec
t
an
d
id
en
t
if
y
o
b
jects
in
in
clem
en
t
wea
th
er
c
o
n
d
it
io
n
.
YOL
Ov
8
o
b
ject
id
en
tif
icatio
n
an
d
r
ec
o
g
n
itio
n
,
a
GAN
-
b
ased
im
ag
e
en
h
an
ce
m
e
n
t
m
o
d
u
le
,
d
ata
p
r
ep
ar
atio
n
,
an
d
s
tan
d
ar
d
ize
d
ass
ess
m
en
t
m
eth
o
d
s
ar
e
all
co
m
b
in
ed
in
to
a
s
in
g
le
p
ip
elin
e.
T
o
en
s
u
r
e
th
o
r
o
u
g
h
b
en
ch
m
ar
k
in
g
,
th
e
p
r
o
ce
s
s
s
tar
ts
b
y
b
u
ild
in
g
a
m
u
lti
-
w
ea
th
er
d
ataset
th
at
in
clu
d
es
clea
r
,
r
ain
,
f
o
g
,
s
n
o
w,
an
d
lo
w
lig
h
t
lev
els.
GAN
–
YOL
Ov
8
f
r
am
ewo
r
k
.
Prio
r
to
b
ein
g
s
en
t
in
to
th
e
YOL
Ov
8
d
etec
tio
n
f
r
am
ewo
r
k
,
d
am
ag
e
d
im
ag
es
ar
e
p
r
ep
r
o
ce
s
s
ed
u
s
in
g
a
Gen
er
ativ
e
A
d
v
er
s
ar
ial
Netwo
r
k
(
GAN)
to
im
p
r
o
v
e
in
p
u
t
q
u
a
lity
.
W
h
ile
o
p
tim
ized
h
y
p
er
p
ar
am
eter
an
d
tr
an
s
f
er
lea
r
n
in
g
ar
e
u
tili
ze
d
f
o
r
tr
ain
in
g
,
m
etr
ics
lik
e
as
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
m
AP@
0
.
5
,
an
d
m
AP@
0
.
5
–
0
.
9
5
a
r
e
u
s
ed
f
o
r
ev
al
u
atio
n
.
T
h
e
ar
ch
itectu
r
e,
tr
ain
in
g
s
etu
p
,
ass
ess
m
en
t
p
r
o
ce
s
s
,
an
d
d
ataset
p
r
ep
ar
atio
n
o
f
t
h
e
s
y
s
tem
ar
e
co
v
e
r
ed
in
d
etail
in
th
e
en
s
u
in
g
s
u
b
s
ec
tio
n
s
.
3
.
1
.
Da
t
a
s
et
p
r
e
p
a
r
a
t
i
o
n
I
n
o
r
d
er
to
ass
ess
th
e
r
esil
ien
ce
o
f
o
b
ject
d
etec
tio
n
an
d
id
en
tific
atio
n
m
o
d
els
u
n
d
e
r
d
if
f
e
r
en
t
en
v
ir
o
n
m
en
tal
c
o
n
d
itio
n
,
th
is
s
tu
d
y
u
s
ed
wea
th
e
r
-
d
e
g
r
ad
e
d
d
atasets
.
T
h
e
f
o
llo
win
g
f
i
v
e
co
n
d
itio
n
s
wer
e
tak
en
in
to
c
o
n
s
id
er
atio
n
:
lo
w
lig
h
t,
f
o
g
,
s
n
o
w,
r
ain
,
a
n
d
clea
r
.
T
o
m
ak
e
s
u
r
e
th
at
a
r
ea
li
s
tic
d
is
tr
ib
u
tio
n
of
o
b
ject
class
es,
in
clu
d
in
g
au
to
m
o
b
ile,
b
u
s
,
tr
u
ck
,
m
o
to
r
cy
cle,
b
icy
cle,
an
d
p
e
r
s
o
n
,
each
d
ataset
was
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
E
n
h
a
n
ce
men
t o
f YOLOv8
fo
r
o
b
ject
d
etec
tio
n
in
a
d
ve
r
s
e
w
ea
th
er c
o
n
d
itio
n
s
…
(
Ta
lifh
a
n
i
C
a
lvin
Ts
h
ip
o
ta
)
2233
m
eticu
lo
u
s
ly
s
elec
ted
.
T
h
e
s
ig
n
if
ican
ce
o
f
th
ese
class
es
in
s
u
r
v
eillan
ce
an
d
s
m
ar
t
d
r
iv
in
g
s
y
s
tem
s
led
to
th
eir
s
elec
tio
n
[
2
1
]
,
[
2
2
]
.
T
o
en
s
u
r
e
co
m
p
atib
ilit
y
with
YOL
Ov
8
tr
ain
in
g
,
d
ata
a
n
n
o
t
atio
n
was
d
o
n
e
u
s
in
g
th
e
C
OC
O
f
o
r
m
at.
T
h
e
in
clu
s
io
n
of
u
n
f
av
o
r
ab
le
wea
th
er
d
atasets
was
m
o
tiv
ated
by
ea
r
lier
r
esear
ch
s
u
ch
as
AC
D
C
[
2
3
]
,
DAWN
[
2
3
]
,
an
d
Fo
g
g
y
C
ity
s
ca
p
es
[
2
2
]
,
wh
ich
h
ig
h
lig
h
te
d
th
e
im
p
o
r
tan
ce
of
d
iv
er
s
if
y
in
g
tr
ain
in
g
s
am
p
les
to
en
h
an
ce
g
en
er
aliza
tio
n
in
ch
allen
g
in
g
s
itu
atio
n
s
.
T
ab
le
1
s
u
m
m
ar
izes
th
e
d
ataset
d
is
tr
ib
u
tio
n
ac
co
r
d
in
g
to
w
ea
th
er
co
n
d
itio
n
s
,
wh
ile
T
ab
le
2
p
r
esen
ts
th
e
d
is
tr
ib
u
tio
n
o
f
o
b
ject
in
s
tan
ce
s
f
o
r
ea
ch
tar
g
et
class
.
Fig
u
r
e
1
as
well
p
r
o
v
id
es
r
ep
r
ese
n
tativ
e
im
ag
es
th
at
h
ig
h
lig
h
t
th
e
im
p
ac
t
o
f
ea
ch
c
o
n
d
itio
n
on
v
is
ib
ilit
y
.
Fo
r
in
s
tan
ce
,
s
n
o
w
p
r
o
d
u
ce
s
o
p
ac
ity
f
r
o
m
f
lak
es,
but
f
o
g
in
tr
o
d
u
ce
s
b
lu
r
r
in
g
a
n
d
p
o
o
r
co
n
tr
ast.
T
h
ese
d
eter
io
r
atio
n
s
ar
e
e
x
am
p
les
of
r
ea
l
-
wo
r
ld
p
r
o
b
lem
s
th
at
o
b
ject
d
etec
to
r
s
f
r
eq
u
e
n
tly
can
no
t so
lv
e
[
2
3
]
,
[
2
4
]
.
T
ab
le
1.
Dis
tr
ib
u
tio
n
of
an
n
o
ta
ted
im
ag
es
p
er
wea
th
e
r
co
n
d
iti
o
n
in
th
e
d
a
t
a
s
e
t
W
e
a
t
h
e
r
c
o
n
d
i
t
i
o
n
T
r
a
i
n
V
a
l
i
d
a
t
i
o
n
T
e
s
t
T
o
t
a
l
Clear
2706
7
6
2
3
4
3
3811
Rain
2281
6
4
2
2
8
9
3212
F
o
g
2067
5
8
2
2
6
2
2911
S
n
o
w
2209
6
2
2
2
8
0
3111
Lo
w
l
i
g
h
t
1855
5
2
2
2
3
5
2612
T
o
t
a
l
1
1
1
1
8
3130
1409
1
5
6
5
7
T
ab
le
2
.
Ob
ject
class
d
is
tr
ib
u
tio
n
in
th
e
d
ataset
O
b
j
e
c
t
c
l
a
ss
N
u
mb
e
r
o
f
i
n
st
a
n
c
e
s
Car
5
1
5
5
P
e
r
so
n
2
6
2
6
B
u
s
2
6
2
5
Tr
u
c
k
2
6
2
5
M
o
t
o
r
c
y
c
l
e
2
6
2
5
B
i
c
y
c
l
e
2
6
2
5
To
t
a
l
1
5
6
5
7
Fig
u
r
e
1
.
I
m
ag
es r
ep
r
esen
tin
g
d
if
f
er
en
t w
ea
th
e
r
co
n
d
itio
n
s
: c
lear
,
r
ain
,
f
o
g
,
s
n
o
w,
an
d
lo
w
l
ig
h
t
3
.
2
.
G
AN
–
YO
L
O
v
8
f
ra
m
ew
o
rk
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
u
s
es
a
h
y
b
r
id
GAN
–
YOL
Ov
8
ar
ch
itectu
r
e
to
m
ak
e
o
b
ject
d
etec
tio
n
m
o
r
e
r
eliab
le
in
b
ad
wea
th
e
r
.
T
h
e
f
r
am
ewo
r
k
h
as
two
clo
s
ely
co
n
n
ec
ted
p
a
r
ts
:
i)
a
GAN
-
b
ased
i
m
ag
e
en
h
an
ce
m
e
n
t
m
o
d
u
le,
a
n
d
ii)
a
YOL
Ov
8
o
b
ject
d
etec
tio
n
m
o
d
u
le.
Un
lik
e
tr
ad
itio
n
al
m
eth
o
d
s
th
at
h
an
d
le
p
r
ep
r
o
ce
s
s
in
g
an
d
d
etec
tio
n
s
ep
ar
ately
,
th
is
ap
p
r
o
ac
h
co
m
b
in
es
th
em
in
t
o
o
n
e
p
r
o
ce
s
s
.
T
h
e
GAN
is
tr
ain
ed
to
h
ig
h
lig
h
t
f
ea
tu
r
es
im
p
o
r
tan
t
f
o
r
d
etec
tio
n
,
n
o
t
ju
s
t
t
o
im
p
r
o
v
e
h
o
w
th
e
im
ag
e
lo
o
k
s
.
As
a
r
esu
lt,
th
e
en
h
a
n
ce
d
im
ag
es
h
elp
th
e
s
y
s
tem
d
etec
t
o
b
jects
m
o
r
e
ac
cu
r
ately
in
to
u
g
h
c
o
n
d
itio
n
s
lik
e
f
o
g
,
r
ain
,
s
n
o
w,
an
d
lo
w
l
ig
h
t.
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
2
2
3
0
-
2246
2234
T
h
r
ee
ess
en
tial
elem
en
ts
m
ak
e
u
p
th
e
YOL
Ov
8
m
o
d
el,
wh
ich
ad
h
er
es
to
th
e
co
n
v
e
n
tio
n
al
s
in
g
le
-
s
h
o
t
d
etec
tio
n
p
ar
a
d
ig
m
:
i
)
a
d
etec
tio
n
h
ea
d
f
o
r
class
if
icatio
n
an
d
l
o
ca
lizatio
n
;
ii)
a
PANet
n
ec
k
f
o
r
m
u
lti
-
s
ca
le
f
ea
tu
r
e
f
u
s
io
n
;
an
d
iii)
a
C
SP
Dar
k
Net
b
ac
k
b
o
n
e
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
[
2
5
]
.
Ou
r
m
eth
o
d
o
lo
g
y
c
r
ea
tes
an
in
teg
r
ated
en
h
a
n
ce
m
en
t
–
d
ete
ctio
n
p
ip
elin
e
b
y
in
s
er
tin
g
t
h
e
GAN
as
a
p
r
ep
r
o
ce
s
s
in
g
b
lo
ck
p
r
i
o
r
to
th
e
b
ac
k
b
o
n
e.
Fig
u
r
e
2
s
h
o
ws an
i
llu
s
tr
atio
n
o
f
th
e
wo
r
k
f
lo
w.
Fig
u
r
e
2.
W
o
r
k
f
lo
w
of
th
e
GA
N
–
YOL
Ov
8
d
etec
tio
n
p
i
p
e
l
i
n
e
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
s
tan
d
s
out
f
o
r
th
r
e
e
m
ain
r
e
a
s
o
n
s
:
a.
Dete
ctio
n
-
d
r
iv
en
GAN
o
p
tim
izatio
n
:
T
h
e
GAN
is
tr
ain
ed
to
en
h
an
ce
im
ag
es
an
d
also
to
b
o
o
s
t
o
b
ject
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
.
b.
Un
if
ied
m
u
lti
-
wea
th
er
f
r
a
m
e
wo
r
k
:
T
h
e
m
o
d
el
is
test
ed
i
n
s
ev
er
al
ty
p
es
o
f
b
ad
wea
t
h
er
,
wh
ile
m
o
s
t
ex
is
tin
g
m
eth
o
d
s
o
n
ly
h
a
n
d
le
o
n
e
s
ce
n
ar
io
.
c.
L
ig
h
tweig
h
t
in
teg
r
atio
n
with
YOL
Ov
8
:
T
h
e
GAN
m
o
d
u
le
i
m
p
r
o
v
es
i
n
p
u
t
q
u
ality
b
u
t
d
o
e
s
n
o
t
ad
d
m
u
c
h
co
m
p
u
tatio
n
al
l
o
ad
,
s
o
it k
ee
p
s
alm
o
s
t r
ea
l
-
tim
e
p
er
f
o
r
m
an
c
e.
Fin
d
in
g
s
in
[
2
4
]
,
[
2
6
]
,
w
h
er
e
p
r
ep
r
o
ce
s
s
in
g
en
h
a
n
ce
d
lo
w
-
q
u
ality
in
p
u
t
im
ag
es,
an
d
d
etec
to
r
-
ce
n
tr
ic
d
ev
elo
p
m
e
n
ts
lik
e
YOL
Ov
8
-
STE
[
2
3
]
,
wh
ich
illu
s
tr
ated
t
h
e
ad
v
an
tag
es
of
ar
ch
itectu
r
al
ch
an
g
es,
s
er
v
ed
as
th
e
im
p
etu
s
f
o
r
th
is
h
y
b
r
id
m
eth
o
d
.
B
y
c
o
m
b
in
i
n
g
th
e
r
o
b
u
s
t
b
aselin
e
p
er
f
o
r
m
an
ce
o
f
YOL
Ov
8
with
GAN
p
r
ep
r
o
ce
s
s
in
g
,
o
u
r
ap
p
r
o
ac
h
i
n
co
r
p
o
r
ates th
e
b
en
e
f
its
o
f
b
o
th
ap
p
r
o
ac
h
es.
B
y
co
m
b
in
in
g
GAN
with
YOL
Ov
8
,
f
ea
tu
r
e
d
ef
in
itio
n
is
im
p
r
o
v
e
d
an
d
wea
th
er
d
is
to
r
ti
o
n
-
in
d
u
ce
d
im
ag
e
d
eg
r
ad
atio
n
is
less
en
ed
.
Prio
r
to
p
ass
in
g
th
e
d
ata
i
n
to
YOL
Ov
8
f
o
r
d
etec
tio
n
,
t
h
e
GAN
co
m
p
o
n
en
t
en
h
an
ce
s
th
e
q
u
ality
of
t
h
e
im
ag
es.
T
h
e
GAN’
s
p
r
im
ar
y
tr
ai
n
in
g
g
o
al
is
ex
p
r
ess
ed
as
a
m
i
n
im
ax
o
p
tim
izatio
n
p
r
o
b
lem
i
n
v
o
lv
i
n
g
th
e
d
is
cr
im
in
ato
r
D
an
d
g
en
e
r
ato
r
G:
(
,
)
=
∼
(
)
[
(
)
]
+
∼
(
)
[
(
1
−
(
(
)
)
)
]
(
1
)
w
h
e
r
e
:
r
ep
r
esen
ts
r
ea
l
im
ag
e
s
am
p
les
f
r
o
m
th
e
d
ataset
(
e.
g
.
,
f
o
g
g
y
or
r
ain
y
f
r
a
m
e
s
)
.
is
a
n
o
is
e
v
ec
to
r
s
am
p
led
f
r
o
m
a
p
r
io
r
d
is
tr
ib
u
tio
n
(
)
.
(
)
g
en
er
ates
s
y
n
th
etic
wea
th
e
r
-
e
n
h
an
ce
d
i
m
a
g
e
s
.
(
)
o
u
tp
u
ts
th
e
p
r
o
b
a
b
ilit
y
th
at
th
e
in
p
u
t
is
r
ea
l
r
ath
er
th
a
n
g
e
n
e
r
a
t
e
d
.
W
h
ile
D
s
ee
k
s
to
m
ax
im
ize
th
is
v
alu
e
b
y
ac
cu
r
ately
d
if
f
er
en
tiatin
g
b
etwe
en
g
en
er
ated
a
n
d
g
e
n
u
in
e
im
a
g
e
s
,
G
s
e
e
k
s
to
lo
w
er
it
by
cr
e
at
in
g
r
e
al
i
s
ti
c
i
m
ag
e
s
t
h
at
ca
n
“
f
oo
l
”
D
.
T
h
e
vi
s
u
al
l
y
i
m
pr
o
v
e
d
f
r
a
m
e
s
p
r
o
d
u
ce
d
by
th
e
ad
v
er
s
ar
ial
o
p
tim
izatio
n
in
cr
ea
s
e
Y
OL
Ov
8
’
s
d
etec
tio
n
ac
cu
r
ac
y
,
p
ar
ticu
lar
l
y
in
lo
w
lig
h
t,
f
o
g
,
a
nd
r
ai
n
y
s
et
t
in
g
s
.
W
h
en
co
m
b
in
e
d
,
th
e
YOL
Ov
8
d
etec
tio
n
p
ip
elin
e
r
ec
eiv
e
s
th
e
GAN
-
en
h
an
ce
d
f
r
am
es
as
in
p
u
t,
ef
f
ec
tiv
ely
f
o
r
m
in
g
a
h
y
b
r
id
GAN
–
YOL
Ov
8
ar
ch
itectu
r
e
c
ap
ab
le
o
f
r
o
b
u
s
t
d
etec
tio
n
ac
r
o
s
s
d
iv
er
s
e
ad
v
er
s
e
en
v
i
r
o
n
m
e
n
t
s
.
3
.
2
.
1
.
G
AN
a
rc
hite
ct
ure
a
nd
t
ra
ini
ng
T
h
e
GAN
m
o
d
u
le
u
s
es
a
Pix
2
Pix
-
in
s
p
ir
ed
en
c
o
d
er
–
d
e
co
d
er
g
en
er
ato
r
ar
c
h
itectu
r
e
with
s
k
ip
co
n
n
ec
tio
n
s
.
E
f
f
ec
tiv
e
s
p
atia
l
d
etail
r
esto
r
atio
n
is
m
a
d
e
p
o
s
s
ib
le
b
y
th
e
g
en
e
r
ato
r
’
s
f
o
u
r
c
o
n
v
o
lu
tio
n
al
d
o
wn
s
am
p
lin
g
b
lo
ck
s
(
C
o
n
v
–
B
atch
No
r
m
–
R
eL
U)
an
d
f
o
u
r
u
p
s
am
p
lin
g
b
lo
c
k
s
(
T
r
an
s
p
o
s
ed
C
o
n
v
–
B
atch
No
r
m
–
R
eL
U
)
.
T
o
p
r
o
-
m
o
te
h
ig
h
-
f
r
e
q
u
en
c
y
r
ea
lis
m
,
th
e
d
is
cr
im
in
ato
r
u
s
es
a
Patch
GAN
s
tr
u
ctu
r
e
to
class
if
y
lo
ca
l im
ag
e
p
atch
es.
T
h
e
g
en
e
r
ato
r
is
o
p
tim
ized
u
s
in
g
a
weig
h
te
d
co
m
b
in
atio
n
o
f
ad
v
er
s
ar
ial
lo
s
s
,
L
1
r
ec
o
n
s
tr
u
c
tio
n
lo
s
s
,
an
d
p
er
ce
p
tu
al
lo
s
s
in
ad
d
itio
n
to
th
e
co
n
v
en
tio
n
al
ad
v
er
s
ar
ia
l lo
s
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
E
n
h
a
n
ce
men
t o
f YOLOv8
fo
r
o
b
ject
d
etec
tio
n
in
a
d
ve
r
s
e
w
ea
th
er c
o
n
d
itio
n
s
…
(
Ta
lifh
a
n
i
C
a
lvin
Ts
h
ip
o
ta
)
2235
=
+
1
1
+
2
(
2
)
wh
er
e
p
er
ce
p
tu
al
lo
s
s
is
ca
lcu
l
ated
u
s
in
g
f
ea
tu
r
e
m
ap
s
tak
en
f
r
o
m
a
p
r
etr
ain
ed
VGG
-
16
n
et
wo
r
k
,
an
d
L
1
lo
s
s
en
f
o
r
ce
s
s
tr
u
ctu
r
al
co
n
s
is
ten
cy
.
To
g
u
a
r
an
tee
s
tab
le
c
o
n
v
er
g
en
ce
,
t
h
e
GAN
is
f
r
o
ze
n
d
u
r
in
g
YOL
Ov
8
tr
ain
in
g
af
ter
b
ein
g
p
r
etr
ain
e
d
in
d
ep
en
d
en
tly
o
n
wea
th
er
-
d
eg
r
ad
ed
d
atasets
.
3
.
3
.
T
ra
ini
ng
co
nfig
ura
t
io
n
T
h
e
YOL
Ov
8
im
p
lem
en
tatio
n
f
r
o
m
Ultr
aly
tics
an
d
Py
T
o
r
ch
wer
e
u
s
ed
f
o
r
m
o
d
el
tr
ain
in
g
.
B
ased
on
em
p
ir
ical
tu
n
in
g
an
d
r
ec
o
m
m
en
d
atio
n
s
f
r
o
m
p
r
ev
io
u
s
r
esear
ch
,
h
y
p
er
p
ar
am
eter
s
wer
e
c
h
o
s
en
[
2
7
]
,
[
2
8
]
.
T
o
s
tr
ik
e
a
b
alan
ce
b
etwe
en
s
p
ee
d
an
d
p
r
ec
is
io
n
,
a
b
atc
h
s
ize
of
16,
a
lear
n
in
g
r
ate
of
0
.
0
1
an
d
an
in
p
u
t
r
eso
lu
tio
n
of
6
4
0
×6
4
0
we
r
e
u
s
ed
.
E
ac
h
m
o
d
el
was
tr
ain
e
d
u
s
in
g
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
with
m
o
m
en
tu
m
ac
r
o
s
s
1
0
0
e
p
o
c
h
s
.
T
h
e
YOL
Ov
8
tr
ain
in
g
l
o
s
s
c
o
m
b
in
es
box
r
eg
r
ess
io
n
,
o
b
je
ctn
ess
,
an
d
class
if
icatio
n
t
e
r
m
s
:
=
+
+
w
h
e
r
e
:
=
1
−
(
,
)
=
(
−
ˆ
)
2
=
=
1
lo
g
(
ˆ
)
Sin
ce
tr
an
s
f
er
lear
n
in
g
h
as
b
e
en
d
em
o
n
s
tr
ated
to
in
cr
ea
s
e
r
o
b
u
s
tn
ess
in
d
o
m
ain
-
s
h
if
t
s
ce
n
ar
io
s
,
p
r
e
-
tr
ain
ed
weig
h
ts
o
n
th
e
C
OC
O
d
ataset
wer
e
em
p
lo
y
ed
f
o
r
in
itializatio
n
in
o
r
d
e
r
to
s
p
ee
d
u
p
co
n
v
er
g
e
n
ce
an
d
av
o
id
o
v
er
f
itti
n
g
[
2
2
]
.
I
n
ac
c
o
r
d
an
ce
with
tactics
d
escr
ib
ed
i
n
[
2
1
]
,
[
2
4
]
,
d
ata
au
g
m
en
tatio
n
ap
p
r
o
ac
h
es
wer
e
u
s
ed
to
f
u
r
th
er
im
p
r
o
v
e
m
o
d
el
g
en
er
aliza
tio
n
.
T
h
ese
tech
n
iq
u
es
in
clu
d
ed
r
an
d
o
m
f
lip
s
,
s
ca
lin
g
,
an
d
wea
th
er
-
s
p
ec
if
ic
ch
an
g
es
(
b
lu
r
,
b
r
ig
h
tn
ess
r
ed
u
ctio
n
,
a
n
d
n
o
is
e
in
jectio
n
)
.
T
a
b
le
3
co
n
tain
th
e
en
tire
tr
ain
in
g
a
r
r
a
n
g
e
m
e
n
t
.
T
ab
le
3.
T
r
ain
i
n
g
h
y
p
er
p
ar
am
eter
s
f
o
r
YOL
Ov
8
an
d
GAN
–
YOL
Ov
8
m
o
d
e
l
s
H
y
p
e
r
p
a
r
a
m
e
t
e
r
Y
O
L
O
v
8
GAN
–
Y
O
L
O
v
8
Le
a
r
n
i
n
g
Rate
0
.
0
0
1
0
.
0
0
0
5
B
a
t
c
h
S
i
z
e
16
16
O
p
t
i
m
i
z
e
r
S
G
D
A
d
a
m
M
o
m
e
n
t
u
m
0
.
9
3
7
–
W
e
i
g
h
t
D
e
c
a
y
0
.
0
0
0
5
0
.
0
0
0
1
E
p
o
c
h
s
1
0
0
1
2
0
I
n
p
u
t
I
mag
e
S
i
z
e
6
4
0
×
640
6
4
0
×
640
D
a
t
a
A
u
g
m
e
n
t
a
t
i
o
n
M
o
s
a
i
c
,
F
l
i
p
,
H
S
V
M
o
s
a
i
c
,
F
l
i
p
,
H
S
V
,
GAN
-
b
a
s
e
d
e
n
h
a
n
c
e
m
e
n
t
Lo
ss
F
u
n
c
t
i
o
n
C
I
o
U
+
B
C
E
C
I
o
U
+
B
C
E
+
GAN
L
o
s
s
P
r
e
t
r
a
i
n
e
d
W
e
i
g
h
t
s
C
O
C
O
C
O
C
O
+
GAN
-
p
r
e
t
r
a
i
n
e
d
w
e
i
g
h
t
s
3
.
4
.
E
x
perim
ent
a
l
s
e
t
u
p
E
x
p
er
im
en
ts
u
tili
ze
d
a
wo
r
k
s
t
atio
n
f
ea
tu
r
in
g
an
NVI
DI
A
R
T
X
-
s
er
ies
GPU,
I
n
tel
C
o
r
e
i
7
p
r
o
ce
s
s
o
r
,
an
d
3
2
GB
R
AM
.
T
h
e
GAN
–
YOL
Ov
8
f
r
am
ew
o
r
k
was
im
p
lem
en
ted
in
Py
t
h
o
n
with
th
e
Py
T
o
r
ch
d
ee
p
lear
n
in
g
lib
r
a
r
y
an
d
tr
ain
e
d
u
s
in
g
th
e
Ultr
aly
tics
YOL
Ov
8
f
r
am
ewo
r
k
.
T
h
e
GAN
m
o
d
u
le
s
er
v
e
d
as
a
p
r
ep
r
o
ce
s
s
in
g
s
tag
e
to
en
h
an
ce
im
ag
es
d
eg
r
ad
e
d
b
y
ad
v
er
s
e
wea
th
er
co
n
d
itio
n
s
p
r
io
r
to
o
b
ject
d
e
tectio
n
with
YOL
Ov
8
.
E
x
p
er
im
en
tal
ev
alu
atio
n
em
p
l
o
y
ed
p
u
b
licly
av
ailab
l
e
d
atasets
,
in
clu
d
in
g
AC
DC
,
DAWN,
an
d
Fo
g
g
y
C
ity
s
ca
p
es,
wh
ich
en
co
m
p
ass
wea
th
er
co
n
d
itio
n
s
s
u
ch
as
r
ain
,
f
o
g
,
s
n
o
w,
an
d
l
o
w
-
lig
h
t e
n
v
ir
o
n
m
en
ts
.
T
r
ain
in
g
an
d
test
in
g
wer
e
co
n
d
u
cted
u
n
d
er
id
e
n
tical
ex
p
er
i
m
en
tal
s
ettin
g
s
to
en
s
u
r
e
f
air
co
m
p
ar
is
o
n
with
b
aselin
e
m
eth
o
d
s
.
T
h
e
Ad
am
o
p
tim
izer
was
em
p
lo
y
ed
with
an
in
itial
lear
n
in
g
r
ate
o
f
0
.
0
0
1
,
b
atch
s
ize
o
f
1
6
,
an
d
im
a
g
e
r
eso
lu
tio
n
o
f
6
4
0
×
6
4
0
p
ix
els.
Per
f
o
r
m
an
ce
was
ev
alu
ated
u
s
in
g
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e,
an
d
m
ea
n
av
er
ag
e
p
r
ec
is
io
n
(
m
AP@
0
.
5
).
Fig
u
r
e
3
p
r
esen
ts
th
e
f
u
ll
e
x
p
er
im
en
tal
wo
r
k
f
lo
w
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
Ad
v
er
s
e
wea
th
er
im
ag
es
f
r
o
m
d
atasets
s
u
ch
as
AC
DC
,
DAWN,
an
d
Fo
g
g
y
C
ity
s
ca
p
es
ar
e
in
itially
p
r
o
ce
s
s
ed
b
y
th
e
GAN
-
b
ased
en
h
an
ce
m
en
t
m
o
d
u
le
t
o
im
p
r
o
v
e
v
is
ib
ilit
y
an
d
f
ea
t
u
r
e
q
u
ality
.
T
h
e
r
esu
ltin
g
en
h
an
ce
d
im
a
g
es
ar
e
s
u
b
s
eq
u
en
tly
in
p
u
t
i
n
to
th
e
YOL
Ov
8
d
etec
to
r
f
o
r
o
b
ject
d
etec
tio
n
an
d
class
if
icatio
n
.
Dete
ctio
n
r
esu
lts
ar
e
th
en
ev
alu
ated
u
s
in
g
s
tan
d
ar
d
p
er
f
o
r
m
an
ce
m
etr
ics,
in
cl
u
d
in
g
Pre
cisi
o
n
,
R
ec
all,
F1
-
s
co
r
e,
an
d
m
AP@
0
.
5
.
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
2
2
3
0
-
2246
2236
Fig
u
r
e
3
.
E
x
p
e
r
im
en
tal
s
etu
p
a
n
d
wo
r
k
f
lo
w
of
th
e
p
r
o
p
o
s
ed
GAN
–
YOL
Ov
8
f
r
am
ewo
r
k
f
o
r
o
b
ject
d
etec
tio
n
u
n
d
er
a
d
v
er
s
e
wea
th
er
co
n
d
itio
n
s
3.
5
.
E
v
a
lua
t
i
o
n
pro
t
o
co
l
W
i
d
e
l
y
u
s
e
d
o
b
j
e
c
t
d
et
e
c
t
io
n
m
e
t
r
i
c
s
,
i
n
c
l
u
d
i
n
g
p
r
e
c
is
i
o
n
,
r
e
c
a
l
l
,
F
1
-
s
c
o
r
e
,
m
AP@
0
.
5
,
a
n
d
m
A
P@
0
.
5
–
0
.
9
5
,
wer
e
u
s
ed
to
ass
es
s
th
e
m
o
d
el’
s
p
er
f
o
r
m
a
n
ce
.
B
o
th
d
etec
tio
n
ac
c
u
r
ac
y
a
n
d
r
o
b
u
s
tn
ess
u
n
d
er
d
if
f
er
en
t
lo
ca
lizatio
n
c
r
iter
ia
ar
e
ca
p
tu
r
ed
by
th
ese
m
etr
ic
s
[
2
2
]
,
[
2
3
]
.
B
est
p
r
ac
tices
d
ev
elo
p
ed
in
ea
r
lier
b
en
ch
m
ar
k
in
g
s
tu
d
ies
of
Y
O
L
O
v
a
r
i
a
ti
o
n
s
u
n
d
e
r
u
n
f
a
v
o
r
a
b
le
w
e
at
h
e
r
c
i
r
c
u
m
s
t
a
n
c
es
[
2
1
]
,
[
2
3
]
,
a
n
d
[
2
4
]
w
e
r
e
a
d
h
e
r
e
d
to
by
t
h
e
e
v
al
u
a
tio
n
p
r
o
ce
s
s
.
Un
d
er
th
e
s
am
e
co
n
d
itio
n
s
,
th
e
b
aselin
e
YOL
Ov
8
an
d
th
e
s
u
g
g
ested
GAN
–
YOL
Ov
8
f
r
am
ewo
r
k
w
er
e
co
m
p
ar
e
d
.
Per
f
o
r
m
an
ce
b
en
ef
its
co
u
ld
b
e
d
ir
ec
tly
lin
k
e
d
to
th
e
s
u
g
g
ested
d
esig
n
th
an
k
s
to
ad
d
itio
n
al
b
en
ch
m
ar
k
i
n
g
ag
ai
n
s
t
p
u
b
lis
h
ed
f
in
d
in
g
s
of
YOL
Ov
8
-
STE
[
2
3
]
,
im
a
g
e
-
ad
a
p
tiv
e
YOL
O
(IA
-
YOL
O)
[
2
4
]
,
an
d
GAN
-
p
r
ep
r
o
ce
s
s
ed
p
ip
elin
es
[
2
6
]
,
[
2
9
]
.
T
h
e
f
o
llo
win
g
f
o
r
m
u
la
was
u
s
ed
to
d
eter
m
in
e
th
e
F1
-
s
co
r
e,
p
r
ec
is
io
n
(
P),
an
d
r
ec
all
(
R
)
:
We
ca
lcu
lated
th
e
m
ea
n
Av
e
r
a
g
e
Pre
cisi
o
n
(
m
AP)
f
o
r
I
o
U
th
r
esh
o
ld
s
b
etwe
en
0
.
5
an
d
0
.
9
5
:
=
+
,
=
+
,
1
=
2
×
×
+
(
3
)
W
e
ca
lcu
lated
th
e
m
ea
n
av
e
r
a
g
e
p
r
ec
is
io
n
(
m
AP)
f
o
r
I
o
U
th
r
esh
o
ld
s
b
etwe
en
0
.
5
an
d
0
.
9
5
:
m
AP
=
1
∑
=
1
,
=
∫
(
)
1
0
(
4
)
T
h
is
s
tatis
tic
m
ea
s
u
r
es
th
e
m
o
d
el’
s
d
etec
tio
n
q
u
ality
a
n
d
s
h
o
ws
th
e
ar
ea
u
n
d
er
t
h
e
p
r
ec
is
io
n
–
r
ec
all
cu
r
v
e.
C
o
n
f
i
d
en
ce
lo
s
s
co
n
v
er
g
en
ce
,
b
o
x
lo
s
s
,
an
d
class
if
ica
tio
n
lo
s
s
wer
e
u
s
ed
to
tr
ac
k
th
e
tr
ain
in
g
s
tab
ilit
y
.
A
d
d
i
t
i
o
n
a
l
l
y
,
to
e
v
al
u
a
t
e
t
h
e
GA
N
–
Y
O
L
O
v
8
f
r
a
m
e
w
o
r
k
’
s
v
ia
b
i
l
i
t
y
in
r
e
al
-
t
i
m
e
,
i
n
f
e
r
e
n
c
e
t
im
e
a
n
d
FP
S
(
f
r
a
m
es
p
er
s
ec
o
n
d
)
wer
e
ass
ess
ed
.
3
.
6
.
H
a
rdwa
re
a
nd
s
o
f
t
wa
re
e
n
v
i
r
o
n
men
t
E
v
e
r
y
e
x
p
e
r
i
m
e
n
t
i
n
t
h
is
s
t
u
d
y
w
a
s
c
o
n
d
u
c
t
e
d
o
n
a
h
i
g
h
-
p
e
r
f
o
r
m
a
n
c
e
w
o
r
k
s
t
a
ti
o
n
t
h
a
t
h
as
a
n
I
n
t
e
l®
X
e
o
n
®
S
il
v
e
r
4
2
1
0
R
C
P
U
r
u
n
n
i
n
g
a
t
2
.
4
0
G
H
z
a
n
d
a
n
N
V
I
D
I
A
T
e
s
l
a
T
4
G
PU
(
1
6
G
B
VR
A
M
)
,
i
n
a
d
d
it
i
o
n
t
o
6
4
G
B
o
f
R
A
M
.
W
i
t
h
C
UD
A
1
2
.
1
a
n
d
c
u
D
N
N
8
.
9
i
n
s
t
a
ll
e
d
o
n
t
h
e
6
4
-
b
i
t
U
b
u
n
t
u
2
0
.
0
4
L
T
S
c
o
m
p
u
t
e
r
s
y
s
t
e
m
,
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
l
i
c
at
i
o
n
s
w
er
e
e
f
f
e
c
t
i
v
e
l
y
ac
c
e
l
e
r
a
t
e
d
b
y
t
h
e
G
PU
.
T
h
r
o
u
g
h
o
u
t
t
h
e
m
o
d
e
l
’
s
t
r
a
i
n
i
n
g
a
n
d
a
s
s
e
s
s
m
e
n
t
p
h
a
s
e
s
,
t
h
i
s
c
o
n
f
i
g
u
r
a
t
i
o
n
g
u
a
r
a
n
t
e
e
d
p
e
a
k
p
e
r
f
o
r
m
a
n
c
e
a
n
d
c
o
n
s
i
s
t
e
n
t
p
r
o
c
es
s
i
n
g
.
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
E
n
h
a
n
ce
men
t o
f YOLOv8
fo
r
o
b
ject
d
etec
tio
n
in
a
d
ve
r
s
e
w
ea
th
er c
o
n
d
itio
n
s
…
(
Ta
lifh
a
n
i
C
a
lvin
Ts
h
ip
o
ta
)
2237
T
h
e
s
o
f
t
w
a
r
e
e
n
v
i
r
o
n
m
e
n
t
wa
s
b
u
i
l
t
wi
t
h
P
y
t
h
o
n
3
.
1
0
,
a
n
d
P
y
T
o
r
c
h
2
.
1
.
0
a
n
d
t
h
e
U
l
t
r
al
y
t
i
cs
Y
O
L
O
v
8
f
r
a
m
e
w
o
r
k
(
v
e
r
s
i
o
n
8
.
1
.
0
)
w
e
r
e
u
t
i
li
z
e
d
t
o
d
e
v
el
o
p
t
h
e
m
o
d
el
.
T
h
e
m
a
j
o
r
it
y
o
f
t
h
e
r
es
e
a
r
c
h
w
a
s
c
o
n
d
u
c
t
e
d
i
n
t
h
e
G
o
o
g
l
e
C
o
l
a
b
P
r
o
+
e
n
v
i
r
o
n
m
e
n
t
,
w
h
i
c
h
o
f
f
e
r
s
r
e
li
a
b
l
e
GP
U
a
cc
e
s
s
a
n
d
m
a
k
e
s
i
t
p
o
s
s
i
b
l
e
t
o
r
e
p
e
at
r
es
u
l
ts
a
c
r
o
s
s
m
u
l
t
i
p
l
e
r
u
n
s
.
M
a
t
p
l
o
tl
i
b
(
v
3
.
7
)
w
as
u
s
e
d
f
o
r
d
a
t
a
v
i
s
u
a
li
z
a
ti
o
n
a
n
d
c
h
a
r
ti
n
g
,
N
u
m
P
y
(
v
1
.
2
5
)
f
o
r
n
u
m
e
r
i
c
a
l
c
o
m
p
u
t
a
t
i
o
n
s
,
a
n
d
Op
e
n
C
V
(
v
4
.
8
)
f
o
r
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
i
n
o
r
d
e
r
t
o
i
n
c
r
e
as
e
w
o
r
k
f
lo
w
e
f
f
i
c
i
e
n
c
y
.
T
h
i
s
c
o
m
b
i
n
e
d
h
a
r
d
wa
r
e
-
s
o
f
tw
a
r
e
a
r
c
h
it
e
c
t
u
r
e
a
l
l
o
ws
f
o
r
q
u
i
c
k
t
r
a
i
n
i
n
g
o
f
b
o
t
h
Y
O
L
O
v
8
a
n
d
G
A
N
-
Y
O
L
O
v
8
m
o
d
e
ls
i
n
a
r
a
n
g
e
o
f
w
e
a
t
h
e
r
-
d
e
g
r
a
d
e
d
c
o
n
d
i
ti
o
n
s
b
y
o
f
f
e
r
i
n
g
a
b
a
l
a
n
c
e
d
c
o
m
b
i
n
a
t
i
o
n
o
f
p
r
o
c
e
s
s
i
n
g
p
o
w
e
r
a
n
d
s
c
a
l
a
b
il
i
t
y
.
A
d
d
i
ti
o
n
a
l
l
y
,
c
o
n
s
i
s
t
e
n
c
y
,
r
e
p
r
o
d
u
c
i
b
il
i
t
y
,
a
n
d
a
l
i
g
n
m
e
n
t
w
it
h
s
t
at
e
-
of
-
t
h
e
-
a
r
t
c
o
m
p
u
t
e
r
v
i
s
i
o
n
r
e
s
e
a
r
c
h
p
r
o
c
e
d
u
r
e
s
we
r
e
g
u
a
r
a
n
t
e
e
d
b
y
t
h
e
u
s
e
o
f
s
t
a
n
d
a
r
d
i
z
e
d
f
r
a
m
e
w
o
r
k
s
a
n
d
G
P
U
-
a
c
c
e
le
r
a
t
e
d
l
i
b
r
a
r
i
es
.
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SU
L
T
S
AND
DI
SC
US
SI
O
N
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
is
a
G
AN
-
e
n
h
a
n
c
e
d
Y
O
L
O
v
8
o
b
j
e
c
t
r
e
c
o
g
n
i
t
i
o
n
a
n
d
i
d
e
n
ti
f
i
c
a
ti
o
n
f
r
a
m
e
w
o
r
k
,
a
n
d
i
t
w
a
s
r
i
g
o
r
o
u
s
l
y
t
e
s
t
e
d
o
n
a
v
a
r
i
e
t
y
o
f
m
u
l
t
i
-
w
e
a
t
h
e
r
d
a
t
a
s
et
s
,
i
n
cl
u
d
i
n
g
l
o
w
-
l
i
g
h
t
,
d
a
m
p
,
f
o
g
g
y
,
s
n
o
w
y
,
a
n
d
c
l
e
a
r
we
a
t
h
e
r
c
o
n
d
i
ti
o
n
.
T
h
e
s
e
s
c
e
n
a
r
i
o
s
w
e
r
e
s
p
e
ci
f
i
c
al
ly
s
e
l
e
ct
e
d
b
e
c
a
u
s
e
t
h
e
y
s
h
o
ws
s
o
m
e
o
f
t
h
e
m
o
s
t
i
m
p
o
r
t
a
n
t
is
s
u
es
i
n
i
n
t
el
l
i
g
e
n
t
s
u
r
v
e
il
l
a
n
c
e
,
a
u
t
o
n
o
m
o
u
s
d
r
i
v
i
n
g
,
a
n
d
s
m
a
r
t
t
r
a
n
s
p
o
r
t
a
t
io
n
s
y
s
t
e
m
s
,
w
h
e
r
e
u
n
p
r
e
d
i
c
t
a
b
l
e
i
ll
u
m
i
n
a
t
i
o
n
a
n
d
w
e
a
t
h
e
r
f
r
e
q
u
e
n
t
l
y
i
m
p
a
i
r
th
e
a
c
c
u
r
a
c
y
o
f
o
b
j
e
c
t
d
e
t
e
ct
i
o
n
a
n
d
i
d
e
n
t
i
f
i
c
a
ti
o
n
[
1
4
]
,
[
1
5
]
.
U
n
l
i
k
e
s
t
a
n
d
a
r
d
e
v
a
l
u
a
t
i
o
n
m
e
t
h
o
d
s
t
h
a
t
m
o
s
t
l
y
f
o
c
u
s
o
n
c
l
e
a
r
-
we
a
t
h
e
r
m
e
t
r
i
c
s
,
o
u
r
s
t
u
d
y
s
p
e
c
i
f
i
ca
l
l
y
a
d
d
r
es
s
es
a
d
v
e
r
s
e
s
c
e
n
a
r
i
o
s
t
o
p
r
o
v
i
d
e
a
m
o
r
e
a
cc
u
r
a
t
e
m
e
a
s
u
r
e
o
f
r
es
i
li
e
n
c
e
a
n
d
a
d
a
p
t
a
t
i
o
n
.
B
y
c
o
m
b
i
n
i
n
g
q
u
a
n
ti
t
a
ti
v
e
a
n
d
q
u
a
l
i
t
a
ti
v
e
v
i
ew
p
o
i
n
t
s
,
t
h
e
ev
a
l
u
a
t
i
o
n
m
e
t
h
o
d
w
a
s
c
r
e
a
te
d
t
o
o
f
f
e
r
a
c
o
m
p
r
e
h
e
n
s
i
v
e
e
v
a
l
u
at
i
o
n
o
f
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
.
T
o
ass
e
s
s
d
e
t
e
ct
i
o
n
a
c
c
u
r
a
c
y
u
n
d
e
r
d
e
f
e
r
e
n
t
e
n
v
i
r
o
n
m
e
n
t
al
w
e
a
t
h
e
r
c
o
n
d
it
i
o
n
s
,
s
ta
n
d
a
r
d
m
e
t
r
i
c
s
s
u
c
h
a
s
p
r
e
c
is
i
o
n
,
r
e
ca
l
l
,
F
1
-
s
c
o
r
e
,
a
n
d
m
e
a
n
a
v
e
r
ag
e
p
r
e
c
i
s
i
o
n
(
m
AP
)
w
e
r
e
c
a
l
c
u
l
at
e
d
.
B
e
ca
u
s
e
t
h
ey
r
e
f
l
e
c
t
t
r
a
d
e
-
o
f
f
s
b
e
t
w
e
e
n
f
a
l
s
e
p
o
s
it
i
v
es
a
n
d
f
a
l
s
e
n
e
g
a
ti
v
e
s
i
n
a
v
a
r
i
e
t
y
o
f
c
o
n
d
i
t
i
o
n
s
,
t
h
es
e
c
r
i
te
r
i
a
h
a
v
e
b
e
c
o
m
e
s
t
a
n
d
a
r
d
i
n
e
v
a
l
u
a
ti
n
g
c
o
n
t
e
m
p
o
r
a
r
y
d
e
e
p
-
l
e
a
r
n
i
n
g
d
e
t
e
c
t
o
r
s
[
1
]
,
[
3
]
.
W
e
c
o
n
d
u
c
t
e
d
a
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
is
w
it
h
s
t
at
e
-
of
-
t
h
e
-
a
r
t
m
o
d
e
ls
(
e
.
g
.
,
Y
O
L
O
v
5
,
Y
O
L
O
v
7
,
F
ast
e
r
R
-
C
NN
)
i
n
o
r
d
e
r
t
o
f
u
r
t
h
e
r
c
o
n
t
e
x
t
u
a
l
i
z
e
o
u
r
f
i
n
d
i
n
g
s
w
i
t
h
i
n
t
h
e
b
r
o
a
d
e
r
r
e
s
e
a
r
c
h
f
r
a
m
e
w
o
r
k
.
W
h
i
l
e
e
m
p
h
a
s
i
z
i
n
g
t
h
e
a
d
v
a
n
t
a
g
es
o
f
t
h
e
p
r
o
p
o
s
e
d
p
a
r
a
d
i
g
m
i
n
m
a
n
a
g
i
n
g
d
i
v
e
r
s
e
w
e
a
t
h
e
r
c
o
n
d
i
tio
n
s
,
t
h
i
s
e
v
a
l
u
at
i
o
n
r
e
c
o
g
n
i
z
e
d
t
h
e
d
r
a
w
b
a
c
k
s
o
f
s
p
e
c
i
a
l
i
ze
d
a
p
p
r
o
a
c
h
e
s
[
3
0
]
,
[
3
1
]
.
I
n
a
d
d
i
t
i
o
n
t
o
n
u
m
e
r
i
c
al
d
a
t
a
,
tr
a
i
n
i
n
g
d
i
a
g
n
o
s
t
ic
s
w
e
r
e
a
n
al
y
z
e
d
t
o
c
h
a
r
a
c
te
r
i
z
e
t
h
e
m
o
d
e
l
’
s
b
e
h
a
v
i
o
r
d
u
r
i
n
g
o
p
t
i
m
i
z
a
t
i
o
n
.
C
o
n
v
e
r
g
e
n
c
e
s
t
a
b
il
i
t
y
,
p
e
r
-
cl
a
s
s
p
er
f
o
r
m
a
n
c
e
v
a
r
i
a
b
i
l
i
t
y
,
a
n
d
co
n
f
i
d
e
n
c
e
-
t
h
r
e
s
h
o
l
d
s
e
n
s
i
t
i
v
it
y
w
e
r
e
g
i
v
e
n
s
p
e
c
i
al
c
o
n
s
i
d
e
r
a
t
i
o
n
.
B
e
c
a
u
s
e
t
h
e
y
d
e
m
o
n
s
t
r
a
t
e
w
h
e
t
h
e
r
i
m
p
r
o
v
e
m
e
n
t
s
a
r
e
at
t
r
i
b
u
t
a
b
le
t
o
g
e
n
e
r
a
l
r
e
s
i
li
e
n
c
e
o
r
o
v
e
r
f
i
t
ti
n
g
t
o
s
p
e
c
i
f
i
c
c
o
n
d
i
t
i
o
n
s
,
t
h
e
s
e
d
i
a
g
n
o
s
t
i
c
i
n
s
i
g
h
t
s
a
r
e
i
m
p
o
r
t
a
n
t
[
2
7
]
,
[
1
7
]
.
T
h
e
m
o
d
e
l
’
s
a
b
i
l
it
y
t
o
h
a
n
d
l
e
r
e
a
l
-
w
o
r
l
d
p
r
o
b
l
e
m
s
i
n
c
l
u
d
i
n
g
m
o
t
io
n
b
l
u
r
,
l
i
m
i
t
e
d
e
y
e
s
i
g
h
t
,
s
e
n
s
o
r
n
o
i
s
e
,
a
n
d
p
a
r
t
i
a
l
o
c
c
l
u
s
i
o
n
w
as
f
i
n
a
ll
y
s
h
o
w
e
r
e
d
b
y
t
h
e
v
i
s
u
a
l
d
e
t
e
c
ti
o
n
r
e
s
u
l
t
s
o
f
a
q
u
a
l
i
t
at
i
v
e
t
es
t
.
T
h
es
e
v
i
s
u
al
e
v
a
l
u
a
ti
o
n
p
r
o
v
i
d
e
h
e
l
p
f
u
l
c
o
n
f
i
r
m
a
t
i
o
n
o
f
t
h
e
q
u
a
n
t
i
t
a
t
i
v
e
m
e
as
u
r
e
s
an
d
b
o
l
s
t
e
r
t
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
’
s
a
p
p
l
i
c
a
b
il
i
t
y
i
n
a
r
a
n
g
e
o
f
o
p
e
r
a
t
i
o
n
a
l
c
o
n
t
e
x
t
s
[
9
]
,
[
1
9
]
.
E
v
e
r
y
e
x
p
e
r
i
m
e
n
t
w
a
s
c
o
n
d
u
c
te
d
t
h
r
e
e
t
i
m
e
s
wi
t
h
a
d
i
f
f
e
r
e
n
t
s
e
t
o
f
r
a
n
d
o
m
s
e
e
d
s
.
T
h
e
r
e
p
o
r
t
e
d
r
e
s
u
l
ts
m
a
t
c
h
t
h
e
m
e
a
n
p
e
r
f
o
r
m
a
n
c
e
,
a
n
d
a
l
l
m
e
t
r
i
c
s
s
h
o
w
s
ta
b
l
e
a
n
d
c
o
n
s
i
s
t
e
n
t
c
o
n
v
e
r
g
e
n
c
e
,
w
i
t
h
s
t
a
n
d
a
r
d
d
e
v
i
at
i
o
n
s
s
t
a
y
i
n
g
b
e
l
o
w
±
1
.
2
%
.
4
.
1
.
Q
u
a
n
t
i
t
a
t
i
v
e
p
e
r
f
o
r
ma
nc
e
un
d
e
r
we
a
t
h
e
r
c
o
n
di
t
i
o
ns
T
h
e
p
r
o
p
o
s
ed
GAN
-
en
h
a
n
ce
d
YOL
Ov
8
d
etec
tio
n
s
y
s
tem
’
s
r
o
b
u
s
tn
ess
an
d
g
en
e
r
aliza
tio
n
ca
p
ac
ity
wer
e
r
ig
o
r
o
u
s
ly
ev
alu
ated
in
f
i
v
e
d
if
f
er
en
t
en
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
:
clea
r
,
r
ain
y
,
f
o
g
g
y
,
s
n
o
wy
,
an
d
lo
w
-
lig
h
t.
C
o
m
m
o
n
o
b
ject
d
etec
tio
n
an
d
id
en
tific
atio
n
m
etr
ics
u
s
ed
in
th
e
ass
ess
m
en
t
wer
e
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
m
ea
n
av
er
a
g
e
p
r
ec
is
io
n
at
I
o
U
t
h
r
e
s
h
o
l
d
0
.
5
(
m
AP@
0
.
5
),
a
n
d
m
ea
n
Av
er
ag
e
Pre
cisi
o
n
ac
r
o
s
s
I
o
U
th
r
esh
o
l
d
s
f
r
o
m
0
.
5
to
0
.
9
5
(
m
AP@
0
.
5
:0
.
9
5
)
.
T
h
ese
m
etr
ics
wer
e
c
h
o
s
en
b
ec
au
s
e
th
e
y
ca
p
tu
r
e
b
o
th
class
if
icatio
n
r
eliab
ilit
y
an
d
lo
ca
lizatio
n
a
cc
u
r
ac
y
(
I
o
U
-
b
ased
p
e
r
f
o
r
m
a
n
ce
)
,
o
f
f
er
in
g
a
f
air
ev
alu
ati
o
n
o
f
th
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
u
n
d
er
d
if
f
icu
lt w
ea
th
er
co
n
d
itio
n
s
[
3
2
]
,
[
2
7
]
.
T
ab
le
4
s
h
o
ws
th
e
m
o
d
el’
s
p
ea
k
p
e
r
f
o
r
m
an
ce
with
an
F1
-
s
co
r
e
o
f
0
.
8
9
an
d
m
AP@
0
.
5
=0
.
9
2
,
s
h
o
win
g
th
e
ar
ch
itectu
r
e
’
s
ef
f
ec
tiv
en
ess
u
n
d
er
id
ea
l
co
n
d
itio
n
.
C
lear
wea
th
er
co
n
d
itio
n
s
er
v
ed
as
a
b
aselin
e.
As
ex
p
ec
ted
,
p
er
f
o
r
m
an
ce
d
e
clin
ed
u
n
d
er
a
d
v
er
s
e
co
n
d
itio
n
s
,
with
th
e
lar
g
est
d
ec
lin
e
o
cc
u
r
r
in
g
at
lo
w
lig
h
t
lev
els
co
n
d
itio
n
s
(
m
AP@
0
.
5
=
0
.
8
1
,
m
AP@
0
.
5
:0
.
9
5
=
0
.
5
4
)
.
T
h
is
ex
em
p
lifie
s
th
e
well
-
k
n
o
wn
v
u
ln
er
a
b
ilit
y
o
f
o
b
ject
d
etec
tio
n
an
d
id
en
tific
atio
n
to
in
ad
eq
u
ate
illu
m
in
ati
o
n
,
wh
ich
r
e
d
u
ce
s
th
e
d
e
p
en
d
ab
ilit
y
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
h
in
d
er
s
p
r
ec
is
e
o
b
ject
b
o
r
d
er
r
ec
o
g
n
itio
n
[
6
]
.
Nev
er
th
eless
,
th
e
m
o
d
el
d
id
r
ath
er
well
in
r
ain
y
en
v
i
r
o
n
m
en
t
(
m
AP@
0
.
5
=0
.
8
8
)
,
d
e
s
p
ite
s
lig
h
t
p
r
ec
is
io
n
–
r
ec
all
tr
ad
e
-
o
f
f
s
ca
u
s
ed
b
y
m
o
tio
n
b
l
u
r
,
wate
r
d
r
o
p
lets
,
an
d
r
ef
lectio
n
s
.
C
lo
u
d
y
co
n
d
itio
n
s
also
d
ec
r
ea
s
ed
co
n
tr
ast
an
d
ey
esig
h
t,
wh
ich
led
to
a
r
ec
all
o
f
0
.
7
8
.
Similar
d
if
f
icu
lties
wer
e
o
b
s
er
v
ed
in
s
n
o
wy
en
v
ir
o
n
m
en
ts
,
wh
er
e
t
h
e
ac
cu
r
ac
y
o
f
b
o
u
n
d
i
n
g
b
o
x
l
o
ca
lizatio
n
was
d
ec
r
ea
s
ed
b
y
p
ar
tial
o
cc
lu
s
io
n
an
d
d
en
s
e
s
n
o
wf
all.
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.
1
6
,
No
.
4
,
Au
g
u
s
t
20
2
6
:
2
2
3
0
-
2246
2238
T
ab
le
4.
YOL
Ov
8
p
er
f
o
r
m
an
c
e
ac
r
o
s
s
wea
th
er
c
o
n
d
i
t
i
o
n
s
W
e
a
t
h
e
r
C
o
n
d
i
t
i
o
n
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
S
c
o
r
e
mA
P
@
0
.
5
mA
P
@
0
.
5
:
0
.
9
5
N
o
t
e
s
C
l
e
a
r
0
.
9
0
0
.
8
8
0
.
8
9
0
.
9
2
0
.
6
5
B
a
se
l
i
n
e
o
r
o
p
t
i
ma
l
s
c
e
n
a
r
i
o
.
R
a
i
n
0
.
8
5
0
.
8
2
0
.
8
3
0
.
8
8
0
.
6
0
S
l
i
g
h
t
p
e
r
f
o
r
m
a
n
c
e
d
r
o
p
d
u
e
t
o
m
o
t
i
o
n
b
l
u
r
a
n
d
r
e
f
l
e
c
t
i
o
n
s.
F
o
g
0
.
8
1
0
.
7
8
0
.
7
9
0
.
8
4
0
.
5
8
R
e
d
u
c
e
d
v
i
s
i
b
i
l
i
t
y
d
e
c
r
e
a
s
e
s
d
e
t
e
c
t
i
o
n
a
c
c
u
r
a
c
y
.
S
n
o
w
0
.
8
2
0
.
7
6
0
.
7
9
0
.
8
3
0
.
5
6
S
n
o
w
f
l
a
k
e
s
i
n
t
r
o
d
u
c
e
n
o
i
se
a
n
d
p
a
r
t
i
a
l
o
b
j
e
c
t
o
c
c
l
u
si
o
n
.
Lo
w
L
i
g
h
t
0
.
8
0
0
.
7
4
0
.
7
7
0
.
8
1
0
.
5
4
P
o
o
r
i
l
l
u
mi
n
a
t
i
o
n
n
e
g
a
t
i
v
e
l
y
a
f
f
e
c
t
s
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
ma
n
c
e
.
Desp
ite
th
ese
d
ip
s
,
it
is
im
p
o
r
tan
t
th
at
p
r
ec
is
io
n
v
alu
es
co
n
s
is
ten
tly
r
e
m
ain
o
v
er
0
.
8
0
in
all
s
ce
n
ar
io
s
,
d
em
o
n
s
tr
atin
g
th
e
m
o
d
el’
s
r
esil
ien
ce
to
o
cc
lu
s
io
n
,
n
o
is
e,
b
lu
r
,
an
d
lo
w
v
is
ib
il
ity
.
E
v
en
wh
ile
th
e
m
o
d
el
m
a
y
s
o
m
etim
es
m
is
s
d
etec
tio
n
s
(
lo
wer
r
ec
all)
,
th
is
c
o
n
s
is
ten
cy
s
h
o
ws
th
at
it
m
ain
t
ain
s
a
r
elativ
ely
lo
w
f
alse
-
p
o
s
itiv
e
r
ate,
g
u
a
r
an
teei
n
g
r
eliab
le
p
e
r
f
o
r
m
an
ce
u
n
d
er
a
r
an
g
e
o
f
s
ce
n
ar
i
o
s
(
1
)
.
T
h
ese
f
in
d
in
g
s
s
h
o
ws
th
at,
in
d
if
f
e
r
en
ce
to
c
o
n
v
e
n
tio
n
al
YOL
O
-
b
ased
m
o
d
els,
th
e
in
clu
s
io
n
o
f
GAN
-
b
ased
p
r
e
-
p
r
o
ce
s
s
in
g
e
n
h
an
ce
s
th
e
d
etec
tio
n
p
i
p
elin
e’
s
r
esil
ien
ce
,
p
ar
ticu
lar
ly
u
n
d
er
u
n
c
er
tain
an
d
n
o
is
y
en
v
ir
o
n
m
en
t
[
3
3
]
,
[
3
4
]
.
Ov
er
all,
th
e
m
ath
em
atica
l
a
n
aly
s
is
s
h
o
ws
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
b
alan
ce
s
r
o
b
u
s
tn
ess
an
d
ad
ap
tab
ilit
y
,
m
ak
in
g
it
p
er
f
ec
t
f
o
r
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
in
a
u
to
n
o
m
o
u
s
d
r
iv
i
n
g
,
tr
a
f
f
ic
m
o
n
ito
r
in
g
,
an
d
s
m
ar
t su
r
v
eill
an
ce
s
y
s
tem
s
.
T
h
e
ab
latio
n
r
esu
lts
in
T
ab
l
e
5
s
h
o
w
th
at
tr
ad
itio
n
al
d
a
ta
au
g
m
en
tatio
n
b
y
its
elf
o
f
f
er
s
litt
l
e
r
esil
ien
ce
in
b
ad
wea
th
er
.
G
AN
-
b
ased
p
r
e
p
r
o
ce
s
s
in
g
,
on
th
e
o
th
e
r
h
a
n
d
,
r
esu
lts
in
s
ig
n
if
ican
t
p
er
f
o
r
m
a
n
ce
g
ain
s
,
in
d
icatin
g
th
at
th
e
im
p
r
o
v
em
e
n
ts
ar
e
m
ain
l
y
d
u
e
t
o
im
p
r
o
v
ed
in
p
u
t
q
u
ality
r
at
h
er
th
an
ju
s
t
m
o
r
e
d
iv
er
s
e
d
ata.
T
ab
le
5.
Ab
latio
n
s
tu
d
y
on
GA
N
p
r
e
p
r
o
c
e
s
s
i
n
g
M
o
d
e
l
V
a
r
i
a
n
t
mA
P
@
0
.
5
F1
-
S
c
o
r
e
Y
O
LO
v
8
(
B
a
s
e
l
i
n
e
)
0
.
8
4
0
.
8
1
Y
O
LO
v
8
+
D
a
t
a
A
u
g
me
n
t
a
t
i
o
n
0
.
8
7
0
.
8
3
Y
O
LO
v
8
+
G
A
N
(
P
r
o
p
o
se
d
)
0
.
9
0
0
.
8
6
T
h
e
s
u
g
g
ested
f
r
a
m
ewo
r
k
m
ain
tain
s
n
ea
r
r
ea
l
-
tim
e
p
e
r
f
o
r
m
a
n
ce
(
3
0
FP
S),
wh
ich
m
ak
es
it
ap
p
r
o
p
r
iate
f
o
r
p
r
ac
tical
d
ep
l
o
y
m
en
t
in
in
tellig
en
t
tr
a
n
s
p
o
r
ta
tio
n
an
d
s
u
r
v
eillan
ce
s
y
s
tem
s
ev
en
th
o
u
g
h
GAN
p
r
e
-
p
r
o
ce
s
s
in
g
ad
d
s
ex
tr
a
c
o
m
p
u
tatio
n
al
o
v
er
h
ea
d
.
T
h
e
p
r
o
p
o
s
ed
s
tr
u
ctu
r
e
m
ain
tai
n
s
n
ea
r
r
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
(
3
0
FP
S)
d
esp
ite
th
e
ex
tr
a
GAN
p
r
ep
r
o
ce
s
s
in
g
s
tep
,
s
u
g
g
esti
n
g
th
at
it
is
s
u
itab
le
f
o
r
p
r
ac
tical
im
p
lem
en
tatio
n
in
au
t
o
n
o
m
o
u
s
d
r
iv
in
g
an
d
s
u
r
v
eillan
ce
s
y
s
tem
s
.
T
h
e
in
f
er
en
ce
-
tim
e
p
er
f
o
r
m
an
ce
is
s
h
o
wn
in
T
ab
le
6
.
T
ab
le
6.
I
n
f
e
r
en
ce
-
tim
e
p
e
r
f
o
r
m
a
n
c
e
M
o
d
e
l
FPS
I
n
f
e
r
e
n
c
e
Ti
me
(
ms)
Y
O
LO
v
8
(
B
a
s
e
l
i
n
e
)
52
1
9
.
2
Y
O
LO
v
8
+
G
A
N
(
P
r
o
p
o
se
d
)
38
2
6
.
0
4
.
2
.
C
o
mp
a
r
a
t
i
v
e
e
v
a
l
u
a
t
i
o
n
o
f
y
o
l
o
v
8
b
a
s
e
d
wi
t
h
r
el
a
t
e
d
w
o
r
ks
I
n
o
r
d
er
f
o
r
u
s
to
d
eter
m
in
e
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
e
d
YOL
Ov
8
-
b
ased
m
o
d
el
im
p
r
o
v
ed
with
GAN
p
r
e
-
p
r
o
ce
s
s
in
g
,
we
m
ea
s
u
r
ed
its
p
er
f
o
r
m
a
n
ce
to
d
ef
er
e
n
t
co
n
tem
p
o
r
ar
y
d
etec
tio
n
alg
o
r
ith
m
s
d
esig
n
ed
o
r
ad
ju
s
ted
f
o
r
a
d
v
er
s
e
wea
th
er
c
o
n
d
itio
n
s
.
T
a
b
le
7
h
i
g
h
lig
h
t th
e
s
u
m
m
ar
y
o
f
th
is
co
m
p
ar
ativ
e
s
tu
d
y
.
T
ab
le
7
.
C
o
m
p
ar
ativ
e
an
aly
s
is
with
r
elate
d
s
t
u
d
i
e
s
S
t
u
d
y
M
e
t
h
o
d
W
e
a
t
h
e
r
F
o
c
u
s
m
AP
@
0
.
5
N
o
t
e
s
F
an
g
et
a
l
.
[
2
7
]
F
a
st
e
r
R
-
C
N
N
F
o
g
,
R
a
i
n
0
.
8
9
(f
o
g
),
0
.
8
2
(
r
a
i
n
)
S
t
ron
g
p
erforman
ce
i
n
fog
,
weak
er
i
n
rai
n
Z
h
an
g
e
t
a
l
.
[
2
6
]
Y
O
L
O
v
5
S
n
o
w
0
.
8
7
Opt
i
mi
ze
d
for
sn
o
w
-
sp
ec
i
fi
c
v
i
s
i
b
i
l
i
t
y
i
s
s
u
e
s
L
ee
e
t
a
l
.
[
2
9
]
Y
O
L
O
v
7
L
o
w
L
i
g
h
t
0
.
8
3
S
p
ec
i
al
i
ze
d
for
n
i
g
h
t
an
d
i
l
l
u
mi
n
ati
o
n
ch
al
l
e
n
g
e
s
P
r
o
p
o
s
e
d
M
o
d
e
l
(
2
0
2
5
)
Y
O
L
O
v
8
+
G
A
N
M
u
l
t
i
-
w
e
a
t
h
e
r
0
.
9
2
(cl
e
a
r
),
0
.
8
8
(ra
i
n
),
0
.
8
4
(
f
o
g
)
,
0
.
8
3
(sn
o
w),
0
.
8
1
(l
o
w
l
i
g
h
t
)
B
a
l
a
n
c
e
d
p
e
r
fo
r
m
a
n
c
e
a
c
r
o
ss
a
l
l
c
o
n
d
i
t
i
o
n
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
E
n
h
a
n
ce
men
t o
f YOLOv8
fo
r
o
b
ject
d
etec
tio
n
in
a
d
ve
r
s
e
w
ea
th
er c
o
n
d
itio
n
s
…
(
Ta
lifh
a
n
i
C
a
lvin
Ts
h
ip
o
ta
)
2239
T
h
e
co
m
p
ar
is
o
n
d
em
o
n
s
tr
ate
s
h
o
w
well
-
s
u
ited
m
o
d
els
p
e
r
f
o
r
m
e
d
in
t
h
eir
r
esp
ec
tiv
e
f
ield
s
.
Fo
r
ex
am
p
le,
F
a
n
g
et
al
.
[
2
7
]
s
h
o
wed
th
at
Fas
ter
R
-
C
NN
p
er
f
o
r
m
ed
b
etter
th
a
n
m
a
n
y
m
o
d
els
in
f
o
g
(
m
AP@
0
.
5
=
0
.
8
9
)
,
wh
ile
it
p
er
f
o
r
m
e
d
wo
r
s
e
in
r
ain
(
0
.
8
2
)
.
Similar
ly
,
L
e
e
et
al
.
[
2
9
]
d
em
o
n
s
tr
ated
th
at
YOL
Ov
7
ad
ap
ted
ef
f
ec
tiv
el
y
to
lo
w
-
lig
h
t
cir
cu
m
s
tan
ce
s
with
m
AP@
0
.
5
=
0
.
8
3
,
wh
ile
Z
ha
ng
et
al
.
[
2
6
]
s
tated
th
at
th
eir
YOL
Ov
5
-
b
ased
s
n
o
w
-
o
p
tim
ized
d
etec
to
r
r
ea
ch
e
d
m
AP@
0
.
5
=
0
.
8
7
.
T
h
ese
s
p
ec
ialized
m
o
d
els,
h
o
wev
er
,
f
r
e
q
u
en
tly
f
o
r
g
o
g
en
er
aliza
tio
n
in
f
av
o
r
o
f
s
p
ec
ializ
atio
n
;
th
ey
p
er
f
o
r
m
well
in
th
eir
i
n
ten
d
ed
s
u
r
r
o
u
n
d
in
g
s
b
u
t
s
u
f
f
er
g
r
ea
tly
o
u
ts
id
e
o
f
th
em
.
O
n
t
h
e
o
th
er
h
an
d
,
th
e
m
AP@
0
.
5
o
f
o
u
r
s
u
g
g
ested
Y
OL
Ov
8
+
GAN
m
o
d
el
r
an
g
ed
f
r
o
m
0
.
8
1
(
lo
w
lig
h
t)
to
0
.
9
2
(
clea
r
co
n
d
itio
n
s
)
,
d
em
o
n
s
tr
atin
g
c
o
n
s
is
ten
t
f
lex
ib
ilit
y
ac
r
o
s
s
a
v
ar
iety
o
f
wea
th
er
s
itu
atio
n
s
.
Fo
r
r
ea
l
-
wo
r
ld
im
p
lem
en
tatio
n
i
n
au
to
n
o
m
o
u
s
d
r
iv
in
g
s
y
s
tem
s
,
in
tellig
en
t
tr
af
f
ic
m
o
n
ito
r
in
g
,
an
d
s
u
r
v
eillan
ce
n
etwo
r
k
s
—
wh
er
e
u
n
p
r
ed
ictab
le
wea
th
er
n
ec
ess
itates
d
etec
to
r
s
th
at
p
er
f
o
r
m
co
n
s
is
ten
tly
ac
r
o
s
s
s
ce
n
ar
io
s
r
ath
er
th
an
e
x
ce
llin
g
in
ju
s
t
o
n
e
—
th
is
b
alan
ce
is
esp
ec
ially
cr
u
cial
[
3
5
]
,
[
2
9
]
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
YOL
O
co
m
p
ar
is
o
n
s
in
d
if
f
er
en
t
wea
th
er
s
itu
atio
n
s
is
d
is
p
lay
ed
in
Fig
u
r
es
3
a
n
d
4
.
Fig
u
r
e
4
s
h
o
ws
a
s
id
e
-
by
-
s
id
e
b
a
r
ch
a
r
t
o
v
er
clim
atic
cir
cu
m
s
tan
ce
s
,
wh
ile
Fig
u
r
e
3
co
m
p
ar
es
th
e
s
u
g
g
ested
m
o
d
e
l
m
ea
n
av
e
r
ag
e
p
r
ec
is
io
n
(
m
A
P@
0
.
5
)
to
ea
r
lier
a
p
p
r
o
ac
h
es.
B
o
th
f
ig
u
r
es
s
h
o
w
th
at
ev
en
if
s
p
ec
ialized
d
etec
to
r
s
m
ay
p
er
f
o
r
m
b
etter
th
an
o
u
r
m
o
d
el
in
s
o
m
e
s
itu
atio
n
s
,
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
o
f
f
er
s
o
v
er
all
r
o
b
u
s
tn
ess
an
d
b
alan
ce
d
ac
cu
r
ac
y
in
a
v
ar
iety
o
f
co
n
d
itio
n
.
T
h
is
im
p
r
o
v
e
d
ad
a
p
tab
ilit
y
i
s
ex
p
lain
ed
b
y
th
e
ad
d
itio
n
o
f
GAN
-
b
ased
p
r
e
-
p
r
o
ce
s
s
in
g
,
wh
ich
im
p
r
o
v
es
im
a
g
e
q
u
ality
an
d
lo
wer
s
n
o
is
e
b
ef
o
r
e
d
etec
ti
o
n
,
allo
win
g
th
e
YOL
Ov
8
b
ac
k
b
o
n
e
to
e
x
tr
ac
t
f
ea
tu
r
es
m
o
r
e
s
u
cc
ess
f
u
lly
ev
en
in
m
o
r
e
d
if
f
ic
u
lt
co
n
d
itio
n
[
2
5
]
,
[
3
6
]
.
B
ec
au
s
e
th
e
p
r
o
p
o
s
ed
m
o
d
el
co
m
b
i
n
es
r
esil
ien
ce
p
o
wer
ed
b
y
GANs
with
a
s
tr
o
n
g
b
aselin
e
d
etec
ti
o
n
ca
p
ab
ilit
y
,
it
is
well
-
p
o
s
iti
o
n
ed
f
o
r
u
s
e
in
n
ex
t
-
g
en
er
atio
n
s
m
ar
t t
r
an
s
p
o
r
tatio
n
s
y
s
tem
s
an
d
s
m
ar
t c
ity
in
f
r
a
s
tr
u
ctu
r
e.
Fig
u
r
e
4.
C
o
m
p
ar
is
o
n
of
YOL
O
v8
p
er
f
o
r
m
an
c
e
u
n
d
er
d
if
f
er
en
t
wea
th
er
c
o
n
d
i
t
i
o
n
4
.
3
.
T
r
a
i
ni
n
g
di
a
g
n
o
s
t
i
c
s
u
nd
e
r
r
a
in
y
c
o
nd
i
t
i
o
ns
Usi
n
g
a
d
i
f
f
er
en
t
ty
p
e
o
f
d
i
ag
n
o
s
tic
g
r
a
p
h
s
th
at
d
is
p
lay
d
if
f
er
en
t
asp
ec
ts
o
f
tr
ain
in
g
d
y
n
am
ics,
co
n
v
er
g
en
ce
s
tab
ilit
y
,
an
d
d
et
ec
tio
n
r
eliab
ilit
y
,
we
ass
ess
ed
our
YOL
Ov
8
m
o
d
el
with
GA
N
p
r
e
-
p
r
o
ce
s
s
in
g
in
r
ain
y
co
n
d
itio
n
s
.
T
h
ese
p
lo
ts
d
em
o
n
s
tr
ate
how
well
th
e
m
o
d
el
b
alan
ce
s
co
n
f
id
en
ce
lev
els
,
f
alse
p
o
s
itiv
es,
an
d
f
alse
n
eg
ativ
es
u
n
d
er
wea
th
er
-
in
d
u
ce
d
d
is
to
r
tio
n
s
s
u
ch
as
m
o
tio
n
b
lu
r
,
r
ain
f
all,
an
d
r
ef
lectio
n
s
.
T
h
ese
ex
am
p
les
h
elp
u
s
c
o
m
p
r
e
h
en
d
h
o
w
th
e
m
o
d
el
wo
r
k
s
an
d
wh
e
th
er
it is
ap
p
r
o
p
r
iate
f
o
r
p
r
ac
ti
ca
l a
p
p
licatio
n
s
.
4
.
3
.
1
.
F1
–
c
o
n
f
i
d
e
n
c
e
c
u
r
v
e
T
h
e
lin
k
b
etwe
en
th
e
c
o
n
f
id
e
n
ce
th
r
esh
o
ld
a
n
d
th
e
F1
-
s
co
r
e,
wh
ich
is
th
e
h
ar
m
o
n
ic
m
ea
n
of
p
r
ec
is
io
n
an
d
r
ec
all,
is
d
ep
icted
b
y
th
e
F1
–
co
n
f
id
e
n
ce
cu
r
v
e
(
Fig
u
r
e
5
)
.
T
h
e
id
ea
l
tr
a
d
e
-
o
f
f
f
o
r
c
lass
if
icatio
n
in
wet
co
n
d
itio
n
s
is
in
d
icate
d
b
y
t
h
e
cu
r
v
e’
s
p
ea
k
,
wh
ich
was
at
0
.
7
7
with
a
co
n
f
id
en
ce
t
h
r
esh
o
ld
o
f
0
.
4
3
.
T
h
is
r
esu
lt
d
em
o
n
s
tr
ates
th
at
wh
ile
ex
ce
s
s
iv
ely
s
ev
er
e
lim
its
d
r
a
s
tically
lo
wer
r
ec
all,
to
o
lo
w
th
r
esh
o
ld
s
r
esu
lt
in
an
o
v
er
wh
elm
in
g
n
u
m
b
e
r
o
f
f
alse
p
o
s
itiv
es.
I
n
p
r
ac
tical
a
p
p
licatio
n
s
lik
e
au
to
n
o
m
o
u
s
d
r
iv
in
g
,
wh
er
e
h
i
g
h
p
r
ec
is
io
n
is
f
r
eq
u
en
tly
s
o
u
g
h
t,
th
is
tr
ad
e
-
o
f
f
is
cr
itical
s
in
ce
f
alse
n
eg
ativ
es
(
m
is
s
ed
d
etec
t
io
n
s
o
f
p
ed
estrian
s
o
r
au
to
m
o
b
iles
)
m
ay
b
e
m
o
r
e
ex
p
en
s
iv
e
th
an
f
alse p
o
s
itiv
es
[
2
7
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.