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
.
1
,
Feb
r
u
ar
y
20
2
6
,
p
p
.
450
~
462
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
1
.
pp
4
5
0
-
4
6
2
450
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
From Y
O
LO
V
1
t
o
YO
L
O
V
1
1
:
co
mpa
ra
tive
a
na
ly
sis
of
YO
L
O
a
lg
o
rithm
(
r
ev
iew)
I
m
a
ne
B
eqqa
li H
a
s
s
a
ni
1
,
So
ufia
B
enhi
da
1
,
Na
bil
L
a
m
ii
1
,
K
ha
lid
O
qa
idi
1
,
Ahm
ed
O
ui
dd
a
d
2
,
So
uk
a
ina
G
hia
di
3
1
R
e
s
e
a
r
c
h
,
D
e
v
e
l
o
p
me
n
t
,
a
n
d
I
n
n
o
v
a
t
i
o
n
L
a
b
o
r
a
t
o
r
y
,
M
u
n
d
i
a
p
o
l
i
s Un
i
v
e
r
si
t
y
,
C
a
sa
b
l
a
n
c
a
,
M
o
r
o
c
c
o
2
LA
M
S
A
D
L
a
b
o
r
a
t
o
r
y
,
N
a
t
i
o
n
a
l
S
c
h
o
o
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
H
a
ss
a
n
1
s
t
U
n
i
v
e
r
si
t
y
,
B
e
r
r
e
c
h
i
d
,
M
o
r
o
c
c
o
3
I
n
t
e
r
d
i
sci
p
l
i
n
a
r
y
L
a
b
o
r
a
t
o
r
y
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
N
a
t
i
o
n
a
l
S
c
h
o
o
l
o
f
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s,
H
a
ss
a
n
1
st
U
n
i
v
e
r
si
t
y
,
B
e
r
r
e
c
h
i
d
,
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
Ap
r
1
,
2
0
2
5
R
ev
is
ed
Oct
2
4
,
2
0
2
5
Acc
ep
ted
No
v
2
3
,
2
0
2
5
Ob
jec
t
d
e
tec
ti
o
n
i
n
ima
g
e
s
o
r
v
i
d
e
o
s
fa
c
e
s
se
v
e
ra
l
c
h
a
ll
e
n
g
e
s
b
e
c
a
u
se
th
e
d
e
tec
ti
o
n
m
u
st
b
e
a
c
c
u
ra
te,
e
ffi
c
ien
t
a
n
d
fa
st.
Th
e
y
o
u
o
n
ly
l
o
o
k
o
n
c
e
(
YO
LO
)
a
lg
o
rit
h
m
wa
s
in
v
e
n
ted
to
m
e
e
t
th
e
se
c
rit
e
ria.
Bu
t
with
th
e
c
re
a
ti
o
n
o
f
se
v
e
ra
l
v
e
rsio
n
s
o
f
t
h
is
a
lg
o
ri
t
h
m
(fro
m
V1
to
V1
1
),
it
b
e
c
o
m
e
s
d
iffi
c
u
lt
fo
r
re
se
a
rc
h
e
rs
to
c
h
o
o
se
t
h
e
b
e
st
o
n
e
.
T
h
e
m
a
in
o
b
jec
ti
v
e
o
f
th
is
re
v
iew
is
to
p
re
se
n
t
a
n
d
c
o
m
p
a
re
th
e
e
lev
e
n
v
e
rsio
n
s
o
f
th
e
y
o
l
o
a
lg
o
rit
h
m
in
o
r
d
e
r
t
o
k
n
o
w
w
h
e
n
u
sin
g
th
e
a
p
p
ro
p
riat
e
o
n
e
fo
r
th
e
stu
d
y
.
Th
e
m
e
th
o
d
o
lo
g
y
u
se
d
fo
r
t
h
is
w
o
rk
is
a
l
i
g
n
e
d
wit
h
p
re
f
e
rre
d
re
p
o
rt
in
g
i
tem
s
fo
r
sy
ste
m
a
ti
c
re
v
iew
s
a
n
d
m
e
ta
-
a
n
a
ly
se
s
(
P
RI
S
M
A
)
p
ri
n
c
ip
les
a
n
d
th
e
re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
th
e
c
h
o
ice
o
f
t
h
e
b
e
st
v
e
rsio
n
m
a
in
l
y
d
e
p
e
n
d
s
o
n
t
h
e
p
r
io
rit
ies
o
f
t
h
e
stu
d
y
.
If
th
e
st
u
d
y
p
rio
r
it
ize
s
a
c
c
u
ra
c
y
a
n
d
d
e
tec
ti
o
n
o
f
sm
a
ll
o
b
jec
ts,
it
s
h
o
u
ld
u
se
YO
LO
V4
,
YO
LO
V5
,
YO
LO
V6
,
YO
LO
V7
,
YO
LO
V8
,
YO
LO
V9
,
YO
LO
V1
0
o
r
YO
LO
V
1
1
.
W
h
il
e
stu
d
ies
t
h
a
t
p
r
io
rit
ize
d
e
tec
ti
o
n
sp
e
e
d
sh
o
u
l
d
u
se
YO
LO
V
5
,
YO
LO
V6
,
YO
LO
V
7
,
YO
LO
V8
,
YO
LO
V1
0
o
r
YO
LO
V1
1
.
In
c
o
m
p
lex
e
n
v
ir
o
n
m
e
n
t,
re
se
a
rc
h
e
rs
sh
o
u
ld
a
v
o
id
u
sin
g
YO
LO
V1
,
YO
LO
V2
,
YO
LO
V3
,
YO
LO
V5
,
YO
LO
V7
a
n
d
YO
LO
V9
.
An
d
re
se
a
rc
h
e
rs
wh
o
a
re
lo
o
k
i
n
g
f
o
r
a
g
o
o
d
a
c
c
u
ra
c
y
a
n
d
sp
e
e
d
a
n
d
a
re
d
u
c
e
d
n
u
m
b
e
r
o
f
p
a
ra
m
e
ters
sh
o
u
l
d
u
se
YO
LO
V1
0
o
r
YO
LO
V
1
1
.
K
ey
w
o
r
d
s
:
C
o
m
p
u
ter
v
is
io
n
E
v
o
lu
tio
n
o
f
YOL
O
Ma
ch
in
e
lear
n
in
g
Ob
ject
-
b
ased
d
etec
tio
n
YOL
O
a
lg
o
r
ith
m
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
:
I
m
an
e
B
eq
q
ali
Hass
an
i
R
esear
ch
,
Dev
elo
p
m
en
t,
a
n
d
I
n
n
o
v
atio
n
L
ab
o
r
ato
r
y
,
Mu
n
d
i
ap
o
lis
Un
iv
er
s
ity
C
asab
lan
ca
,
Mo
r
o
cc
o
E
m
ail: i.
b
eq
q
ali@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Dee
p
lear
n
in
g
h
as
ad
v
an
ce
d
s
ig
n
if
ican
tly
in
r
ec
en
t
y
ea
r
s
an
d
h
as
co
n
tr
ib
u
ted
to
th
e
ad
v
a
n
ce
m
en
t
o
f
s
ev
er
al
f
ield
s
(
h
ea
lth
,
ed
u
ca
tio
n
,
b
io
l
o
g
y
,
ae
r
o
n
a
u
tics
,
an
d
au
to
m
o
tiv
e)
.
T
h
is
is
d
u
e
to
th
e
av
ailab
ilit
y
o
f
r
eso
u
r
ce
s
s
u
ch
as
d
atasets
,
r
o
b
u
s
tn
ess
o
f
h
ar
d
war
e
an
d
th
e
d
ev
el
o
p
m
e
n
t
o
f
s
o
f
twar
e
an
d
wo
r
k
to
o
ls
.
Ho
wev
er
,
th
er
e
is
s
till
a
lo
t
o
f
s
p
ac
e
f
o
r
r
esear
ch
an
d
d
ev
e
lo
p
m
en
t
[
1
]
,
esp
ec
ially
in
th
e
f
ield
o
f
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
.
R
ea
l
-
tim
e
is
cr
itical,
b
ec
au
s
e
it
d
o
es
n
o
t
to
ler
ate
an
y
m
ar
g
in
o
f
e
r
r
o
r
.
Ob
jects
m
u
s
t
b
e
d
etec
ted
o
n
e
h
u
n
d
r
e
d
p
e
r
ce
n
t
if
we
wan
t
to
h
a
v
e
a
p
er
f
ec
t,
co
n
s
is
ten
t
an
d
r
eliab
le
r
es
u
lt.
T
o
m
e
et
th
ese
r
eq
u
ir
em
e
n
ts
,
r
esear
ch
er
s
h
av
e
u
s
ed
th
e
YOL
O
alg
o
r
ith
m
b
ec
au
s
e
ac
co
r
d
in
g
to
Alah
d
al
et
a
l.
[
2
]
,
d
i
f
f
er
en
t
v
er
s
io
n
s
o
f
YOL
O
h
a
v
e
d
e
m
o
n
s
tr
ated
s
p
ee
d
an
d
ac
c
u
r
a
cy
in
d
etec
tin
g
o
b
jects.
Sin
ce
2
0
1
6
,
th
e
YOL
O
alg
o
r
ith
m
h
as
ev
o
lv
ed
an
d
r
esear
ch
er
s
h
av
e
b
ee
n
ab
l
e
to
cr
ea
te
elev
en
v
e
r
s
io
n
s
o
f
th
is
alg
o
r
ith
m
(
f
r
o
m
YOL
O
V1
t
o
YOL
O
V1
1
)
.
E
ac
h
o
f
th
em
h
as its
o
wn
ch
ar
ac
ter
is
tics
,
ad
v
an
tag
es a
n
d
d
is
ad
v
an
tag
es.
Sev
er
al
s
tu
d
ies
in
th
e
d
o
m
ain
o
f
co
m
p
u
ter
v
is
io
n
an
d
r
ea
l
-
ti
m
e
d
etec
tio
n
e
m
p
h
asize
th
e
i
m
p
o
r
tan
c
e
o
f
s
tr
ik
in
g
an
eq
u
ilib
r
iu
m
b
et
wee
n
p
r
ec
is
io
n
an
d
s
p
ee
d
.
H
o
wev
er
,
p
r
e
v
io
u
s
wo
r
k
ten
d
s
to
f
o
cu
s
o
n
ly
o
n
a
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
F
r
o
m
YOLO V
1
to
YOLO V
1
1
:
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f
YOLO
…
(
I
ma
n
e
B
eq
q
a
li Ha
s
s
a
n
i
)
451
s
in
g
le
v
er
s
io
n
o
f
th
e
YOL
O
alg
o
r
ith
m
an
d
f
ails
to
o
f
f
er
c
o
m
p
ar
ativ
e
s
tu
d
ies
o
f
m
u
ltip
le
YOL
O
v
er
s
io
n
s
(
esp
ec
ially
th
o
s
e
in
clu
d
i
n
g
r
e
ce
n
t
v
er
s
io
n
s
)
in
o
r
d
er
to
p
r
o
v
id
e
co
n
c
r
ete
r
ec
o
m
m
en
d
atio
n
s
.
B
y
co
v
er
in
g
al
l
v
er
s
io
n
s
o
f
YOL
O
u
p
to
v
e
r
s
io
n
1
1
,
th
is
liter
atu
r
e
r
ev
iew
f
ills
a
cr
itical
g
ap
i
n
th
e
liter
atu
r
e
an
d
a
d
d
r
ess
es
th
e
p
r
ac
tical
n
ee
d
s
o
f
r
esear
ch
er
s
in
s
elec
tin
g
th
e
b
est o
b
ject
d
et
ec
tio
n
m
o
d
el
b
ased
o
n
th
eir
r
e
q
u
ir
em
en
ts
.
I
n
th
is
co
n
tex
t,
we
p
r
esen
t
th
is
r
ev
iew
in
o
r
d
er
to
an
s
wer
to
th
e
f
o
llo
win
g
r
esear
ch
q
u
esti
o
n
:
“
W
h
at
ar
e
th
e
ch
an
g
es
m
ad
e
to
th
e
YOL
O
alg
o
r
ith
m
in
ea
ch
o
f
its
v
er
s
io
n
s
,
wh
at
ar
e
th
e
ad
v
an
tag
es
an
d
d
is
ad
v
an
tag
es
o
f
ev
er
y
YOL
O
alg
o
r
ith
m
v
er
s
io
n
a
n
d
wh
at
is
th
e
b
est
v
er
s
io
n
to
u
s
e
f
o
r
e
v
er
y
r
eq
u
ir
em
en
t?
”
.
T
o
an
s
wer
th
is
q
u
esti
o
n
,
we
p
r
o
p
o
s
e
th
e
p
r
esen
t
wo
r
k
,
wh
o
s
e
o
b
jectiv
e
is
to
p
r
o
v
id
e
a
co
m
p
ar
ativ
e
an
aly
s
is
o
f
th
e
d
if
f
er
en
t
v
e
r
s
io
n
s
o
f
th
e
YOL
O
alg
o
r
ith
m
.
Mo
r
e
p
r
ec
is
ely
,
we
id
en
tify
t
h
e
ad
v
an
tag
es
an
d
d
is
ad
v
an
tag
es
an
d
p
r
esen
t
th
e
co
n
tr
ib
u
tio
n
o
f
ea
ch
v
e
r
s
io
n
o
f
YOL
O
in
ter
m
s
o
f
ar
ch
itectu
r
e,
ac
cu
r
ac
y
,
s
p
ee
d
,
an
d
im
p
r
o
v
em
e
n
ts
in
o
r
d
er
t
o
k
n
o
w
th
e
b
est
v
er
s
io
n
to
u
s
e
f
o
r
ev
er
y
r
eq
u
ir
em
en
t.
T
h
u
s
,
th
e
p
r
esen
t
p
ap
er
is
s
tr
u
ctu
r
e
d
as
f
o
llo
ws:
as
a
f
ir
s
t
s
tep
,
we
d
escr
ib
e
th
e
m
eth
o
d
o
lo
g
y
o
f
th
is
liter
atu
r
e
r
ev
iew.
T
h
en
we
d
escr
ib
e
an
o
v
e
r
v
iew
ab
o
u
t
YOL
O
alg
o
r
ith
m
b
y
p
r
esen
tin
g
its
elev
en
v
er
s
io
n
s
.
Af
ter
th
at,
we
r
ep
o
r
t
th
e
r
esu
lts
an
d
f
in
ally
we
d
is
cu
s
s
th
em
.
2.
M
E
T
H
O
DO
L
O
G
Y
T
h
e
m
eth
o
d
o
l
o
g
y
o
f
th
is
r
e
v
iew
is
b
ased
o
n
th
e
m
eth
o
d
o
lo
g
y
o
f
[
3
]
alig
n
e
d
with
PR
I
SMA
p
r
in
cip
les
[
4
]
.
I
t
in
cl
u
d
es
th
e
f
o
llo
win
g
s
tep
s
:
i)
R
esear
ch
q
u
esti
o
n
s
f
o
r
t
h
e
liter
atu
r
e
r
e
v
iew,
ii)
Do
cu
m
e
n
t
s
ea
r
ch
s
tr
ateg
y
,
iii)
Do
cu
m
en
t
s
elec
tio
n
cr
iter
ia,
iv
)
Do
cu
m
en
t
v
alid
ity
ass
ess
m
en
t
an
d
v
)
Data
co
llectio
n
.
I
n
th
is
r
ev
iew
we
an
s
wer
th
e
f
o
llo
win
g
r
esear
ch
q
u
esti
o
n
s
as
s
h
o
wn
in
T
ab
le
1
wh
o
s
e
f
o
r
m
u
latio
n
was
m
ad
e
b
ased
o
n
th
e
p
o
p
u
latio
n
,
i
n
ter
v
en
tio
n
,
c
o
m
p
ar
is
o
n
,
o
u
tco
m
e
(
PICO)
tech
n
iq
u
e
[
5
]
:
T
ab
le
1
.
R
esear
ch
q
u
esti
o
n
s
b
ased
o
n
th
e
PICO te
ch
n
iq
u
e
R
e
se
a
r
c
h
q
u
e
st
i
o
n
C
o
r
r
e
s
p
o
n
d
i
n
g
P
I
C
O
C
e
l
e
m
e
n
t
W
h
a
t
a
r
e
t
h
e
c
h
a
n
g
e
s
ma
d
e
t
o
t
h
e
Y
O
LO
a
l
g
o
r
i
t
h
m
i
n
e
a
c
h
o
f
i
t
s
v
e
r
s
i
o
n
s
?
P
o
p
u
l
a
t
i
o
n
:
Y
O
LO
A
l
g
o
r
i
t
h
m
v
e
r
si
o
n
s
I
n
t
e
r
v
e
n
t
i
o
n
:
C
h
a
n
g
e
s
a
n
d
m
o
d
i
f
i
c
a
t
i
o
n
s
a
c
r
o
ss
v
e
r
si
o
n
s
W
h
i
c
h
v
e
r
si
o
n
o
f
t
h
e
y
o
l
o
a
l
g
o
r
i
t
h
m
i
s t
h
e
b
e
s
t
?
C
o
m
p
a
r
i
so
n
:
C
o
m
p
a
r
i
so
n
b
e
t
w
e
e
n
v
e
r
si
o
n
s
t
o
d
e
t
e
r
m
i
n
e
t
h
e
b
e
s
t
o
n
e
W
h
a
t
a
r
e
t
h
e
a
d
v
a
n
t
a
g
e
s a
n
d
d
i
s
a
d
v
a
n
t
a
g
e
s
o
f
e
a
c
h
v
e
r
si
o
n
o
f
t
h
e
Y
O
LO
a
l
g
o
r
i
t
h
m?
O
u
t
c
o
m
e
:
A
d
v
a
n
t
a
g
e
s/
D
i
sa
d
v
a
n
t
a
g
e
s
2
.
1
.
Sea
rc
h str
a
t
eg
y
a
nd
s
elec
t
io
n pro
ce
s
s
T
h
e
ar
ticles
u
s
ed
in
th
is
r
ev
iew
ar
e
E
n
g
lis
h
wr
itten
,
ex
t
r
ac
ted
f
r
o
m
Scien
ce
Dir
ec
t
an
d
g
o
o
g
le
s
ch
o
lar
d
atab
ases
an
d
wer
e
p
u
b
lis
h
ed
b
etwe
en
2
0
1
2
an
d
2
0
2
5
as
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
e
m
ajo
r
ity
o
f
th
e
ar
ticles
ar
e
v
er
y
r
ec
en
t.
I
n
d
e
ed
3
ar
ticles
wer
e
p
u
b
lis
h
ed
in
2
0
2
5
,
2
9
ar
ticles
wer
e
p
u
b
lis
h
ed
in
2
0
2
4
,
5
ar
ticles
wer
e
p
u
b
lis
h
ed
in
2
0
2
3
,
3
ar
ticles
wer
e
p
u
b
lis
h
ed
f
o
r
ea
ch
y
ea
r
2
0
1
8
,
2
0
2
0
,
2
0
2
1
an
d
2
0
2
2
.
An
d
f
o
r
2
0
1
2
,
2
0
1
3
,
2
0
1
4
,
2
0
1
5
,
2
0
1
6
an
d
2
0
1
7
o
n
ly
o
n
e
ar
ticle
i
s
u
s
ed
f
o
r
ea
ch
y
ea
r
.
T
h
e
ar
ti
cles
u
s
e
th
e
YOL
O
alg
o
r
ith
m
ap
p
lied
in
d
if
f
er
e
n
t
f
ield
s
an
d
in
d
if
f
er
en
t c
o
u
n
t
r
ies.
Fig
u
r
e
1
.
R
esear
ch
p
r
o
ce
d
u
r
e
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
.
1
,
Feb
r
u
ar
y
20
2
6
:
4
5
0
-
462
452
T
h
e
k
ey
wo
r
d
s
/s
ea
r
ch
s
tr
in
g
s
u
s
e
ad
v
an
ce
d
B
o
o
lean
q
u
er
ies
o
n
s
cien
tific
d
atab
ases
a
n
d
ar
e
as:
i)
YOL
O
alg
o
r
ith
m
OR
Yo
lo
OR
Ob
ject
d
etec
tio
n
u
s
in
g
Y
OL
O
alg
o
r
ith
m
.
ii)
YOL
O
v
er
s
io
n
(
Nu
m
b
er
)
OR
YOL
O
V
(
Nu
m
b
er
)
OR
YOL
O
V
(
Nu
m
b
er
)
.
An
d
iii)
A
d
v
an
tag
es
OR
/AND
d
is
ad
v
an
tag
es
o
f
YOL
O
V
(
Nu
m
b
er
)
.
W
e
r
ep
lace
th
e
(
N
u
m
b
er
)
,
with
a
n
u
m
b
er
f
r
o
m
1
t
o
1
1
f
o
r
ea
c
h
s
ea
r
ch
.
T
h
e
f
o
llo
win
g
cr
iter
ia
s
er
v
e
as
th
e
f
o
u
n
d
atio
n
f
o
r
th
e
s
elec
tio
n
p
r
o
ce
s
s
:
−
Yea
r
o
f
p
u
b
licatio
n
: th
e
m
o
s
t r
ec
en
t a
r
ticles ar
e
th
e
m
o
s
t p
r
e
f
er
r
ed
−
E
n
g
lis
h
lan
g
u
a
g
e:
On
ly
E
n
g
lis
h
-
lan
g
u
a
g
e
p
u
b
licatio
n
s
ar
e
u
s
ed
.
−
T
itle: o
n
ly
ar
ticles m
en
tio
n
in
g
“
YOL
O
”
,
“
Ob
ject
d
etec
tio
n
”
in
th
e
titl
e
ar
e
r
ev
iewe
d
.
−
Fin
ally
,
we
ex
clu
d
e
s
tu
d
ies th
at
d
o
n
o
t a
d
d
r
ess
o
b
ject
d
etec
tio
n
,
an
d
th
at
d
o
n
o
t
u
s
e
th
e
Y
OL
O
alg
o
r
ith
m
.
2
.
2
.
Do
cu
m
ent
v
a
lid
it
y
a
s
s
es
s
m
ent
a
nd
da
t
a
co
llect
io
n
I
n
o
r
d
e
r
to
ev
alu
ate
th
e
q
u
ality
o
f
s
tu
d
ies,
we
u
s
e
q
u
esti
o
n
s
,
an
d
we
ass
ig
n
a
s
co
r
e
to
ea
ch
q
u
esti
o
n
.
Acc
o
r
d
in
g
ly
,
th
e
v
alu
es
“
0,
”
“
0
.
5
,
”
an
d
“
1
”
ar
e
ass
ig
n
ed
to
th
e
a
n
s
wer
s
“
No
,
”
“
Par
tially
,
”
an
d
“
Yes,
”
r
esp
ec
tiv
ely
.
T
h
e
s
u
m
o
f
t
h
e
s
co
r
es
f
o
r
ea
ch
s
tu
d
y
is
th
e
n
ca
lcu
lated
.
T
h
e
q
u
ality
ass
ess
m
en
t
q
u
esti
o
n
s
(
Q)
ar
e
as
f
o
llo
ws:
(
Q1
)
Do
es
th
e
s
tu
d
y
p
r
o
v
id
e
an
ex
p
licit
u
s
e
o
f
th
e
YOL
O
alg
o
r
ith
m
?
(
Q2
)
I
s
o
b
ject
d
etec
tio
n
th
e
m
ain
o
b
jectiv
e
o
f
t
h
e
s
tu
d
y
?
Fo
r
ar
ticles
th
at
d
is
cu
s
s
ed
th
e
YOL
O
alg
o
r
ith
m
,
we
ex
tr
ac
ted
th
e
f
o
llo
win
g
d
ata:
y
ea
r
o
f
p
u
b
licatio
n
,
v
er
s
io
n
o
f
th
e
YOL
O
alg
o
r
ith
m
,
ar
c
h
itectu
r
e,
s
t
u
d
y
r
esu
lts
,
an
d
ad
v
a
n
tag
es
an
d
d
is
ad
v
a
n
tag
es
o
f
th
e
u
s
ed
v
er
s
io
n
.
3.
YO
L
O
AL
G
O
RI
T
H
M
:
AN
O
VE
RVI
E
W
Acc
o
r
d
in
g
to
G
h
eo
r
g
h
e
et
a
l.
[
6
]
,
YOL
O
alg
o
r
ith
m
h
as
b
ec
o
m
in
g
v
er
y
p
o
p
u
lar
am
o
n
g
th
e
s
er
ies
o
f
o
b
ject
d
etec
tio
n
m
o
d
els.
T
h
e
alg
o
r
ith
m
h
as
b
ee
n
im
p
r
o
v
ed
s
ig
n
if
ican
tly
s
in
ce
its
f
ir
s
t
p
u
b
licatio
n
,
wit
h
v
er
s
io
n
s
s
p
an
n
in
g
f
r
o
m
V
1
to
V
11
[
7
]
.
Du
r
in
g
its
ev
o
lu
tio
n
,
ea
ch
v
e
r
s
io
n
o
f
YOL
O
h
as
b
ee
n
im
p
r
o
v
e
d
to
h
av
e
m
o
r
e
p
r
ec
is
io
n
[
8
]
,
s
m
aller
v
o
lu
m
e
an
d
h
ig
h
e
r
s
p
ee
d
[
7
]
.
T
h
e
f
ir
s
t
two
v
er
s
io
n
s
o
f
y
o
lo
u
s
e
g
e
n
er
al
a
r
ch
itectu
r
es,
b
u
t
s
tar
tin
g
with
YOL
O
V3
,
th
e
m
ai
n
ar
ch
itectu
r
e
o
f
th
e
YOL
O
alg
o
r
ith
m
in
clu
d
es
th
r
ee
m
ain
p
a
r
ts
,
n
am
ely
th
e
b
ac
k
b
o
n
e
,
n
ec
k
an
d
h
ea
d
[
9
]
.
T
h
e
m
ain
r
o
le
o
f
th
e
b
ac
k
b
o
n
e
is
to
ex
tr
ac
t th
e
m
o
s
t im
p
o
r
tan
t
f
ea
tu
r
es o
f
th
e
im
a
g
e
an
d
t
r
an
s
m
it th
em
to
th
e
h
ea
d
th
r
o
u
g
h
th
e
n
ec
k
[
9
]
.
T
h
e
m
ain
f
u
n
ctio
n
o
f
th
e
n
ec
k
is
v
e
r
y
im
p
o
r
tan
t
in
th
is
p
r
o
ce
s
s
.
B
ec
au
s
e
it
co
m
p
iles
th
e
f
ea
tu
r
e
m
ap
s
g
e
n
er
ated
b
y
th
e
b
ac
k
b
o
n
e
[
9
]
.
I
n
a
d
d
itio
n
to
th
is
,
th
e
n
ec
k
h
elp
s
im
p
r
o
v
in
g
r
ec
eiv
ed
im
a
g
e’
s
f
ea
tu
r
es,
b
y
b
u
ild
in
g
f
ea
tu
r
e
p
y
r
a
m
id
s
th
at
h
elp
m
ak
in
g
m
u
ltis
ca
le
o
b
ject
d
etec
tio
n
ea
s
ier
[
9
]
.
O
n
ce
p
r
o
ce
s
s
in
g
is
c
o
m
p
leted
,
th
e
n
ec
k
s
en
d
s
th
e
d
ata
to
th
e
h
e
ad
,
wh
ich
in
tu
r
n
m
a
k
es
th
e
f
in
al
p
r
e
d
ictio
n
s
in
ter
m
s
o
f
class
if
icatio
n
an
d
g
e
n
er
aliza
tio
n
.
3
.
1
.
YO
L
O
V
1
First
in
tr
o
d
u
ce
d
b
y
R
ed
m
o
n
et
a
l.
[
1
0
]
,
YOL
O
V1
is
b
ased
o
n
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs
)
to
d
etec
t
o
b
ject
th
r
o
u
g
h
u
s
in
g
r
eg
r
ess
io
n
[
1
1
]
.
I
t d
ir
ec
tly
im
p
r
o
v
es
d
etec
tio
n
p
er
f
o
r
m
an
ce
b
y
tr
ain
in
g
o
n
co
m
p
lete
im
ag
es
[
1
0
]
.
T
h
e
o
n
e
-
s
tate
o
b
ject
d
etec
tio
n
m
o
d
e
is
m
ain
ly
u
s
ed
in
YOL
O
V1
.
T
h
is
allo
ws
to
s
im
u
ltan
eo
u
s
ly
p
r
ed
ict
all
b
o
u
n
d
in
g
b
o
x
es with
th
eir
class
es
u
s
in
g
ju
s
t a
s
in
g
le
p
ass
th
r
o
u
g
h
th
e
m
o
d
el
[
1
1
]
.
T
h
e
n
etwo
r
k
ar
ch
itectu
r
e
o
f
YOL
O
V1
is
in
s
p
ir
ed
b
y
th
e
Go
o
g
L
eNe
t
m
o
d
el
[
1
2
]
f
o
r
im
ag
e
class
if
icatio
n
[
1
0
]
.
T
h
is
f
ir
s
t
v
er
s
io
n
o
f
YOL
O
u
s
es
a
g
r
id
f
o
r
o
b
ject
d
etec
tio
n
.
E
ac
h
ce
ll
o
f
th
e
g
r
id
allo
ws
th
e
p
r
ed
ictio
n
o
f
th
e
b
o
u
n
d
in
g
b
o
x
es
an
d
class
es,
k
n
o
win
g
th
at
ea
ch
en
tr
y
is
d
iv
id
ed
i
n
to
S
x
S
[
8
]
.
T
h
e
ar
ch
itectu
r
e
o
f
YOL
O
V1
co
n
s
is
t
s
o
f
2
4
co
n
v
o
lu
tio
n
al
lay
er
s
with
2
f
u
lly
co
n
n
ec
te
d
lay
er
s
[
8
]
.
T
o
r
e
d
u
ce
th
e
n
u
m
b
er
o
f
r
eq
u
ir
ed
ch
a
n
n
els,
ea
ch
lay
er
u
s
es
a
1
x
1
co
n
v
o
lu
tio
n
th
at
p
r
ec
ed
es
a
3
x
3
c
o
n
v
o
lu
ti
o
n
.
E
x
ce
p
t
th
e
last
lay
er
t
h
at
u
s
es
a
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
,
ea
ch
lay
er
o
f
YOL
O
V1
u
s
es
leak
y
r
ec
tifie
d
lin
ea
r
u
n
it
(
L
ea
k
y
R
eL
U)
as its
ac
tiv
atio
n
f
u
n
ctio
n
[
8
]
.
Ho
wev
er
,
th
e
p
e
r
f
o
r
m
an
ce
o
f
YOL
O
V1
is
s
u
b
-
o
p
tim
al
in
s
ce
n
ar
io
s
wh
er
e
o
b
jects
ar
e
clo
s
e
to
ea
ch
o
th
er
[
9
]
.
T
h
e
n
ew
v
e
r
s
io
n
s
o
f
YOL
O
ar
e
m
o
r
e
ef
f
icien
t
an
d
f
aster
b
u
t
t
h
e
f
ir
s
t
v
er
s
io
n
o
f
th
is
alg
o
r
ith
m
r
ep
r
esen
ts
a
r
ea
l
leap
f
o
r
war
d
in
th
e
h
is
to
r
y
o
f
o
b
ject
d
etec
tio
n
.
I
n
d
ee
d
,
ac
c
o
r
d
i
n
g
to
[
9
]
,
in
th
e
h
is
to
r
y
o
f
co
m
p
u
ter
v
is
io
n
,
YOL
O
V1
i
s
a
s
ig
n
if
ican
t
ac
c
o
m
p
lis
h
m
en
t
s
in
ce
it
o
p
e
n
ed
th
e
d
o
o
r
f
o
r
o
th
er
s
o
p
h
is
ticated
d
etec
tio
n
alg
o
r
it
h
m
s
.
3
.
2
.
YO
L
O
V
2
R
elea
s
ed
b
y
R
ed
m
o
n
an
d
Far
h
ad
i
in
2
0
1
7
,
YOL
O
V2
is
b
u
ild
in
g
u
p
o
n
YOL
O
V1
,
an
d
o
f
ten
r
ef
er
r
ed
to
as
YOL
O9
0
0
0
.
T
h
e
r
esear
ch
er
s
ca
lled
it
y
o
lo
9
0
0
0
b
ec
au
s
e
it
ca
n
d
etec
t
o
v
er
9
,
0
0
0
o
b
ject
ca
teg
o
r
ies
[
1
3
]
.
I
t
o
p
tim
izes
d
etec
tio
n
an
d
class
if
icatio
n
[
1
3
]
an
d
f
o
cu
s
es
p
r
im
ar
ily
o
n
im
p
r
o
v
in
g
o
b
ject
lo
ca
lizatio
n
an
d
r
ec
all
to
en
h
an
ce
o
b
ject
d
etec
tio
n
p
er
f
o
r
m
an
ce
[
1
1
]
.
Usi
n
g
a
s
in
g
le
n
eu
r
al
n
etwo
r
k
,
th
e
s
ec
o
n
d
v
er
s
io
n
o
f
YOL
O
d
ir
ec
tly
p
r
ed
icts
th
e
b
o
u
n
d
in
g
b
o
x
an
d
th
e
ass
o
ciate
d
ca
te
g
o
r
y
[
1
4
]
.
T
h
e
m
o
d
el
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
F
r
o
m
YOLO V
1
to
YOLO V
1
1
:
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f
YOLO
…
(
I
ma
n
e
B
eq
q
a
li Ha
s
s
a
n
i
)
453
YOL
O
V2
u
tili
ze
s
an
ch
o
r
b
o
x
es
to
p
r
ed
ict
b
o
u
n
d
in
g
b
o
x
e
s
[
1
3
]
.
YOL
O
V2
'
s
an
ch
o
r
id
ea
s
b
u
ild
o
n
th
at
o
f
f
aster
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
Fas
ter
R
-
C
NN)
,
wh
ich
s
am
p
les
o
n
t
h
e
co
n
v
o
l
u
tio
n
f
ea
tu
r
e
m
ap
with
s
lid
in
g
win
d
o
w
[
1
5
]
.
Acc
o
r
d
in
g
to
Yan
et
a
l.
[
1
1
]
,
in
th
e
tr
ain
in
g
s
et,
clu
s
ter
in
g
t
h
e
b
o
u
n
d
in
g
b
o
x
es
in
to
in
s
tan
ce
s
allo
ws d
eter
m
in
in
g
th
e
s
izes o
f
th
e
a
n
ch
o
r
b
o
x
es b
y
u
s
in
g
K
-
m
ea
n
s
.
T
h
e
s
tr
u
ctu
r
e
o
f
YOL
O
V2
is
b
ased
o
n
th
e
d
ar
k
n
et
f
r
am
ewo
r
k
wh
ich
is
s
im
ilar
to
th
e
ar
c
h
itectu
r
e
o
f
YOL
O
V1
[
8
]
.
T
h
e
d
if
f
er
e
n
c
e
b
etwe
en
th
e
s
tr
u
ctu
r
e
o
f
Y
OL
O
V1
an
d
th
at
o
f
YOL
O
V2
is
th
at
f
o
r
YOL
O
V2
we
f
in
d
o
n
ly
1
9
co
n
v
o
lu
t
io
n
al
lay
er
s
wh
ich
p
r
ec
ed
e
5
m
ax
p
o
o
lin
g
lay
e
r
s
an
d
th
e
ac
tiv
atio
n
f
u
n
ctio
n
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
is
u
s
ed
f
o
r
ea
c
h
co
n
v
o
lu
tio
n
al
lay
er
[
8
]
.
I
n
a
d
d
itio
n
t
o
th
at,
Y
OL
O
V2
u
s
es
b
atch
n
o
r
m
aliza
tio
n
af
ter
ea
ch
co
n
v
o
lu
tio
n
al
lay
er
,
wh
ich
h
elp
s
to
ac
ce
ler
ate
co
n
v
er
g
e
n
ce
,
in
c
r
ea
s
e
ac
cu
r
ac
y
an
d
im
p
r
o
v
e
t
r
ain
in
g
s
tab
ilit
y
.
3
.
3
.
YO
L
O
V
3
C
r
ea
ted
b
y
Far
h
ad
i
an
d
R
ed
m
o
n
in
2
0
1
8
,
YOL
O
V3
f
ea
tu
r
ed
s
ig
n
if
ican
tly
m
o
r
e
a
d
v
a
n
ce
m
en
ts
.
I
n
d
ee
d
,
co
m
p
ar
e
d
to
YOL
O
V2
,
th
e
Av
er
a
g
e
ac
c
u
r
ac
y
is
e
s
s
en
tially
im
p
r
o
v
ed
in
YOL
O
V3
[
1
6
]
.
Acc
o
r
d
in
g
to
Yan
et
a
l.
[
1
1
]
,
f
o
r
ea
ch
b
o
u
n
d
in
g
b
o
x
,
in
o
r
d
er
to
esti
m
ate
th
e
o
b
jectn
ess
s
co
r
e,
YOL
O
V3
u
s
es
lo
g
is
tic
r
eg
r
ess
io
n
.
T
h
e
s
tr
u
ctu
r
e
o
f
YOL
O
V3
is
m
o
s
tly
m
ad
e
u
p
o
f
th
e
f
ea
tu
r
e
p
y
r
am
id
n
et
wo
r
k
(
FP
N)
an
d
th
e
d
ar
k
n
et
-
5
3
b
ac
k
b
o
n
e
[
1
7
]
.
I
n
d
ee
d
,
t
h
e
b
ac
k
b
o
n
e
n
etwo
r
k
a
r
ch
itectu
r
e
u
s
ed
in
YOL
O
V3
is
th
e
Dar
k
n
et
-
5
3
ar
ch
itectu
r
e
[
1
8
]
wh
ich
c
o
n
s
is
ts
o
f
5
3
co
n
v
o
lu
tio
n
al
lay
e
r
s
,
k
n
o
win
g
t
h
at
ea
ch
lay
er
u
s
es
a
leak
y
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
[
8
]
.
I
n
ad
d
itio
n
,
to
f
o
r
m
a
n
ch
o
r
b
o
x
es,
K
-
m
ea
n
s
clu
s
ter
in
g
is
u
s
ed
b
y
YOL
O
V3
[
1
9
]
an
d
h
as
th
e
ab
ilit
y
to
p
r
ed
ict
b
o
u
n
d
in
g
b
o
x
es
o
n
th
r
ee
d
if
f
er
en
t
s
ca
les
[
1
1
]
.
YOL
O
V3
d
if
f
er
s
f
r
o
m
o
th
er
o
b
ject
d
etec
tio
n
alg
o
r
ith
m
s
th
at
allo
w
r
eg
io
n
g
e
n
er
atio
n
b
ec
au
s
e
i
t
r
etu
r
n
s
t
h
e
d
i
r
ec
t
p
o
s
itio
n
o
f
th
e
b
o
u
n
d
in
g
b
o
x
an
d
its
ca
teg
o
r
y
b
y
u
s
in
g
th
e
en
tire
im
ag
e
as
th
e
n
etwo
r
k
'
s
in
p
u
t
[
1
9
]
.
YOL
O
V3
in
c
o
r
p
o
r
ates
th
e
latest
tech
n
o
lo
g
ies
s
u
c
h
as
u
p
-
s
am
p
lin
g
,
s
k
ip
-
c
o
n
n
e
ctio
n
,
an
d
r
es
id
u
al
b
l
o
ck
s
.
Dar
k
Net
-
5
3
is
a
b
le
to
o
b
tain
m
o
r
e
s
ca
lin
g
f
ea
tu
r
e
m
ap
s
,
s
u
ch
a
s
th
e
s
m
all,
m
ed
iu
m
,
an
d
la
r
g
e
s
ca
les
o
f
tar
g
et
f
ea
tu
r
es
[
1
9
]
.
Mu
lti
-
s
ca
le
p
r
ed
ictio
n
s
an
d
r
esid
u
al
n
et
wo
r
k
s
tr
u
ctu
r
e
ca
n
p
er
f
o
r
m
b
etter
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
im
p
r
o
v
e
th
e
m
ea
n
av
er
ag
e
p
r
ec
is
io
n
(
m
AP)
an
d
s
m
all
o
b
ject
d
etec
tio
n
r
esu
lts
[
1
9
]
.
I
n
o
r
d
er
to
d
etec
t
m
u
ltip
le
o
b
jects
with
d
if
f
er
en
t
class
es
in
th
e
s
am
e
g
r
id
an
d
t
r
ain
l
o
g
is
tic
class
if
ier
s
f
o
r
class
if
icatio
n
,
YOL
O
V3
u
s
es
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
.
T
h
at
e
n
ab
les to
lab
el
o
b
jects th
at
ar
e
in
th
e
s
am
e
g
r
i
d
[
8
]
.
3
.
4
.
YO
L
O
V
4
YOL
O
V4
is
th
e
f
ir
s
t v
er
s
io
n
o
f
YOL
O
cr
ea
ted
in
2
0
2
0
b
y
B
o
ch
k
o
v
s
k
iy
et
a
l.
a
n
d
in
wh
i
ch
R
ed
m
o
n
was
n
o
t
in
v
o
lv
ed
.
T
o
ac
h
ie
v
e
its
o
b
jectiv
e,
th
e
n
etwo
r
k
ar
ch
itectu
r
e
o
f
YOL
O
V4
u
s
es
s
ev
er
al
co
m
p
o
n
en
ts
(
[
1
1
]
,
[
2
0
]
)
.
L
i
k
e
YOL
O
V3
,
th
e
ar
c
h
itectu
r
e
o
f
YOL
O
V4
co
n
s
is
ts
o
f
5
3
co
n
v
o
lu
tio
n
al
lay
er
s
.
T
h
e
d
if
f
er
en
ce
b
etwe
en
b
o
th
ar
ch
i
tectu
r
es
is
th
at
YOL
O
V4
u
s
es
f
o
r
ea
c
h
lay
e
r
th
e
m
is
h
ac
ti
v
atio
n
f
u
n
ctio
n
an
d
allo
ws
th
e
u
s
ag
e
o
f
cr
o
s
s
-
s
tag
e
p
ar
tial
co
n
n
ec
tio
n
s
[
8
]
.
T
h
e
b
ac
k
b
o
n
e
in
clu
d
es
th
e
ar
c
h
itectu
r
e
o
f
cr
o
s
s
-
s
tag
e
p
ar
tial
C
SP
Dar
k
n
et
-
53
[
2
1
]
.
C
o
m
p
ar
ed
to
th
e
YOL
O
V3
ap
p
r
o
ac
h
,
th
e
lear
n
in
g
ca
p
ab
i
lity
o
f
th
e
C
NN
is
im
p
r
o
v
e
d
th
r
o
u
g
th
C
SP
Dar
k
n
et5
3
b
ac
k
b
o
n
e
m
o
d
el
[
8
]
.
Acc
o
r
d
in
g
to
Fah
im
an
d
Hasan
[
9
]
,
in
C
SP
Dar
k
Net5
3
,
th
e
f
ea
tu
r
e
m
ap
o
f
th
e
b
ase
lay
er
is
d
iv
id
ed
in
to
2
s
ec
tio
n
s
an
d
th
e
n
m
er
g
ed
b
y
u
s
in
g
a
cr
o
s
s
s
tag
e
p
ar
tial
n
etwo
r
k
(
C
SP
Net)
.
T
h
e
n
ec
k
in
clu
d
es
th
e
p
ath
-
ag
g
r
e
g
atio
n
n
etwo
r
k
s
(
P
ANe
t)
[
2
2
]
an
d
th
e
ad
d
itio
n
al
s
p
atial
p
y
r
am
id
p
o
o
lin
g
(
SP
P)
m
o
d
u
le
[
2
3
]
.
I
n
o
r
d
er
to
im
p
r
o
v
e
ac
c
u
r
ac
y
a
n
d
r
ed
u
ce
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
,
th
e
tr
ad
itio
n
al
f
ea
tu
r
e
p
y
r
am
i
d
n
etwo
r
k
(
FP
N)
n
ec
k
u
s
ed
in
YOL
O
V
3
h
as
b
ee
n
r
ep
lace
d
in
YOL
O
V4
b
y
th
e
PANet
n
ec
k
f
o
r
p
ar
a
m
eter
a
g
g
r
eg
atio
n
[
8
]
.
T
h
e
h
ea
d
in
cl
u
d
es
th
e
YOL
O
V3
an
ch
o
r
-
b
ased
ar
c
h
itectu
r
e.
Acc
o
r
d
in
g
to
X
u
et
a
l.
[
2
4
]
,
u
s
in
g
m
o
s
aic
d
ata
au
g
m
e
n
ta
tio
n
im
p
r
o
v
ed
lear
n
i
n
g
m
u
ch
b
etter
in
YOL
O
V4
.
T
h
is
tech
n
iq
u
e
allo
ws
co
m
b
in
in
g
4
o
r
ig
in
al
in
p
u
t
im
ag
es
in
o
r
d
er
to
cr
ea
te
a
n
ew
tr
ain
in
g
im
ag
e
s
o
th
at
th
e
n
etwo
r
k
lear
n
r
ec
o
g
n
izin
g
s
o
m
e
o
b
jects
th
at
ar
e
n
o
t
f
o
u
n
d
in
th
eir
u
s
u
al
en
v
ir
o
n
m
e
n
t
[
8
]
.
E
ac
h
im
ag
e
tak
es
u
p
a
q
u
ar
ter
o
f
t
h
e
f
in
al
im
ag
e,
an
d
th
e
o
b
ject
s
(
b
o
u
n
d
in
g
b
o
x
es)
ar
e
r
ea
d
ju
s
ted
ac
co
r
d
in
g
ly
in
o
r
d
er
t
o
in
cr
ea
s
e
th
e
v
ar
iety
o
f
d
ata
an
d
im
p
r
o
v
e
th
e
m
o
d
el'
s
ab
ilit
y
to
g
en
er
alize
.
Pro
v
id
in
g
an
id
ea
l
eq
u
ilib
r
i
u
m
b
etwe
en
o
b
ject
d
etec
tio
n
ac
c
u
r
ac
y
an
d
co
m
p
u
tatio
n
al
co
m
p
lex
ity
is
a
s
ig
n
if
ican
t
ass
et
f
o
r
YOL
O
V4
an
d
th
at
r
e
p
r
esen
ts
an
in
cr
em
en
tal
im
p
r
o
v
em
e
n
t
o
v
er
YOL
O
V3
[
8
]
.
T
h
e
im
p
lem
en
tatio
n
o
f
d
iv
er
s
e
t
r
a
in
in
g
tech
n
iq
u
es
t
h
at
ar
e
p
ar
t
o
f
th
e
YOL
O
V
4
n
etwo
r
k
a
r
ch
itectu
r
e
allo
ws
o
b
tain
in
g
an
im
p
r
o
v
e
d
p
er
f
o
r
m
an
ce
an
d
f
aster
p
r
o
ce
s
s
in
g
[
1
1
]
.
I
n
a
d
d
itio
n
to
th
at,
YOL
O
V4
is
d
es
ig
n
ed
to
b
e
tr
ain
ab
le
o
n
a
s
in
g
le
g
r
ap
h
i
cs
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
ca
r
d
,
m
ak
in
g
it
ac
ce
s
s
ib
le
to
r
esea
r
ch
er
s
with
lim
ited
r
eso
u
r
ce
s
.
3
.
5
.
YO
L
O
V
5
Glen
n
J
o
ch
er
in
tr
o
d
u
ce
d
t
h
e
YOL
O
V5
in
J
u
n
e
2
0
2
0
[
2
5
]
,
wh
ich
h
as
s
o
m
e
n
o
tab
le
im
p
r
o
v
em
en
ts
an
d
s
o
m
e
d
is
tin
ctio
n
s
[
9
]
.
T
h
e
k
ey
i
n
n
o
v
atio
n
o
f
YOL
O
V
5
is
th
at
it
i
s
th
e
1
s
t
v
er
s
io
n
o
f
YOL
O
b
u
ilt
u
s
in
g
Py
T
o
r
ch
in
p
y
th
o
n
an
d
n
o
t
b
ased
o
n
d
ar
k
n
et
ar
ch
i
tectu
r
e,
allo
win
g
f
o
r
a
m
o
r
e
s
tr
aig
h
tf
o
r
war
d
im
p
lem
en
tatio
n
p
r
o
ce
s
s
an
d
d
ev
elo
p
m
e
n
t
[
8
]
.
YOL
O
V5
in
teg
r
ated
th
e
an
ch
o
r
b
o
x
s
elec
tio
n
p
r
o
ce
s
s
[
2
6
]
.
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
.
1
,
Feb
r
u
ar
y
20
2
6
:
4
5
0
-
462
454
I
n
d
ee
d
,
f
o
r
o
b
jects
in
th
e
d
at
aset,
th
e
d
is
tr
ib
u
tio
n
o
f
th
e
b
o
u
n
d
in
g
b
o
x
lo
ca
tio
n
co
n
tr
ib
u
tes
to
au
to
lear
n
i
n
g
an
ch
o
r
b
o
x
es
[
8
]
.
Sev
er
al
n
etwo
r
k
ar
ch
itectu
r
es
ar
e
p
r
o
p
o
s
ed
b
y
YOL
O
V5
an
d
ar
e
s
u
itab
le
f
o
r
s
ev
er
al
s
ce
n
ar
io
s
with
v
ar
io
u
s
in
p
u
t
s
izes
s
u
ch
as
Y
OL
O
V5
s
,
YOL
O
V5
m
an
d
YOL
O
V5
l
[
1
1
]
.
I
n
ea
ch
o
f
th
ese
m
o
d
els
th
e
s
tr
u
ctu
r
e
o
f
YOL
O
V5
d
if
f
e
r
s
g
r
ea
tly
in
ter
m
s
o
f
wid
th
an
d
d
ep
t
h
o
f
ea
c
h
co
n
v
o
lu
tio
n
al
m
o
d
u
le
[
8
]
.
Acc
o
r
d
in
g
to
Ah
m
ed
et
a
l.
[
7
]
,
th
ese
v
a
r
ian
ts
o
f
YOL
O
V5
o
f
f
e
r
v
ar
y
in
g
p
er
f
o
r
m
a
n
ce
an
d
d
etec
tio
n
ac
cu
r
ac
y
,
ac
h
iev
ed
b
y
ad
j
u
s
tin
g
lay
er
co
u
n
t a
n
d
n
etwo
r
k
d
e
p
th
.
T
h
er
e
is
a
g
r
ea
t
s
im
ilar
ity
b
etwe
en
th
e
ar
ch
itectu
r
e
o
f
YOL
O
V5
an
d
th
at
o
f
YOL
O
V
4
[
1
1
]
.
T
h
e
th
r
ee
m
ain
co
m
p
o
n
e
n
ts
o
f
t
h
e
ar
ch
itectu
r
e
o
f
YOL
O
V5
a
r
e:
th
e
h
e
ad
f
o
r
p
r
ed
ictin
g
cl
ass
es
an
d
b
o
u
n
d
in
g
b
o
x
es,
th
e
PANet
n
ec
k
f
o
r
co
l
lectin
g
f
ea
tu
r
e
m
ap
s
,
an
d
th
e
cr
o
s
s
s
tag
e
p
ar
tial
(
C
SP
)
b
ac
k
b
o
n
e
f
o
r
ex
tr
ac
tin
g
im
ag
e
f
ea
tu
r
es
(
[
8
]
,
[
1
3
]
,
[
2
7
]
)
.
YOL
O
V5
u
s
es
a
m
o
d
if
ie
d
cr
o
s
s
s
tag
e
p
ar
tial
d
a
r
k
n
et
-
5
3
(
C
SP
Dar
k
Net5
3
)
b
ac
k
b
o
n
e
[
8
]
.
I
n
o
r
d
er
to
d
etec
t
o
b
jects
o
f
v
a
r
io
u
s
s
izes,
C
SP
Dar
k
Net5
3
p
r
o
p
o
s
es
5
3
co
n
v
o
l
u
tio
n
al
lay
e
r
s
an
d
g
e
n
er
ates f
ea
tu
r
es a
t m
u
lti
p
le
s
ca
les with
a
m
o
d
if
ied
PANet
[
8
]
.
3
.
6
.
YO
L
O
V
6
I
n
2
0
2
2
,
L
i
et
a
l
.
[
2
8
]
d
ev
elo
p
ed
an
d
lau
n
c
h
ed
th
e
YOL
O
V6
alg
o
r
ith
m
.
T
h
e
au
th
o
r
s
f
o
cu
s
ed
o
n
cr
ea
tin
g
an
o
b
ject
d
etec
to
r
wit
h
an
in
d
u
s
tr
y
f
o
c
u
s
as
a
d
esig
n
s
tr
ateg
y
[
3
0
]
.
T
h
er
e
ar
e
d
if
f
er
en
t
co
n
f
i
g
u
r
atio
n
s
o
f
YOL
O
V6
s
u
ch
as
YOL
O
V6
n
,
YOL
O
V6
s
,
YOL
O
V6
m
,
YOL
O
V6
l
an
d
YOL
O
V6
l6
t
h
at
allo
w
ad
ap
tin
g
to
d
if
f
er
en
t a
p
p
licatio
n
s
ce
n
ar
io
s
.
I
n
th
at
alg
o
r
ith
m
,
th
e
au
th
o
r
s
p
r
esen
ted
a
n
u
p
d
ated
r
ep
ar
am
etr
ized
b
ac
k
b
o
n
e
an
d
n
ec
k
,
p
r
o
p
o
s
ed
as
th
e
ef
f
icien
tR
ep
b
ac
k
b
o
n
e
an
d
th
e
r
e
p
r
esen
tatio
n
p
ath
a
g
g
r
eg
atio
n
n
etwo
r
k
(
R
ep
-
PAN)
n
ec
k
(
[
9
]
,
[
2
9
]
)
.
Un
lik
e
YOL
O
V5
,
YOL
O
V
6
f
ea
tu
r
es
an
an
ch
o
r
-
f
r
ee
d
esi
g
n
[
7
]
m
ak
i
n
g
it
5
1
p
er
ce
n
t
q
u
ick
er
t
h
an
a
n
ch
o
r
-
b
ased
m
eth
o
d
s
[
2
9
]
.
T
h
e
r
ep
ar
am
etr
ized
b
ac
k
b
o
n
e
wit
h
C
SP
an
d
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
(
VGG)
b
ac
k
b
o
n
es
is
u
s
ed
in
th
e
‘
‘
m
’
’
,
‘
‘
l
’
’
an
d
‘
‘
l
6
’
’
v
a
r
ian
ts
,
an
d
‘
‘
n
’
’
an
d
‘
‘
s
’
’
v
ar
ian
ts
r
esp
ec
tiv
ely
[
7
]
.
T
h
e
Nec
k
o
f
YOL
O
V6
is
s
im
ilar
to
YO
L
O
V5
,
b
u
t
b
r
ea
k
i
n
g
t
h
e
co
n
v
en
tio
n
,
th
e
h
ea
d
is
ef
f
icien
tly
d
ec
o
u
p
led
,
r
e
d
u
ci
n
g
co
m
p
u
tatio
n
an
d
in
c
r
ea
s
in
g
p
r
ec
is
io
n
b
y
p
r
ev
e
n
tin
g
p
ar
am
eter
s
h
ar
in
g
b
etwe
en
th
e
d
etec
tio
n
an
d
clas
s
if
icatio
n
b
r
an
ch
es
[
7
]
.
A
two
-
lo
s
s
f
u
n
ctio
n
is
r
eq
u
ir
ed
b
y
YOL
O
V6
.
Dis
tr
ib
u
tio
n
f
o
ca
l
lo
s
s
(
DFL)
[
3
0
]
wh
ic
h
allo
ws
to
b
etter
lear
n
w
h
er
e
to
p
lace
b
o
x
es,
with
m
o
r
e
f
in
ess
e
an
d
p
r
ec
is
io
n
,
b
y
t
r
an
s
f
o
r
m
i
n
g
o
n
d
is
cr
ete
d
is
tr
ib
u
tio
n
s
,
r
eg
r
ess
io
n
in
to
a
class
if
icatio
n
task
.
An
d
v
ar
i
f
o
ca
l
lo
s
s
(
VFL)
[
1
9
]
wh
ic
h
is
u
s
ed
as
th
e
class
if
icatio
n
lo
s
s
,
alo
n
g
with
SIo
U/GI
o
U
(
Scy
ll
a
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
/
g
e
n
e
r
alize
d
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
)
[
3
1
]
as
r
eg
r
ess
io
n
lo
s
s
[
2
9
]
.
Ad
d
itio
n
al
im
p
r
o
v
em
en
ts
o
f
YOL
O
V6
f
o
cu
s
ed
o
n
in
d
u
s
tr
ial
ap
p
licatio
n
s
in
clu
d
e
t
h
e
u
s
e
o
f
k
n
o
wled
g
e
d
is
till
atio
n
[
3
2
]
,
i
n
v
o
lv
in
g
a
teac
h
er
-
s
tu
d
en
t
m
o
d
el.
T
h
e
p
r
in
cip
le
o
f
th
is
m
o
d
el
is
th
at
a
s
tu
d
en
t
m
o
d
el
is
tr
ain
ed
b
y
a
teac
h
er
m
o
d
el
b
ec
a
u
s
e
in
o
r
d
er
t
o
tr
a
in
th
e
s
tu
d
en
t,
th
e
p
r
ed
ictio
n
s
o
f
th
e
teac
h
er
a
r
e
u
s
ed
as
s
o
f
t
la
b
els
alo
n
g
with
th
e
g
r
o
u
n
d
tr
u
t
h
[
2
9
]
.
T
h
is
co
m
es
with
o
u
t
f
u
elin
g
th
e
c
o
m
p
u
tatio
n
al
co
s
t
b
ec
au
s
e
b
ased
o
n
[
2
9
]
,
t
h
e
ai
m
is
to
r
ep
licate
th
e
p
o
wer
f
u
l
p
er
f
o
r
m
an
ce
o
f
th
e
lar
g
er
“
t
ea
ch
er
”
m
o
d
el
in
a
s
m
aller
“
s
tu
d
en
t
”
m
o
d
el.
3
.
7
.
YO
L
O
V
7
T
o
im
p
lem
en
t
YOL
O
V7
,
W
an
g
et
a
l.
[
3
3
]
b
u
ilt
o
n
p
r
ev
i
o
u
s
v
er
s
io
n
s
o
f
YOL
O
an
d
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
its
ar
ch
itectu
r
e.
T
h
e
YOL
O
V7
m
o
d
el
co
n
s
is
ts
o
f
s
ev
en
v
ar
ian
ts
:
P6
m
o
d
els
(
d
6
,
e6
,
w
6
,
an
d
e6
e)
an
d
P5
m
o
d
els
(
v
7
,
v
7
x
,
a
n
d
v
7
-
tin
y
)
[
7
]
.
Acc
o
r
d
in
g
to
Yan
et
a
l.
[
1
1
]
,
th
e
ef
f
icien
t
lay
er
ag
g
r
e
g
atio
n
n
etwo
r
k
(
E
-
L
AN)
r
ep
r
esen
ts
th
e
f
o
u
n
d
atio
n
o
f
YOL
O
V7
'
s
ar
ch
itectu
r
e,
wh
ich
aim
s
to
o
p
tim
ize
th
e
in
f
er
en
ce
s
p
ee
d
[
9
]
an
d
s
et
u
p
a
g
o
o
d
n
etwo
r
k
b
y
co
n
tr
o
llin
g
th
e
lo
n
g
est
an
d
s
h
o
r
test
g
r
ad
ien
ts
,
en
a
b
lin
g
d
ee
p
er
n
etwo
r
k
s
to
lea
r
n
ef
f
e
ctiv
ely
an
d
c
o
n
v
e
r
g
e
[
1
1
]
.
W
an
g
et
a
l.
[
3
3
]
also
im
p
r
o
v
e
d
th
e
ar
ch
itectu
r
e
o
f
YOL
O
V7
b
y
u
s
in
g
co
m
p
u
ta
tio
n
al
b
lo
ck
in
th
e
YOL
O
V
7
’
s
b
ac
k
b
o
n
e
in
o
r
d
er
to
in
tr
o
d
u
ce
th
e
ex
ten
d
ed
ef
f
icien
t
lay
er
ag
g
r
eg
atio
n
n
e
two
r
k
(
E
-
E
L
AN)
wh
ich
u
s
es
m
er
g
e,
s
h
u
f
f
le
an
d
ex
p
a
n
d
ca
r
d
in
ality
with
o
u
t
d
is
r
u
p
tin
g
th
e
g
r
ad
ie
n
t p
ath
,
i
n
o
r
d
e
r
to
im
p
r
o
v
e
n
etwo
r
k
le
ar
n
in
g
[
7
]
.
YOL
O
V7
u
s
es
C
S
PDar
k
Net
-
5
3
b
ac
k
b
o
n
e.
T
o
im
p
r
o
v
e
g
r
ad
ien
t
f
lo
w
d
u
r
in
g
tr
ai
n
in
g
,
th
e
C
SP
Dar
k
Net
-
5
3
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
5
3
co
n
v
o
lu
tio
n
al
la
y
er
s
with
leak
y
r
ec
tifie
d
lin
ea
r
u
n
it (
L
ea
k
y
R
eL
U)
ac
tiv
atio
n
s
an
d
well
-
ch
o
s
en
f
ilter
s
izes
[
9
]
.
Acc
o
r
d
in
g
to
t
h
e
s
am
e
r
esear
c
h
er
s
,
th
is
b
ac
k
b
o
n
e
is
en
h
a
n
ce
d
with
th
e
m
o
d
u
le
o
f
SP
P
to
ex
t
r
ac
t
m
u
lti
-
s
ca
le
f
ea
tu
r
es
c
r
u
ci
al
ef
f
ec
tiv
ely
f
o
r
o
b
ject
d
etec
tio
n
.
R
eg
ar
d
i
n
g
th
e
h
ea
d
co
m
p
o
n
e
n
t,
YOL
O
V7
u
s
es
th
e
co
n
v
e
n
tio
n
al
s
h
ar
e
d
-
f
ea
tu
r
e
ap
p
r
o
ac
h
.
[
9
]
af
f
ir
m
t
h
at
in
th
is
ap
p
r
o
ac
h
,
b
ef
o
r
e
r
e
d
ir
ec
tin
g
f
ea
tu
r
es
to
s
ep
ar
ate
b
r
an
ch
es
f
o
r
class
p
r
o
b
ab
ilit
y
class
if
icatio
n
an
d
b
o
u
n
d
in
g
b
o
x
co
o
r
d
in
ate
r
eg
r
ess
io
n
,
th
e
ch
ar
ac
ter
is
tics
ar
e
tr
ea
ted
th
r
o
u
g
h
a
s
er
ies o
f
co
n
v
o
lu
tio
n
al
lay
e
r
s
.
3
.
8
.
YO
L
O
V
8
I
n
2
0
2
3
,
Ultr
aly
tics
in
tr
o
d
u
ce
d
YOL
O
V8
an
d
p
r
o
p
o
s
ed
it
in
5
v
er
s
io
n
s
(
n
,
s
,
m
,
l,
an
d
x
)
[
3
4
]
.
I
t
d
eliv
er
s
s
o
m
e
o
f
th
e
m
o
s
t a
d
v
an
ce
d
p
er
f
o
r
m
a
n
ce
to
d
ate
[
3
5
]
.
I
t
i
s
d
ev
elo
p
e
d
b
y
Py
T
o
r
c
h
an
d
o
f
f
er
s
r
ea
l
-
tim
e
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
F
r
o
m
YOLO V
1
to
YOLO V
1
1
:
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f
YOLO
…
(
I
ma
n
e
B
eq
q
a
li Ha
s
s
a
n
i
)
455
p
r
ed
ictio
n
o
f
o
b
ject
b
o
u
n
d
in
g
b
o
x
es
an
d
class
p
r
o
b
ab
ilit
ies
u
s
in
g
a
f
u
lly
co
n
v
o
l
u
tio
n
al
n
e
two
r
k
to
p
r
o
v
id
e
a
s
in
g
le
-
s
tag
e
o
b
ject
d
etec
tio
n
m
o
d
el
[
3
6
]
.
Acc
o
r
d
in
g
to
Ya
n
et
a
l.
[
1
1
]
,
it
co
n
tain
s
m
o
s
t
o
f
th
e
im
p
r
o
v
em
en
ts
f
r
o
m
p
r
ev
io
u
s
v
er
s
io
n
s
o
f
Y
OL
O.
YOL
O
V8
d
o
es
n
o
t
u
s
e
an
ch
o
r
s
[
9
]
.
An
c
h
o
r
-
f
r
ee
d
etec
tio
n
r
ed
u
ce
s
th
e
n
u
m
b
er
o
f
b
o
x
p
r
ed
ictio
n
s
,
wh
ich
s
p
ee
d
s
u
p
n
o
n
-
m
ax
im
al
s
u
p
p
r
ess
io
n
(
NM
S)
[
9
]
.
T
h
e
m
o
d
if
ied
v
er
s
io
n
o
f
th
e
C
SP
Dar
k
n
et5
3
ar
ch
itectu
r
e,
E
-
E
L
AN,
f
o
r
m
s
th
e
b
asis
o
f
th
e
b
ac
k
b
o
n
e
o
f
YOL
O
V8
an
d
th
e
ef
f
icien
t
Py
T
o
r
ch
im
p
le
m
en
tatio
n
im
p
r
o
v
es
th
e
s
p
ee
d
o
f
in
f
e
r
en
ce
an
d
tr
ain
in
g
[
1
1
]
.
T
h
e
YOL
O
-
S
PP
-
B
o
o
s
t
b
ac
k
b
o
n
e
is
in
tr
o
d
u
ce
d
in
YOL
O
V8
.
D
u
r
in
g
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
e
YOL
O
-
SPP
-
B
o
o
s
t
b
ac
k
b
o
n
e
in
teg
r
ates
r
esid
u
al
co
n
n
ec
tio
n
s
b
etwe
en
co
n
v
o
lu
t
io
n
al
lay
er
s
to
f
ac
ilit
ate
s
m
o
o
th
g
r
a
d
ien
t
f
lo
w
[
9
]
.
Fo
r
ef
f
ic
ien
t
f
ea
tu
r
e
f
u
s
io
n
,
YOL
O
V8
u
s
es
a
Fo
cu
s
-
V3
n
ec
k
.
An
d
f
o
r
m
u
lti
-
s
ca
le
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
is
v
er
s
io
n
o
f
YOL
O
u
s
es
an
SP
P
m
o
d
u
le
[
9
]
.
T
h
e
h
ea
d
o
f
YOL
O
V8
ad
o
p
ts
d
ec
o
u
p
le
d
h
ea
d
s
an
d
b
r
ea
k
s
th
e
co
n
v
e
n
tio
n
al
ap
p
r
o
ac
h
o
f
YOL
O
V7
[
9
]
.
T
h
is
in
v
o
lv
es
th
at
b
ef
o
r
e
m
ak
in
g
th
e
f
in
al
p
r
e
d
ictio
n
s
,
th
er
e
m
u
s
t
b
e
a
s
ep
ar
atio
n
o
f
f
ea
tu
r
es.
Acc
o
r
d
in
g
to
[
9
]
,
YOL
O
V8
f
u
r
th
e
r
in
c
r
ea
s
es
ac
cu
r
ac
y
b
y
u
s
in
g
an
c
h
o
r
-
f
r
ee
d
esig
n
.
T
h
at
ap
p
r
o
ac
h
h
elp
s
im
p
r
o
v
i
n
g
d
etec
tio
n
ac
cu
r
ac
y
an
d
f
lex
ib
ilit
y
[
3
4
]
.
I
t
ca
n
d
ir
ec
tly
p
r
ed
ict
th
e
d
im
en
s
io
n
s
o
f
th
e
b
o
u
n
d
in
g
b
o
x
an
d
th
e
ce
n
ter
p
o
in
t i
n
o
r
d
er
to
s
im
p
lif
y
th
e
n
etwo
r
k
ar
c
h
itectu
r
e
wh
i
le
im
p
r
o
v
in
g
lo
ca
lizatio
n
ac
c
u
r
ac
y
[
9
]
.
3
.
9
.
YO
L
O
V
9
I
n
Feb
r
u
a
r
y
2
0
2
4
,
W
an
g
et
a
l.
[
3
7
]
in
tr
o
d
u
ce
d
YOL
O
V9
.
T
h
at
v
er
s
io
n
o
f
YOL
O,
h
as
im
p
r
o
v
e
d
its
n
etwo
r
k
ar
ch
itectu
r
e
wh
ic
h
is
b
ased
o
n
th
e
r
o
b
u
s
t
co
d
e
b
ase
p
r
o
v
id
e
d
b
y
Ultr
aly
tics
YOL
O
V5
[
3
8
]
in
o
r
d
e
r
to
p
er
f
o
r
m
b
etter
i
n
r
ec
o
g
n
izi
n
g
co
m
p
lex
tar
g
ets
an
d
with
v
ar
io
u
s
s
izes
[
3
9
]
.
I
n
ad
d
itio
n
t
o
th
at,
t
h
e
en
h
an
ce
d
m
o
d
el
d
em
o
n
s
tr
ates
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
m
u
lti
-
ta
r
g
et
s
ce
n
ar
io
s
an
d
h
an
d
lin
g
o
cc
lu
s
io
n
s
,
wh
ich
m
ak
es
its
u
s
e
m
o
r
e
r
eliab
le
at
co
m
p
lex
in
ter
s
ec
tio
n
s
[
3
9
]
.
R
ely
in
g
o
n
f
ewe
r
ca
lc
u
latio
n
s
an
d
p
ar
am
ete
r
s
,
YOL
O
V9
m
a
y
h
av
e
th
e
s
am
e
o
r
b
etter
d
etec
tio
n
r
esu
lts
th
an
p
r
ev
io
u
s
YOL
O
alg
o
r
ith
m
s
b
y
ex
tr
ac
tin
g
with
p
r
ec
is
io
n
an
d
r
etain
in
g
i
n
f
o
r
m
atio
n
r
eq
u
i
r
ed
to
m
ap
d
ata
t
o
t
ar
g
ets
[
4
0
]
.
YOL
O
V9
p
r
esen
ts
two
k
ey
in
n
o
v
atio
n
s
:
T
o
im
p
r
o
v
e
p
a
r
am
eter
u
tili
za
tio
n
ef
f
icien
c
y
,
th
e
f
ir
s
t
in
n
o
v
atio
n
o
f
YOL
O
V9
,
r
ep
r
esen
ted
b
y
a
n
ew
g
en
e
r
alize
d
ef
f
icien
t
lay
er
ag
g
r
eg
atio
n
n
et
wo
r
k
(
GE
L
AN)
,
is
u
s
ed
[
4
1
]
.
T
h
e
s
ec
o
n
d
in
n
o
v
atio
n
o
f
YOL
O
V
9
is
a
n
o
v
e
l
co
n
ce
p
t
o
f
p
r
o
g
r
am
m
ab
le
g
r
ad
ien
t
in
f
o
r
m
atio
n
(
PGI
)
f
r
am
ewo
r
k
[
4
2
]
to
p
r
o
p
ag
ate
m
u
lti
-
lev
el
g
r
ad
ien
t
in
f
o
r
m
atio
n
th
r
o
u
g
h
a
n
a
u
x
iliar
y
r
e
v
er
s
ib
le
b
r
an
ch
[
4
1
]
in
r
esp
o
n
s
e
to
t
h
e
p
r
o
b
le
m
o
f
in
f
o
r
m
atio
n
lo
s
s
in
d
ee
p
n
etwo
r
k
d
ata
tr
a
n
s
m
is
s
io
n
[
4
0
]
.
Sh
ar
m
a
et
a
l.
[
3
6
]
af
f
ir
m
th
at
th
o
s
e
in
n
o
v
atio
n
s
ca
n
ad
d
r
ess
is
s
u
es
r
elate
d
to
co
m
p
u
tatio
n
al
e
f
f
icien
cy
ef
f
ec
tiv
ely
an
d
in
f
o
r
m
atio
n
lo
s
s
an
d
ac
c
o
r
d
i
n
g
to
Ma
r
c
h
i
et
a
l.
[
3
9
]
,
it
r
ep
r
esen
ts
a
s
ig
n
if
ica
n
t
leap
f
o
r
war
d
in
te
r
m
s
o
f
ef
f
icien
cy
,
ac
cu
r
ac
y
,
an
d
ad
ap
tab
ilit
y
.
I
n
d
ee
d
,
p
u
r
s
u
in
g
a
tr
ain
f
r
o
m
s
cr
atch
s
tr
ateg
y
,
W
an
g
et
a
l.
[
3
7
]
o
b
tain
ed
b
etter
d
etec
tio
n
r
esu
lts
th
an
s
tate
o
f
th
e
ar
t m
o
d
els p
r
e
-
tr
ain
ed
with
lar
g
e
d
atasets
[
4
1
]
.
3
.
1
0
.
YO
L
O
V
10
C
r
ea
ted
b
y
W
an
g
et
a
l.
in
2
0
2
4
[
4
3
]
,
th
e
YOL
O
V1
0
o
p
ti
m
ized
th
e
s
p
ee
d
an
d
ac
cu
r
ac
y
o
f
o
b
ject
lo
ca
lizatio
n
in
im
a
g
es
an
d
ca
t
eg
o
r
izatio
n
co
m
p
a
r
ed
t
o
its
ea
r
lier
v
er
s
io
n
s
.
T
h
er
e
ar
e
d
if
f
er
en
t
co
n
f
ig
u
r
atio
n
s
o
f
YOL
O
V1
0
s
u
ch
as
YOL
O
V1
0
n
,
YOL
OV1
0
s
,
YOL
O
V1
0
m
,
YOL
O
V1
0
b
,
YOL
OV1
0
l
an
d
YOL
O
V1
0
x
th
at
allo
w
ad
a
p
tin
g
to
d
if
f
er
en
t
ap
p
licatio
n
s
ce
n
ar
io
s
[
4
4
]
.
YOL
O
V1
0
aim
s
t
o
f
u
r
th
er
a
d
v
an
ce
th
e
p
er
f
o
r
m
an
ce
-
ef
f
icien
cy
b
o
u
n
d
ar
y
o
f
YOL
Os
f
r
o
m
b
o
t
h
th
e
m
o
d
el
ar
ch
itectu
r
e
a
n
d
th
e
p
o
s
t
-
p
r
o
ce
s
s
in
g
[
4
3
]
.
Acc
o
r
d
in
g
to
Sap
k
o
ta
et
a
l.
[
4
4
]
,
YOL
O
V1
0
ef
f
ec
tiv
ely
ad
d
r
ess
es
p
r
ev
io
u
s
ar
ch
itectu
r
al
lim
itatio
n
s
an
d
th
e
r
elian
ce
o
n
NM
S,
wh
ich
is
a
s
ig
n
if
ican
t
s
tep
f
o
r
war
d
in
en
h
an
cin
g
in
f
er
en
ce
s
p
ee
d
,
p
er
f
o
r
m
an
ce
an
d
o
p
er
atio
n
al
ef
f
icien
c
y
.
I
t
p
r
o
v
id
es
two
m
ain
im
p
r
o
v
em
e
n
ts
o
v
er
p
r
ev
i
o
u
s
v
er
s
io
n
s
o
f
YO
L
O:
a
m
o
d
el
d
esig
n
f
o
cu
s
ed
o
n
ac
cu
r
ac
y
an
d
ef
f
i
cien
cy
(
to
in
cr
ea
s
e
o
v
er
all
p
e
r
f
o
r
m
a
n
ce
)
an
d
a
co
n
s
is
ten
t
d
u
al
ass
ig
n
m
en
t
(
in
o
r
d
er
to
h
av
e
an
NM
S
-
f
r
ee
tr
ain
in
g
p
r
o
to
c
o
l)
[
4
3
]
.
T
h
e
NM
S
-
f
r
ee
tr
ain
in
g
p
r
o
t
o
co
l
th
r
o
u
g
h
co
n
s
is
ten
t
d
u
al
ass
ig
n
m
en
ts
s
im
p
lifie
s
th
e
o
u
tp
u
t
s
tag
e,
r
ed
u
ce
s
p
o
s
t
p
r
o
ce
s
s
in
g
tim
e
[
4
4
]
a
n
d
b
r
in
g
s
th
e
co
m
p
etitiv
e
p
er
f
o
r
m
an
ce
an
d
lo
w
in
f
er
en
c
e
laten
cy
[
4
3
]
.
T
h
e
ar
c
h
itectu
r
al
en
h
an
ce
m
en
ts
in
YOL
O
V1
0
in
clu
d
e
th
e
s
p
atial
-
ch
a
n
n
el
d
ec
o
u
p
led
d
o
wn
s
am
p
lin
g
,
th
e
im
p
lem
en
tatio
n
o
f
lig
h
tweig
h
t
class
if
icati
o
n
h
ea
d
s
,
an
d
r
a
n
k
-
g
u
id
ed
b
lo
ck
d
esig
n
,
ea
c
h
co
n
tr
ib
u
tin
g
to
s
u
b
s
tan
tial
r
ed
u
ctio
n
s
in
p
ar
am
eter
c
o
u
n
t
a
n
d
co
m
p
u
tatio
n
al
d
em
an
d
s
[
4
5
]
.
YOL
O
V1
0
h
as
im
p
r
o
v
e
d
its
b
ac
k
b
o
n
e
b
y
u
s
i
n
g
an
u
p
d
ated
v
er
s
io
n
o
f
cr
o
s
s
s
tag
e
p
ar
tial
n
etwo
r
k
(
C
SP
Net)
[
4
4
]
.
I
n
o
r
d
er
to
h
av
e
ef
f
icien
t
an
d
ac
cu
r
ate
f
e
atu
r
e
ex
tr
ac
tio
n
,
t
h
is
u
p
d
ate
i
s
d
esig
n
ed
to
r
ed
u
ce
c
o
m
p
u
t
atio
n
al
r
ed
u
n
d
a
n
cy
an
d
im
p
r
o
v
e
g
r
ad
ien
t
f
lo
w
[
4
4
]
.
R
eg
ar
d
in
g
t
h
e
n
ec
k
o
f
th
e
a
r
ch
itectu
r
e,
it
in
co
r
p
o
r
ate
d
a
p
ath
ag
g
r
eg
atio
n
n
etwo
r
k
t
o
f
ac
ilit
ate
ef
f
icien
t
m
u
lti
-
s
ca
le
f
ea
tu
r
e
f
u
s
io
n
[
4
4
]
.
T
h
is
o
p
tio
n
is
ess
en
tial
b
ec
au
s
e
it
im
p
r
o
v
es
th
e
al
g
o
r
ith
m
'
s
ab
ilit
y
to
d
etec
t
o
b
jects
with
g
r
ea
ter
ac
cu
r
ac
y
an
d
f
o
r
d
if
f
er
en
t
s
izes
[
4
4
]
.
As
f
o
r
th
e
h
ea
d
,
Y
OL
O
V1
0
u
s
es
a
d
u
al
-
h
ea
d
d
esig
n
(
one
-
to
-
m
a
n
y
h
ea
d
an
d
o
n
e
-
to
-
o
n
e
h
ea
d
)
.
On
e
-
to
-
Ma
n
y
h
ea
d
g
en
e
r
ates
m
u
ltip
le
p
r
ed
ictio
n
s
p
er
o
b
ject
to
im
p
r
o
v
e
lear
n
in
g
ac
cu
r
ac
y
b
y
p
r
o
v
id
in
g
r
ich
s
u
p
er
v
is
io
n
s
ig
n
als
[
4
4
]
.
On
e
-
to
-
On
e
h
ea
d
d
eliv
e
r
s
an
o
p
tim
al
a
n
d
s
in
g
le
p
r
ed
icti
o
n
p
er
o
b
ject.
T
h
is
elim
in
ates
th
e
n
ee
d
f
o
r
NM
S
an
d
th
u
s
r
ed
u
ce
s
laten
cy
an
d
s
im
p
lifie
s
th
e
d
etec
tio
n
p
r
o
ce
s
s
[
4
4
]
.
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
.
1
,
Feb
r
u
ar
y
20
2
6
:
4
5
0
-
462
456
3
.
1
1
.
YO
L
O
V
11
I
n
tr
o
d
u
ce
d
b
y
Ultr
aly
tics
in
2
0
2
4
,
YOL
O
V1
1
is
th
e
latest
ad
d
itio
n
to
t
h
e
YOL
O
s
er
ies
alg
o
r
ith
m
s
b
u
ild
in
g
o
n
th
e
f
o
u
n
d
atio
n
o
f
YOL
O
V8
to
d
ate
[
4
6
]
an
d
was
an
n
o
u
n
ce
d
at
th
e
YOL
O
Vis
io
n
2
0
2
4
co
n
f
er
e
n
ce
[
4
6
]
.
T
h
er
e
ar
e
d
if
f
er
en
t m
o
d
els o
f
YOL
O
V1
1
with
a
s
ca
le
th
at
v
ar
ies f
r
o
m
s
m
all
to
lar
g
e
s
u
ch
as
YOL
O
V1
1
n
,
YOL
O
V1
1
s
,
YOL
O
V1
1
m
,
YOL
O
V1
1
l
an
d
YOL
O
V1
1
x
th
at
allo
w
ad
ap
tin
g
to
d
if
f
er
en
t
ap
p
licatio
n
s
ce
n
ar
io
s
[
4
6
]
.
O
p
tim
izin
g
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
m
u
ltip
le
co
m
p
u
ter
v
is
io
n
task
s
is
o
n
e
o
f
t
h
e
o
b
jectiv
es
o
f
th
e
tr
ain
in
g
p
r
o
c
ess
o
f
YOL
O
V1
1
,
ju
s
t
lik
e
f
o
r
YOL
O
V1
0
[
4
4
]
.
T
h
is
latest
v
er
s
io
n
o
f
YOL
O,
b
r
in
g
s
s
o
lid
im
p
r
o
v
em
en
ts
in
th
e
ar
c
h
itectu
r
e
a
n
d
tr
ain
in
g
m
eth
o
d
s
th
at
allo
w
t
o
b
r
in
g
ev
en
m
o
r
e
s
p
ee
d
,
ac
cu
r
ac
y
a
n
d
e
f
f
icien
cy
[
4
6
]
.
T
o
ca
p
tu
r
e
co
m
p
lex
d
etails
in
im
ag
es,
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
ab
ilit
ies
h
av
e
b
ee
n
o
p
tim
ized
in
th
e
YOL
O
V1
1
ar
ch
itectu
r
e
[
4
4
]
.
I
n
d
ee
d
,
ac
c
o
r
d
in
g
to
Ultr
aly
tics
[
4
7
]
,
th
i
s
v
er
s
io
n
en
h
an
ce
s
s
p
ee
d
an
d
ef
f
icien
cy
th
r
o
u
g
h
tr
ain
in
g
p
ip
elin
es
a
n
d
o
p
tim
iz
ed
d
esig
n
s
,
b
alan
ci
n
g
p
r
ec
is
io
n
an
d
p
er
f
o
r
m
an
ce
an
d
im
p
r
o
v
es
f
ea
tu
r
e
ex
tr
ac
ti
o
n
with
an
ad
v
an
ce
d
b
ac
k
b
o
n
e
an
d
n
ec
k
ar
ch
itectu
r
e
YOL
O
V1
1
f
ea
tu
r
es
th
e
co
n
v
o
l
u
tio
n
al
b
lo
c
k
with
p
a
r
allel
s
p
atial
atten
tio
n
(
C
2
PS
A
(
cr
o
s
s
s
tag
e
p
ar
tial
with
s
p
atial
atten
tio
n
)
)
co
m
p
o
n
en
ts
,
s
p
atial
p
y
r
am
id
p
o
o
lin
g
f
ast
(
SP
PF
)
an
d
a
cr
o
s
s
-
s
tag
e
p
o
r
tio
n
o
f
k
er
n
el
s
ize
2
(
C
3
k
2
)
b
lo
ck
,
in
o
r
d
er
t
o
im
p
r
o
v
e
th
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
ca
p
a
b
ilit
ies o
f
th
e
m
o
d
el
[
4
6
]
.
Acc
o
r
d
in
g
to
Sap
k
o
ta
et
a
l.
[
4
4
]
,
YOL
O
V1
1
allo
ws
h
av
i
n
g
b
etter
r
esu
lts
o
n
b
e
n
ch
m
a
r
k
d
atasets
b
ec
au
s
e
it
u
s
es
en
h
a
n
ce
d
t
r
ain
in
g
tech
n
i
q
u
es.
I
n
d
ee
d
,
o
n
th
e
C
OC
O
d
ataset,
b
y
u
s
in
g
2
2
%
f
ewe
r
p
a
r
am
eter
s
co
m
p
ar
ed
to
th
e
YOL
O
V8
m
alg
o
r
ith
m
,
YOL
O
V1
1
m
ac
h
ie
v
ed
a
m
AP sco
r
e
o
f
9
5
%
[
4
4
]
.
T
h
is
d
em
o
n
s
tr
ates
g
r
ea
ter
ef
f
icien
c
y
with
o
u
t
co
m
p
r
o
m
is
in
g
ac
c
u
r
ac
y
.
T
h
r
o
u
g
h
an
av
e
r
ag
e
in
f
e
r
en
ce
s
p
e
ed
2
%
f
aster
th
an
YOL
O
V1
0
,
ev
en
i
n
co
m
p
lex
en
v
ir
o
n
m
en
ts
,
YOL
O
V1
1
g
u
ar
an
tees
f
ast
p
r
o
ce
s
s
in
g
b
ec
au
s
e
it
is
o
p
tim
ized
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
[
4
4
]
.
T
h
ese
d
ata
d
em
o
n
s
tr
ate
th
at
YOL
O
V1
1
r
ep
r
esen
ts
a
s
ig
n
if
ican
t
ad
v
an
ce
m
en
t
in
th
e
f
ield
o
f
ar
tific
ial
in
tellig
en
ce
an
d
p
ar
ticu
lar
ly
i
n
ar
ea
s
wh
er
e
p
r
ec
is
io
n
a
n
d
r
a
p
id
an
al
y
s
is
ar
e
r
eq
u
ir
ed
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Co
m
pa
riso
n bet
wee
n t
he
diff
er
ent
v
er
s
io
ns
o
f
YO
L
O
T
h
is
s
ec
tio
n
co
n
tain
s
th
e
e
v
o
lu
tio
n
o
f
ea
ch
v
e
r
s
io
n
o
f
Y
OL
O
in
clu
d
in
g
th
e
ar
ch
itectu
r
e
o
f
ea
ch
v
er
s
io
n
.
I
n
ad
d
itio
n
to
th
is
w
e
co
m
p
ar
e
all
t
h
e
v
er
s
io
n
s
b
y
p
r
esen
tin
g
th
e
ad
v
a
n
tag
es
an
d
d
is
ad
v
an
tag
es
o
f
ea
ch
o
n
e
.
T
o
b
e
a
b
le
to
co
m
p
ar
e
th
e
1
1
v
er
s
io
n
s
o
f
th
e
Y
OL
O
m
o
d
el,
we
ap
p
lied
it
to
th
e
C
OC
O
d
ataset,
wh
ich
is
a
wid
ely
u
s
ed
d
atase
t
an
d
in
clu
d
es
5
0
0
0
im
ag
es
o
f
ev
er
y
d
ay
o
b
jects.
Fo
r
YOL
O
v
er
s
io
n
s
t
h
at
h
a
v
e
m
u
ltip
le
v
ar
ian
ts
,
we
h
av
e
ch
o
s
en
to
test
th
e
s
m
all(
s
)
v
ar
ia
n
t
o
f
all
v
er
s
io
n
s
.
T
h
e
r
esu
lts
o
f
o
u
r
co
m
p
ar
is
o
n
ar
e
as f
o
llo
ws:
4
.
2
.
DIS
CUSSI
O
N
T
h
is
liter
atu
r
e
r
ev
iew
f
o
cu
s
es
o
n
th
e
YOL
O
alg
o
r
ith
m
;
an
ar
tific
ial
in
tellig
en
ce
alg
o
r
ith
m
d
esig
n
ed
f
o
r
o
b
ject
d
etec
tio
n
in
im
ag
es
o
r
v
id
e
o
s
.
I
t
p
r
o
v
id
es
a
co
m
p
r
eh
e
n
s
iv
e
an
d
co
m
p
ar
at
iv
e
an
aly
s
is
o
f
th
e
ev
o
lu
tio
n
o
f
th
e
YOL
O
alg
o
r
ith
m
f
r
o
m
2
0
1
6
to
2
0
2
4
.
W
h
ile
m
an
y
p
ap
er
s
f
o
cu
s
o
n
in
d
iv
id
u
al
v
er
s
io
n
s
o
f
YOL
O
o
r
in
tr
o
d
u
ce
s
p
ec
if
ic
ch
an
g
es
o
r
im
p
r
o
v
em
en
ts
,
o
u
r
wo
r
k
s
y
n
t
h
esizes
all
m
ajo
r
im
p
r
o
v
em
e
n
ts
an
d
ad
v
an
ce
m
e
n
ts
,
h
ig
h
lig
h
ts
s
tr
en
g
th
s
an
d
lim
itatio
n
s
o
f
ea
ch
v
er
s
io
n
in
o
r
d
er
to
h
el
p
f
u
tu
r
e
r
esear
ch
er
s
ch
o
o
s
in
g
th
e
b
est v
er
s
io
n
b
ased
o
n
th
e
n
ee
d
s
o
f
th
eir
s
tu
d
ies.
T
h
is
s
tu
d
y
aim
s
to
id
en
tif
y
th
e
ad
v
an
ta
g
es
an
d
d
is
ad
v
a
n
tag
es
an
d
p
r
esen
t
th
e
co
n
tr
i
b
u
tio
n
o
f
ea
ch
v
er
s
io
n
o
f
YOL
O
in
ter
m
s
o
f
ar
ch
itectu
r
e,
ac
cu
r
ac
y
,
s
p
ee
d
,
an
d
im
p
r
o
v
em
e
n
ts
in
o
r
d
e
r
to
k
n
o
w
th
e
b
est
v
er
s
io
n
to
u
s
e
f
o
r
ev
er
y
r
eq
u
ir
em
en
t.
T
h
is
r
ep
r
esen
ts
a
d
ec
is
io
n
-
m
ak
in
g
s
u
p
p
o
r
t
f
o
r
r
esear
ch
er
s
wh
o
ar
e
h
esit
atin
g
b
etwe
en
d
if
f
er
en
t
v
er
s
io
n
s
o
f
th
e
YOL
O
alg
o
r
ith
m
.
T
h
e
n
o
v
el
co
n
tr
i
b
u
tio
n
o
f
th
is
m
an
u
s
cr
ip
t
i
s
th
at
th
e
c
o
m
p
ar
is
o
n
m
ad
e
b
et
wee
n
th
e
ele
v
en
v
er
s
io
n
s
o
f
Y
OL
O
h
as
n
o
t
b
ee
n
p
r
e
v
io
u
s
ly
co
n
s
o
lid
ated
in
th
e
liter
atu
r
e
esp
ec
ially
a
co
m
p
a
r
is
o
n
th
at
in
clu
d
es th
e
latest v
er
s
io
n
s
9
,
1
0
a
n
d
1
1
.
T
h
is
liter
atu
r
e
r
ev
iew
is
b
ased
o
n
th
e
m
eth
o
d
o
lo
g
y
o
f
Kitch
en
h
am
a
n
d
C
h
ar
ter
s
[
3
]
ali
g
n
ed
with
PR
I
SMA
2
0
2
0
an
d
t
h
e
u
s
ed
te
ch
n
iq
u
es
ar
e
as
f
o
llo
ws:
1
/Th
e
f
o
r
m
u
latio
n
o
f
th
e
r
esear
ch
q
u
esti
o
n
s
was
m
ad
e
b
ased
o
n
th
e
PICO
tech
n
iq
u
e
as
s
h
o
wn
in
T
ab
le
1
,
2
/f
o
r
t
h
e
s
ea
r
ch
s
tr
ateg
y
,
k
ey
wo
r
d
s
/s
ea
r
ch
s
tr
in
g
s
u
s
e
ad
v
an
ce
d
B
o
o
lean
q
u
er
ies
o
n
s
cien
tific
d
atab
ases
,
3
/s
elec
tio
n
p
r
o
ce
s
s
is
b
ased
o
n
t
h
e
d
ef
i
n
itio
n
o
f
in
clu
s
io
n
an
d
ex
clu
s
io
n
cr
iter
ia,
4
/
f
o
r
d
o
cu
m
e
n
t
v
alid
ity
ass
ess
m
en
t
,
th
e
u
s
ed
tech
n
iq
u
e
is
cr
itical
ass
e
s
s
m
en
t
o
f
th
e
q
u
ality
o
f
r
eso
u
r
ce
s
b
ased
o
n
q
u
ality
ass
ess
m
en
t
q
u
esti
o
n
s
,
5
/
f
in
ally
,
f
o
r
d
ata
co
llectio
n
,
we
u
s
ed
s
tr
u
ctu
r
ed
ex
tr
ac
tio
n
in
T
a
b
le
s
2
,
3
a
n
d
4
.
T
h
e
o
v
er
v
iew
o
f
th
is
ar
ticle
an
s
wer
s
th
e
f
ir
s
t
q
u
esti
o
n
o
f
th
is
r
ev
iew
b
y
d
is
p
lay
in
g
th
e
ch
an
g
es m
ad
e
to
th
e
YOL
O
alg
o
r
ith
m
in
ea
c
h
o
f
its
v
er
s
io
n
s
.
T
a
b
le
2
allo
ws
u
s
to
k
n
o
w
th
at
YOL
O
V1
an
d
YOL
O
V2
u
s
e
an
ar
ch
itectu
r
e
with
o
u
t
b
ac
k
b
o
n
e,
n
ec
k
an
d
h
ea
d
wh
o
a
r
e
C
NN
an
d
Dar
k
n
et
-
1
9
r
esp
ec
ti
v
ely
.
I
n
T
ab
le
3
,
we
d
is
p
lay
th
e
v
er
s
io
n
s
o
f
YOL
O
alg
o
r
ith
m
th
at
u
s
e
an
ar
c
h
itectu
r
e
with
b
ac
k
b
o
n
e,
n
ec
k
an
d
h
ea
d
wh
ic
h
ar
e
f
r
o
m
Yo
lo
V3
to
Yo
l
o
v
1
1
an
d
th
at
allo
w
u
s
to
s
ee
th
e
d
if
f
e
r
en
t a
r
ch
itectu
r
es o
f
ea
ch
v
er
s
io
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
F
r
o
m
YOLO V
1
to
YOLO V
1
1
:
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f
YOLO
…
(
I
ma
n
e
B
eq
q
a
li Ha
s
s
a
n
i
)
457
T
ab
le
2
.
Ar
c
h
itectu
r
e
with
o
u
t
B
ac
k
b
o
n
e,
n
ec
k
an
d
h
ea
d
Y
O
LO
Y
e
a
r
A
r
c
h
i
t
e
c
t
u
r
e
V1
2
0
1
6
C
N
N
V2
2
0
1
7
D
a
r
k
n
e
t
-
19
T
ab
le
3
.
Ar
c
h
itectu
r
e
with
B
ac
k
b
o
n
e
,
n
ec
k
an
d
h
ea
d
Y
O
LO
Y
e
a
r
A
r
c
h
i
t
e
c
t
u
r
e
w
i
t
h
b
a
c
k
b
o
n
e
,
n
e
c
k
a
n
d
h
e
a
d
B
a
c
k
b
o
n
e
N
e
c
k
H
e
a
d
V3
2
0
1
8
D
a
r
k
n
e
t
-
53
F
P
N
,
P
A
N
e
t
A
n
c
h
o
r
-
b
a
se
d
a
p
p
r
o
a
c
h
e
s
V4
2
0
2
0
C
S
P
D
a
r
k
n
e
t
-
53
P
A
N
e
t
,
S
P
P
A
n
c
h
o
r
-
b
a
se
d
a
p
p
r
o
a
c
h
e
s
V5
2
0
2
0
M
o
d
i
f
i
e
d
C
S
P
D
a
r
k
N
e
t
5
3
P
A
N
e
t
A
n
c
h
o
r
-
b
a
se
d
a
p
p
r
o
a
c
h
e
s
V6
2
0
2
2
Ef
f
i
c
i
e
n
t
R
e
p
R
e
p
-
P
A
N
A
n
c
h
o
r
-
f
r
e
e
d
e
si
g
n
V7
2
0
2
3
C
S
P
D
a
r
k
N
e
t
-
5
3
e
n
h
a
n
c
e
d
w
i
t
h
t
h
e
m
o
d
u
l
e
o
f
S
P
P
(E
-
ELA
N
)
P
A
N
e
t
A
n
c
h
o
r
-
b
a
se
d
a
p
p
r
o
a
c
h
e
s
V8
2
0
2
3
E
-
ELA
N
F
o
c
u
s
-
V
3
n
e
c
k
D
e
c
o
u
p
l
e
d
h
e
a
d
s,
A
n
c
h
o
r
-
f
r
e
e
d
e
s
i
g
n
V9
2
0
2
4
G
ELA
N
R
e
p
N
C
S
P
-
ELA
N
4
*
,
A
D
o
w
n
**
A
n
c
h
o
r
-
f
r
e
e
d
e
si
g
n
V
1
0
2
0
2
4
U
p
d
a
t
e
d
v
e
r
si
o
n
o
f
C
S
P
N
e
t
P
A
N
D
u
a
l
-
h
e
a
d
d
e
si
g
n
(
O
n
e
-
to
-
M
a
n
y
H
e
a
d
a
n
d
O
n
e
-
to
-
O
n
e
H
e
a
d
)
.
V
1
1
2
0
2
4
C
3
k
2
S
P
F
F
*
*
*
A
n
c
h
o
r
-
f
r
e
e
d
e
si
g
n
*
:
R
e
p
N
C
S
P
-
ELA
N
4
:
R
e
p
a
r
a
me
t
r
i
z
e
d
n
o
n
-
l
o
c
a
l
c
r
o
s
s st
a
g
e
p
a
r
t
i
a
l
e
f
f
i
c
i
e
n
t
l
a
y
e
r
a
g
g
r
e
g
a
t
i
o
n
n
e
t
w
o
r
k
**:
A
D
o
w
n
;
A
t
t
e
n
t
i
o
n
d
o
w
n
s
a
mp
l
i
n
g
***:
SPFF:
S
p
a
t
i
a
l
p
y
r
a
m
i
d
f
i
n
e
f
u
si
o
n
C
o
n
ce
r
n
in
g
th
e
s
ec
o
n
d
q
u
esti
o
n
o
f
th
is
r
ev
iew,
T
ab
le
4
p
r
esen
ts
th
e
lis
t
o
f
a
d
v
an
tag
es
an
d
d
is
ad
v
an
tag
es
o
f
all
v
er
s
io
n
s
o
f
YOL
O
(
V1
to
V1
1
)
an
d
T
ab
le
5
g
i
v
es
u
s
a
co
m
p
ar
ati
v
e
an
aly
s
is
o
f
th
e
elev
en
v
er
s
io
n
s
o
n
C
OC
O
d
at
aset
an
d
th
is
lead
s
u
s
to
th
e
f
o
llo
win
g
r
esu
lts
.
I
n
ter
m
s
o
f
s
p
ee
d
an
d
ac
cu
r
a
cy
,
YOL
O
V2
is
b
etter
th
an
YOL
O
V1
b
u
t
b
o
th
alg
o
r
ith
m
s
ar
e
n
o
t
s
u
itab
le
f
o
r
co
m
p
lex
co
n
t
ex
ts
wh
er
e
o
b
jects
ar
e
v
er
y
clo
s
e
to
ea
ch
o
th
e
r
b
ec
au
s
e
th
ey
ca
n
’
t
d
etec
t
th
e
m
co
r
r
ec
tly
.
C
o
m
p
a
r
in
g
to
Y
OL
O
V1
an
d
V2
,
YOL
O
V3
im
p
r
o
v
es
th
e
s
m
all
-
s
ize
tar
g
et
d
etec
tio
n
ac
cu
r
ac
y
an
d
p
r
ed
ictio
n
.
As
f
o
r
YOL
O
V4
,
we
ca
n
s
ay
th
at
it is
b
etter
th
an
YOL
O
V3
in
ter
m
s
o
f
p
r
ec
is
io
n
an
d
s
p
ee
d
b
u
t a
cc
o
r
d
in
g
to
[
4
8
]
it m
ay
b
e
s
lo
wer
in
s
o
m
e
s
ce
n
ar
io
s
.
C
o
n
ce
r
n
in
g
YOL
O
V5
,
it h
as 3
7
.
5
% m
AP f
o
r
th
e
s
m
all
v
ar
ian
t a
n
d
it a
llo
ws id
e
n
tify
in
g
v
er
y
s
m
all
o
b
jects f
aster
th
an
YOL
O
V4
b
u
t it
en
co
u
n
ter
s
s
o
m
e
d
if
f
icu
l
ties
wh
en
th
er
e
is
an
o
v
er
lap
o
f
o
b
jects,
a
ch
an
g
e
o
f
lig
h
tin
g
an
d
o
th
er
c
o
m
p
lex
co
n
d
itio
n
s
.
YOL
O
V6
ac
h
iev
es
h
ig
h
ac
c
u
r
ac
y
,
p
er
f
o
r
m
a
n
ce
an
d
s
p
ee
d
co
m
p
a
r
ed
to
p
r
ev
io
u
s
v
er
s
io
n
s
b
u
t
h
as
n
o
t
b
ee
n
m
u
ch
u
s
ed
b
ec
au
s
e
it
is
n
o
t
f
u
lly
o
p
en
-
s
o
u
r
ce
.
A
s
f
o
r
YOL
O
V7
,
we
ca
n
s
ay
th
at
it
p
r
o
v
es
to
b
e
ef
f
ec
tiv
e
in
ter
m
s
o
f
ac
c
u
r
ac
y
co
m
p
a
r
ed
to
p
r
ev
io
u
s
v
er
s
io
n
s
an
d
in
s
o
m
e
s
ce
n
a
r
io
s
,
it
ev
en
o
u
t
p
er
f
o
r
m
s
YOL
O
V8
b
u
t
it
s
t
ill
en
co
u
n
ter
s
p
r
o
b
lem
s
in
ex
tr
em
e
c
o
n
d
itio
n
s
s
u
ch
as
lo
w
lig
h
tin
g
,
o
r
v
er
y
clu
tter
ed
b
ac
k
g
r
o
u
n
d
s
.
YOL
O
V8
i
s
u
s
er
-
f
r
ien
d
ly
alg
o
r
ith
m
ch
a
r
ac
ter
ized
b
y
b
etter
s
p
ee
d
an
d
ac
c
u
r
ac
y
co
m
p
a
r
ed
to
p
r
ev
io
u
s
alg
o
r
ith
m
s
ex
ce
p
t Y
OL
O
V7
wh
ich
lar
g
ely
s
u
r
p
as
s
es it in
ter
m
s
o
f
ac
cu
r
ac
y
in
ce
r
tain
co
n
tex
ts
.
Ou
r
p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
ch
a
r
ts
as
s
h
o
wn
in
T
ab
le
5
s
h
o
ws
th
at
YOL
O
V9
s
allo
ws
a
s
ig
n
if
ican
t
r
ed
u
ctio
n
in
th
e
n
u
m
b
er
o
f
p
ar
am
ete
r
s
u
s
e
d
in
t
h
e
alg
o
r
ith
m
an
d
co
m
p
ar
ed
to
YOL
O
V8
s
,
it
h
as
g
o
o
d
ac
cu
r
ac
y
4
7
%m
AP
b
u
t
its
s
p
ee
d
r
em
ain
s
r
elativ
e
ly
lo
w.
I
n
a
d
d
itio
n
to
t
h
at,
s
o
m
e
au
th
o
r
s
claim
th
at
YOL
O
V9
ac
h
iev
es
g
o
o
d
p
er
f
o
r
m
an
ce
an
d
im
p
r
o
v
es d
et
ec
tio
n
ac
cu
r
ac
y
b
u
t o
th
er
r
esear
ch
er
s
f
in
d
th
at
YOL
O
V9
d
o
es n
o
t d
em
o
n
s
tr
ate
o
u
ts
tan
d
in
g
p
er
f
o
r
m
an
ce
in
t
er
m
s
o
f
s
p
ee
d
an
d
ac
c
u
r
ac
y
an
d
s
till
en
co
u
n
te
r
s
p
r
o
b
lem
s
in
d
etec
tin
g
lo
w
q
u
ality
im
ag
es.
C
o
n
ce
r
n
in
g
Y
OL
O
V1
0
s
,
it
h
as
also
a
r
e
d
u
ce
d
n
u
m
b
er
o
f
p
ar
a
m
eter
s
an
d
h
as
a
g
o
o
d
m
AP
(
4
6
,
2
%)
a
n
d
s
p
ee
d
(
1
3
7
FP
S)
b
u
t
ac
co
r
d
in
g
t
o
[
3
4
]
its
ac
cu
r
ac
y
q
u
ality
r
em
ain
s
in
f
er
io
r
to
YOL
O
V8
f
o
r
s
m
all
o
b
jects
an
d
in
s
o
m
e
c
o
n
tex
ts
it
r
em
ain
s
ev
en
in
f
er
io
r
to
YOL
O
V5
.
Fin
ally
,
YOL
O
V1
1
o
f
f
er
s
in
cr
ea
s
ed
ef
f
icien
cy
an
d
s
p
ee
d
co
m
p
a
r
ed
to
p
r
ev
io
u
s
v
er
s
io
n
s
b
u
t
s
in
ce
it
is
a
v
er
y
r
ec
e
n
t
v
er
s
io
n
,
it
h
as
n
o
t
b
ee
n
test
ed
b
y
s
ev
e
r
al
r
esear
c
h
er
s
an
d
in
s
ev
e
r
al
co
n
te
x
ts
.
T
o
an
s
wer
t
h
e
th
ir
d
an
d
last
q
u
esti
o
n
o
f
th
is
r
e
v
iew,
th
ese
r
esu
lts
s
h
o
w
th
at
t
h
er
e
is
n
o
t
a
p
er
f
ec
t
alg
o
r
ith
m
.
C
h
o
o
s
in
g
th
e
a
p
p
r
o
p
r
iate
v
er
s
io
n
d
ep
e
n
d
s
o
n
th
e
s
tu
d
y
co
n
tex
t,
th
e
in
f
o
r
m
atio
n
u
s
ed
f
o
r
th
e
s
tu
d
y
,
th
e
co
m
p
lex
ity
o
f
th
e
o
b
jects
to
b
e
d
etec
ted
an
d
p
r
io
r
ities
o
f
r
esear
ch
er
s
.
Stu
d
ies
wh
o
s
e
o
b
jectiv
e
is
th
e
d
etec
tio
n
o
f
s
m
all
o
b
jects
s
h
o
u
ld
u
s
e
YOL
O
V4
,
YOL
O
V
5
,
YOL
O
V6
,
YOL
O
V7
,
YO
L
O
V8
,
YOL
O
V9
,
YOL
O
V1
0
o
r
YOL
O
V1
1
.
I
f
th
e
p
r
io
r
ity
o
f
th
e
s
tu
d
y
is
th
e
d
etec
tio
n
o
f
s
m
all
o
b
jects
in
en
v
ir
o
n
m
en
ts
th
at
ar
e
co
m
p
lex
,
h
av
e
lo
w
lig
h
ti
n
g
o
r
h
av
e
v
er
y
clu
tter
ed
b
ac
k
g
r
o
u
n
d
s
,
it
is
n
ec
ess
ar
y
to
av
o
id
wo
r
k
in
g
with
YOL
O
V1
,
YOL
O
V2
,
YOL
O
V3
,
YOL
O
V5
,
YOL
O
V
7
an
d
YOL
O
V9
.
Fo
r
s
tu
d
ies
wh
er
e
p
r
o
ce
s
s
in
g
s
p
ee
d
is
a
p
r
io
r
ity
,
t
h
ey
s
h
o
u
ld
u
s
e
YOL
O
V5
,
YOL
O
V6
,
YOL
O
V7
,
YOL
O
V8
,
YOL
O
V1
0
o
r
YOL
O
V1
1
an
d
av
o
id
wo
r
k
in
g
with
YOL
O
V4
o
r
YOL
O
V9
.
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
.
1
,
Feb
r
u
ar
y
20
2
6
:
4
5
0
-
462
458
T
ab
le
4
.
Ad
v
an
tag
es a
n
d
d
is
ad
v
an
tag
es o
f
th
e
d
if
f
er
en
t
v
er
s
io
n
s
o
f
YOL
O
alg
o
r
ith
m
V
e
r
si
o
n
o
f
Y
O
LO
A
d
v
a
n
t
a
g
e
s
D
i
sad
v
a
n
t
a
g
e
s
Y
O
LO
V
1
1
/
F
o
r
c
a
n
d
i
d
a
t
e
b
o
x
p
r
e
d
i
c
t
i
o
n
a
n
d
o
b
j
e
c
t
c
l
a
ssi
f
i
c
a
t
i
o
n
,
Y
O
LO
V
1
d
i
r
e
c
t
l
y
a
d
o
p
t
s
r
e
g
r
e
ssi
o
n
.
T
h
i
s
i
m
p
r
o
v
e
me
n
t
mak
e
s
Y
O
LO
V
1
t
e
n
t
i
mes
f
a
st
e
r
i
n
t
e
r
ms
o
f
d
e
t
e
c
t
i
o
n
t
h
a
n
f
a
st
e
r
R
-
C
N
N
(
r
e
g
i
o
n
s
w
i
t
h
C
N
N
)
a
n
d
si
mp
l
i
f
i
e
s
t
h
e
n
e
t
w
o
r
k
st
r
u
c
t
u
r
e
[
4
6
]
1
/
I
n
s
c
e
n
a
r
i
o
s
w
h
e
r
e
o
b
j
e
c
t
s
a
r
e
c
l
o
s
e
t
o
e
a
c
h
o
t
h
e
r
,
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
Y
O
L
O
V
1
i
s
n
o
t
o
p
t
i
mal
[
9
]
Y
O
LO
V
2
1
/
F
a
s
t
e
r
i
n
t
e
r
ms
o
f
d
e
t
e
c
t
i
o
n
s
p
e
e
d
t
h
a
n
o
t
h
e
r
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
ms
a
t
t
h
a
t
t
i
me
,
i
n
c
l
u
d
i
n
g
Y
O
LO
V
1
(
[
1
3
]
,
[
4
6
]
)
2
/
F
u
r
t
h
e
r
m
o
r
e
,
i
t
c
a
n
b
e
r
u
n
a
t
a
v
a
r
i
e
t
y
o
f
i
m
a
g
e
si
z
e
s
t
o
p
r
o
v
i
d
e
a
smo
o
t
h
t
r
a
d
e
o
f
f
b
e
t
w
e
e
n
a
c
c
u
r
a
c
y
a
n
d
sp
e
e
d
[
1
3
]
1
/
Y
O
LO
V
2
c
a
n
n
o
t
e
f
f
e
c
t
i
v
e
l
y
d
e
t
e
c
t
sma
l
l
o
b
j
e
c
t
s
a
n
d
v
e
r
y
c
l
o
se
o
b
j
e
c
t
s
i
n
c
o
mp
l
e
x
i
ma
g
e
s
[
1
3
]
Y
O
LO
V
3
1
/
M
o
r
e
e
f
f
e
c
t
i
v
e
t
h
a
n
Y
O
LO
V
2
[
4
6
]
2
/
H
a
s
a
l
o
w
b
a
c
k
g
r
o
u
n
d
f
a
l
se
d
e
t
e
c
t
i
o
n
r
a
t
e
a
n
d
v
e
r
y
f
a
st
d
e
t
e
c
t
i
o
n
sp
e
e
d
,
a
n
d
i
m
p
r
o
v
e
s
t
h
e
d
e
t
e
c
t
i
o
n
a
c
c
u
r
a
c
y
o
f
sma
l
l
o
b
j
e
c
t
s
[
4
6
]
3
/
Th
e
m
o
st
si
g
n
i
f
i
c
a
n
t
i
m
p
r
o
v
e
me
n
t
i
n
Y
O
LO
V
3
l
i
e
s i
n
mu
l
t
i
-
sc
a
l
e
p
r
e
d
i
c
t
i
o
n
s
[
8
]
1
/
P
o
o
r
i
n
t
e
r
ms
o
f
i
t
s
p
r
e
d
i
c
t
i
o
n
a
c
c
u
r
a
c
y
o
f
t
a
r
g
e
t
c
o
o
r
d
i
n
a
t
e
s
[
4
6
]
Y
O
LO
V
4
1
/
C
o
m
p
a
r
e
d
t
o
Y
O
LO
V
3
,
Y
O
LO
V
4
a
c
h
i
e
v
e
s
a
h
i
g
h
e
r
f
r
a
mes
p
e
r
sec
o
n
d
(
F
P
S
)
a
n
d
a
v
e
r
a
g
e
p
r
e
c
i
si
o
n
(
A
P
)
r
a
t
e
[
8
]
1
/
Y
O
LO
V
4
ma
y
b
e
s
l
o
w
e
r
[
4
8
]
Y
O
LO
V
5
1
/
Y
O
LO
V
5
e
n
a
b
l
e
s
t
h
e
mo
d
e
l
t
o
l
e
a
r
n
h
o
w
t
o
i
d
e
n
t
i
f
y
a
v
a
r
i
e
t
y
o
f
o
b
j
e
c
t
s
a
t
a
mu
c
h
sma
l
l
e
r
s
c
a
l
e
t
h
a
n
n
o
r
ma
l
[
8
]
1
/
Y
O
LO
V
5
m
a
y
n
o
t
b
e
r
o
b
u
s
t
e
n
o
u
g
h
u
n
d
e
r
c
o
m
p
l
e
x
c
o
n
d
i
t
i
o
n
s,
su
c
h
a
s
v
a
r
y
i
n
g
d
i
s
t
a
n
c
e
s,
o
b
j
e
c
t
o
v
e
r
l
a
p
,
o
c
c
l
u
s
i
o
n
s,
a
n
d
l
i
g
h
t
i
n
g
c
h
a
n
g
e
s
[
4
9
]
Y
O
LO
V
6
1
/
Y
O
LO
V
6
a
c
h
i
e
v
e
s h
i
g
h
e
r
a
v
e
r
a
g
e
p
r
e
c
i
si
o
n
(
mA
P
)
o
n
st
a
n
d
a
r
d
d
a
t
a
s
e
t
s
l
i
k
e
C
O
C
O
[
2
9
]
2
/
I
t
c
o
m
b
i
n
e
s
a
d
v
a
n
c
e
d
t
e
c
h
n
i
q
u
e
s
l
i
k
e
R
e
p
O
p
t
i
mi
z
e
r
a
n
d
s
t
r
u
c
t
u
r
a
l
o
p
t
i
mi
z
a
t
i
o
n
s
t
o
i
m
p
r
o
v
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
[
2
8
]
.
1
/
C
o
m
p
a
r
e
d
t
o
p
r
e
v
i
o
u
s
v
e
r
s
i
o
n
s
l
i
k
e
Y
O
LO
V
4
o
r
Y
O
LO
V
5
,
Y
O
LO
V
6
h
a
s
b
e
e
n
l
e
ss
u
se
d
b
y
t
h
e
r
e
se
a
r
c
h
c
o
mm
u
n
i
t
y
2
/
Y
O
LO
V
6
i
s n
o
t
f
u
l
l
y
o
p
e
n
-
so
u
r
c
e
i
n
s
o
me
c
a
ses
,
w
h
i
c
h
ma
y
b
e
p
r
o
b
l
e
ma
t
i
c
f
o
r
smal
l
c
o
m
p
a
n
i
e
s
o
r
i
n
d
e
p
e
n
d
e
n
t
r
e
se
a
r
c
h
e
r
s
.
Y
O
LO
V
7
1
/
T
h
r
o
u
g
h
se
v
e
r
a
l
me
a
n
s,
Y
O
LO
V
7
i
mp
r
o
v
e
s a
c
c
u
r
a
c
y
,
n
o
t
a
b
l
y
t
h
r
o
u
g
h
m
o
d
e
l
sc
a
l
i
n
g
a
n
d
p
a
r
a
me
t
e
r
t
u
n
i
n
g
a
n
d
a
l
s
o
t
h
r
o
u
g
h
t
h
e
u
s
e
o
f
E
-
E
LA
N
[
8
]
2
/
R
e
c
e
n
t
w
o
r
k
h
a
s
s
h
o
w
n
t
h
a
t
Y
O
LO
V
5
,
Y
O
LO
V
6
,
a
n
d
e
v
e
n
Y
O
LO
V
8
,
i
n
s
o
me
sc
e
n
a
r
i
o
s,
a
r
e
o
u
t
p
e
r
f
o
r
me
d
b
y
Y
O
LO
V
7
(
[
8
]
,
[
5
0
]
)
3
/
I
n
t
e
r
ms
o
f
r
e
s
o
u
r
c
e
e
f
f
i
c
i
e
n
c
y
a
n
d
f
a
mi
l
i
a
r
i
t
y
,
Y
O
LO
V
7
p
r
o
v
e
s
t
o
b
e
q
u
i
t
e
e
f
f
e
c
t
i
v
e
[
9
]
1
/
A
l
t
h
o
u
g
h
t
h
e
Y
O
LO
V
7
i
s
d
e
si
g
n
e
d
t
o
p
e
r
f
o
r
m
w
e
l
l
i
n
c
o
mp
l
e
x
f
i
e
l
d
e
n
v
i
r
o
n
me
n
t
s
,
i
t
ma
y
s
t
i
l
l
e
n
c
o
u
n
t
e
r
e
x
t
r
e
m
e
c
o
n
d
i
t
i
o
n
s
su
c
h
a
s
p
o
o
r
l
i
g
h
t
i
n
g
,
h
e
a
v
y
o
c
c
l
u
s
i
o
n
,
o
r
v
e
r
y
c
l
u
t
t
e
r
e
d
b
a
c
k
g
r
o
u
n
d
s
[
5
1
]
Y
O
LO
V
8
1
/
T
h
e
a
d
v
a
n
t
a
g
e
s
o
f
Y
O
LO
V
8
a
r
e
sca
l
a
b
i
l
i
t
y
,
h
i
g
h
p
e
r
f
o
r
m
a
n
c
e
,
a
n
d
u
ser
-
f
r
i
e
n
d
l
i
n
e
s
s
[
3
4
]
2
/
Y
O
LO
V
8
o
f
f
e
r
s
s
u
p
e
r
i
o
r
a
c
c
u
r
a
c
y
a
n
d
s
p
e
e
d
c
o
m
p
a
r
e
d
t
o
p
r
e
v
i
o
u
s
Y
O
LO
v
e
r
si
o
n
s
[
4
2
]
3
/
Y
O
LO
V
8
i
s
e
n
t
h
u
si
a
st
i
c
a
b
o
u
t
i
t
s
c
u
t
t
i
n
g
-
e
d
g
e
d
e
s
i
g
n
,
w
h
i
c
h
ma
x
i
m
i
z
e
s
u
n
i
q
u
e
n
e
ss a
n
d
f
u
n
c
t
i
o
n
a
l
i
t
y
[
9
]
1
/
Y
O
LO
V
8
i
s
o
u
t
p
e
r
f
o
r
me
d
b
y
Y
O
LO
V
7
[
5
0
]
Y
O
LO
V
9
1
/
Y
O
LO
V
9
a
c
h
i
e
v
e
d
g
o
o
d
p
e
r
f
o
r
ma
n
c
e
[
5
2
]
a
n
d
i
n
t
r
o
d
u
c
e
s
n
e
w
a
r
c
h
i
t
e
c
t
u
r
e
s
t
h
a
t
i
mp
r
o
v
e
a
c
c
u
r
a
c
y
(
e
.
g
.
,
g
e
n
e
r
a
l
i
z
e
d
h
i
g
h
-
e
f
f
i
c
i
e
n
c
y
l
a
y
e
r
a
g
g
r
e
g
a
t
i
o
n
n
e
t
w
o
r
k
s
a
n
d
p
r
o
g
r
a
mm
a
b
l
e
g
r
a
d
i
e
n
t
i
n
f
o
r
ma
t
i
o
n
)
[
4
2
]
2
/
Y
O
LO
V
9
a
l
l
o
w
s
l
e
a
r
n
i
n
g
e
f
f
e
c
t
i
v
e
l
y
a
n
d
i
d
e
n
t
i
f
y
i
n
g
d
i
f
f
e
r
e
n
t
t
y
p
e
s
o
f
o
b
j
e
c
t
s
i
n
r
e
a
l
t
i
m
e
a
n
d
p
r
o
c
e
ss
l
a
r
g
e
-
sca
l
e
d
a
t
a
s
e
t
s
[
4
0
]
1
/
C
o
m
p
a
r
e
d
t
o
Y
O
LO
V
8
,
Y
O
LO
V
9
f
a
i
l
s
t
o
f
i
n
d
t
h
e
r
i
g
h
t
b
a
l
a
n
c
e
b
e
t
w
e
e
n
a
c
c
u
r
a
c
y
a
n
d
s
p
e
e
d
[
4
2
]
2
/
D
e
t
e
c
t
i
n
g
l
o
w
-
q
u
a
l
i
t
y
i
m
a
g
e
t
a
r
g
e
t
s
b
y
Y
O
LO
V
9
s
t
i
l
l
p
o
ses
c
h
a
l
l
e
n
g
e
s
[
5
2
]
3
/
Y
O
LO
V
9
'
s
p
r
o
c
e
ssi
n
g
s
p
e
e
d
a
n
d
a
c
c
u
r
a
c
y
s
t
i
l
l
n
e
e
d
s
i
m
p
r
o
v
e
me
n
t
(
[
5
3
]
,
[
3
9
]
)
Y
O
LO
V
1
0
1
/
B
y
o
p
t
i
m
i
z
i
n
g
m
o
d
e
l
c
o
m
p
o
n
e
n
t
s a
n
d
r
e
m
o
v
i
n
g
N
M
S
,
Y
O
LO
V
1
0
e
x
c
e
l
s
i
n
l
i
g
h
t
w
e
i
g
h
t
,
a
c
c
u
r
a
c
y
,
a
n
d
sp
e
e
d
[
3
4
]
1
/
I
n
d
e
t
e
c
t
i
n
g
s
mal
l
t
a
r
g
e
t
s
,
Y
O
LO
V
1
0
i
s
sl
i
g
h
t
l
y
i
n
f
e
r
i
o
r
t
o
Y
O
LO
V
8
[
3
4
]
2
/
A
c
c
o
r
d
i
n
g
t
o
G
e
e
t
h
a
a
n
d
H
u
ssa
i
n
[
5
4
]
,
i
n
c
e
r
t
a
i
n
s
c
e
n
a
r
i
o
s,
Y
O
LO
V
5
o
u
t
p
e
r
f
o
r
med
Y
O
LO
V
1
0
a
n
d
Y
O
LO
V
8
i
n
t
e
r
ms
o
f
a
c
c
u
r
a
c
y
Y
O
LO
V
1
1
1
/
Y
O
LO
V
1
1
a
l
l
o
w
s
a
s
i
g
n
i
f
i
c
a
n
t
r
e
d
u
c
t
i
o
n
i
n
t
h
e
n
u
m
b
e
r
o
f
p
a
r
a
me
t
e
r
s
u
se
d
i
n
t
h
e
a
l
g
o
r
i
t
h
m
[
4
7
]
2
/
Y
O
LO
V
1
1
o
f
f
e
r
s
i
n
c
r
e
a
s
e
d
e
f
f
i
c
i
e
n
c
y
a
n
d
sp
e
e
d
c
o
m
p
a
r
e
d
t
o
p
r
e
v
i
o
u
s
v
e
r
si
o
n
s
[
4
7
]
1
/
S
i
n
c
e
i
t
’
s
a
v
e
r
y
r
e
c
e
n
t
v
e
r
si
o
n
;
Y
O
LO
V
1
1
h
a
s
n
o
t
b
e
e
n
t
e
st
e
d
a
n
d
o
p
t
i
m
i
z
e
d
o
v
e
r
a
l
o
n
g
p
e
r
i
o
d
.
T
h
e
s
tu
d
y
'
s
f
in
d
in
g
s
an
d
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
elev
en
v
er
s
io
n
s
o
f
th
e
YOL
O
alg
o
r
ith
m
h
av
e
u
n
d
er
g
o
n
e
co
n
tin
u
o
u
s
im
p
r
o
v
em
en
t
o
v
er
th
e
y
ea
r
s
in
ter
m
s
o
f
ac
cu
r
ac
y
,
a
r
ch
itectu
r
al
co
m
p
lex
ity
,
an
d
o
b
ject
d
etec
tio
n
s
p
ee
d
.
I
n
d
ee
d
,
th
e
f
ir
s
t
th
r
ee
v
er
s
io
n
s
estab
lis
h
e
d
th
e
b
asics
o
f
d
etec
tio
n
,
b
u
t
th
ey
ar
e
n
o
t
v
er
y
ac
cu
r
ate
an
d
s
tr
u
g
g
le
to
d
ete
ct
o
b
jects
in
c
o
m
p
lex
en
v
ir
o
n
m
en
ts
.
Ver
s
io
n
s
4
to
1
1
all
o
w
th
e
d
etec
tio
n
o
f
s
m
all
o
b
jects
b
u
t
th
ey
d
o
n
o
t
all
p
er
f
o
r
m
well
in
co
m
p
lex
e
n
v
ir
o
n
m
en
ts
.
I
n
d
ee
d
,
YOL
O
V5
,
YOL
O
V7
an
d
YOL
O
V9
d
o
n
o
t
p
er
f
o
r
m
wel
l in
th
e
en
v
ir
o
n
m
e
n
ts
th
at
ar
e
n
o
t v
er
y
clea
r
o
r
lack
b
r
ig
h
tn
e
s
s
.
T
h
e
m
o
s
t r
ec
en
t
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
F
r
o
m
YOLO V
1
to
YOLO V
1
1
:
C
o
mp
a
r
a
tive
a
n
a
lysi
s
o
f
YOLO
…
(
I
ma
n
e
B
eq
q
a
li Ha
s
s
a
n
i
)
459
v
er
s
io
n
s
V1
0
an
d
V1
1
o
f
f
e
r
a
f
ai
r
ly
b
alan
ce
d
co
m
p
r
o
m
is
e
b
etwe
en
s
p
ee
d
,
p
r
ec
is
i
o
n
an
d
n
u
m
b
e
r
o
f
p
ar
am
eter
s
an
d
th
ey
a
r
e
esp
ec
ially
v
er
y
e
f
f
icien
t
in
ter
m
s
o
f
s
p
ee
d
.
T
h
is
is
d
u
e
to
i
n
n
o
v
atio
n
s
s
u
ch
as
d
ec
o
u
p
le
d
h
ea
d
s
,
a
n
ch
o
r
-
f
r
ee
d
etec
tio
n
an
d
tr
an
s
f
o
r
m
er
-
b
as
ed
m
o
d
u
les.
T
h
e
co
m
p
ar
is
o
n
b
etwe
en
d
if
f
er
en
t
v
er
s
io
n
s
o
f
YOL
O
s
h
o
ws
th
at
r
ec
en
t
v
er
s
io
n
s
ar
e
n
o
t
alwa
y
s
th
e
b
est
in
all
co
n
tex
ts
.
Fo
r
ex
am
p
le,
YOL
O
V4
m
ay
d
em
o
n
s
tr
ate
ex
ce
llen
t
p
er
f
o
r
m
an
c
e
f
o
r
s
m
all
o
b
ject
d
etec
tio
n
,
wh
ile
YOL
O
V8
p
r
io
r
itizes
s
p
ee
d
an
d
s
ac
r
if
ices
s
o
m
e
ac
cu
r
ac
y
.
T
h
is
s
h
o
ws
th
at
th
e
b
est
v
e
r
s
io
n
o
f
th
e
YOL
O
alg
o
r
ith
m
d
ep
e
n
d
s
o
n
th
e
c
o
n
tex
t
a
n
d
th
at
r
esear
ch
er
s
s
h
o
u
ld
n
o
t
alwa
y
s
ch
o
o
s
e
th
e
latest
v
er
s
io
n
,
as
it
m
ay
n
o
t
y
ield
th
e
d
esire
d
r
esu
lts
in
all
c
o
n
tex
t
s
.
T
h
e
ar
c
h
itectu
r
es
ad
o
p
ted
f
o
r
ea
ch
v
e
r
s
io
n
ar
e
th
e
r
esu
lt o
f
g
o
o
d
b
alan
ce
th
at
d
ev
elo
p
e
r
s
co
n
s
id
er
b
ased
o
n
th
e
co
n
s
tr
ain
ts
im
p
o
s
ed
b
y
th
e
s
tu
d
y
.
T
ab
le
5
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
ch
ar
ts
Y
O
LO
mA
P
5
0
-
95*
F
P
S
*
*
P
a
r
a
me
t
e
r
s (
M
)
*
*
*
Y
O
LO
V
1
1
9
.
6
%
45
6
2
M
Y
O
LO
V
2
2
1
.
6
%
67
5
0
M
Y
O
LO
V
3
3
1
.
7
%
30
6
1
.
9
M
Y
O
LO
V
4
4
4
,
5
%
62
6
4
M
Y
O
LO
V
5
s
3
7
.
5
%
1
3
2
7
.
2
M
Y
O
LO
V
6
s
4
4
.
8
%
4
8
4
1
3
.
1
4
M
Y
O
LO
V
7
5
1
.
4
%
1
3
5
3
7
.
2
M
Y
O
LO
V
8
s
4
4
.
9
%
1
3
3
1
1
.
1
6
M
Y
O
LO
V
9
s
4
7
%
1
2
0
7
.
2
0
M
Y
O
LO
V
1
0
s
4
6
.
2
%
1
3
7
7
.
2
5
M
Y
O
LO
V
1
1
s
4
6
.
8
%
1
4
9
9
.
4
4
M
*
:
mAP
:
me
a
n
A
v
e
r
a
g
e
P
r
e
c
i
s
i
o
n
d
e
t
e
r
mi
n
e
d
f
o
r
I
o
U
l
e
v
e
l
s
b
e
t
w
e
e
n
0
.
5
0
a
n
d
0
.
9
5
i
n
s
t
e
p
s
o
f
0
.
0
5
,
t
h
e
n
a
v
e
r
a
g
e
d
o
v
e
r
a
l
l
c
l
a
ss
e
s
**
:
FPS
:
F
r
a
mes
P
e
r
S
e
c
o
n
d
*
*
*
:
Pa
r
a
m
e
t
e
r
s
(
M
)
:
N
u
m
b
e
r
o
f
p
a
r
a
met
e
r
s
o
f
t
h
e
m
o
d
e
l
5.
CO
NCLU
SI
O
N
B
ased
o
n
o
u
r
r
esu
lts
an
d
d
is
cu
s
s
io
n
,
th
is
li
ter
atu
r
e
r
ev
iew
o
f
f
er
s
a
r
o
ad
m
ap
f
o
r
r
esear
c
h
er
s
to
ch
o
o
s
e
th
e
YOL
O
v
er
s
io
n
b
est s
u
ited
to
th
eir
p
r
io
r
ities
,
wh
eth
er
in
t
er
m
s
o
f
ac
cu
r
ac
y
,
p
e
r
f
o
r
m
an
c
e,
o
r
r
esp
o
n
s
e
tim
e.
T
h
e
im
p
licatio
n
s
o
f
th
is
wo
r
k
in
clu
d
e
r
a
p
id
al
g
o
r
ith
m
s
elec
tio
n
b
ased
o
n
s
p
ec
if
ic
co
n
s
tr
a
in
ts
o
f
a
s
tu
d
y
an
d
h
elp
in
g
r
esear
ch
er
s
a
v
o
id
in
g
er
r
o
r
s
a
n
d
tim
e
-
co
n
s
u
m
in
g
w
h
en
c
h
o
o
s
in
g
a
m
o
d
el.
I
n
ad
d
itio
n
to
th
at,
th
is
r
ev
iew
will
allo
w
r
esear
ch
er
s
h
av
in
g
ef
f
icien
t
an
d
im
p
r
o
v
e
d
p
er
f
o
r
m
a
n
ce
in
th
e
u
s
e
o
f
a
r
t
if
icial
in
tellig
en
ce
.
As
an
o
u
tlo
o
k
f
o
r
th
is
wo
r
k
,
we
r
ec
o
m
m
en
d
t
h
at
f
u
tu
r
e
r
e
s
ea
r
ch
f
o
cu
s
o
n
e
x
p
lo
r
in
g
m
o
d
els
th
at
ac
h
iev
e
a
g
o
o
d
b
alan
ce
b
etwe
en
ac
c
u
r
a
cy
,
s
p
ee
d
an
d
n
u
m
b
er
o
f
p
ar
a
m
eter
s
to
im
p
r
o
v
e
d
etec
tio
n
e
f
f
icien
cy
,
esp
ec
ially
f
o
r
u
s
es wh
er
e
r
ea
l
-
tim
e
d
etec
tio
n
is
n
ec
ess
ar
y
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
No
f
u
n
d
in
g
in
v
o
lv
e
d
.
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
I
m
an
e
B
eq
q
ali
Hass
an
i
✓
✓
✓
✓
✓
✓
So
u
f
ia
B
en
h
id
a
✓
Nab
il L
am
ii
✓
Kh
alid
Oq
aid
i
✓
✓
Ah
m
ed
Ou
id
d
a
d
✓
So
u
k
ain
a
Gh
iad
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
Evaluation Warning : The document was created with Spire.PDF for Python.