I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
4
,
A
u
g
u
s
t
20
26
,
p
p
.
3
5
8
1
~
3
5
9
0
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
5
8
1
-
3
5
9
0
3581
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Rea
l
-
time
objec
t
detec
tion for a
u
tono
mo
us driv
ing
:
a
co
mpa
ra
tive stud
y
of YO
L
O
and
F
a
ster R
-
CNN
M
a
dh
ura
M
.
B
ho
s
a
le
1
,
Yo
g
esh
S.
Ang
a
l
2
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
JS
P
M
's
R
a
j
a
r
s
h
i
S
h
a
h
u
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
P
u
n
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
JS
P
M
's
B
h
i
v
a
r
a
b
a
i
S
a
w
a
n
t
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
R
e
s
e
a
r
c
h
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
30
,
2
0
2
5
R
ev
is
ed
May
13
,
2
0
2
6
Acc
ep
ted
May
24
,
2
0
2
6
Ov
e
r
th
e
p
a
st
fe
w
y
e
a
rs,
o
b
jec
t
d
e
tec
ti
o
n
h
a
s
e
x
p
e
rien
c
e
d
r
e
m
a
rk
a
b
le
p
ro
g
re
ss
a
n
d
d
e
v
e
lo
p
m
e
n
t,
p
rima
ril
y
d
ri
v
e
n
b
y
th
e
d
e
v
e
l
o
p
m
e
n
t
o
f
o
n
e
-
sta
g
e
a
n
d
tw
o
-
sta
g
e
d
e
tec
ti
o
n
a
l
g
o
rit
h
m
s.
Am
o
n
g
t
h
e
se
,
F
a
ste
r
re
g
io
n
-
b
a
se
d
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
(
F
a
ste
r
R
-
CNN
)
a
n
d
y
o
u
o
n
l
y
lo
o
k
o
n
c
e
(YO
LO)
h
a
v
e
a
c
h
iev
e
d
n
o
tab
le
su
c
c
e
ss
d
u
e
t
o
t
h
e
ir
stro
n
g
p
e
rf
o
r
m
a
n
c
e
a
n
d
c
o
m
p
u
tati
o
n
a
l
e
fficie
n
c
y
.
Ob
jec
t
d
e
tec
ti
o
n
p
lay
s
a
c
ru
c
ial
r
o
le
i
n
v
a
ri
o
u
s
a
p
p
li
c
a
ti
o
n
s,
p
a
rti
c
u
larly
i
n
a
u
t
o
n
o
m
o
u
s
d
riv
i
n
g
s
y
ste
m
s,
wh
e
re
a
c
c
u
ra
te
d
e
tec
ti
o
n
o
f
p
e
d
e
strian
s,
v
e
h
icle
s,
a
n
d
r
o
a
d
sig
n
s
is
e
ss
e
n
ti
a
l
fo
r
e
n
su
rin
g
sa
fe
ty
a
n
d
re
li
a
b
il
i
ty
.
T
h
is
p
a
p
e
r
c
o
n
d
u
c
ts
a
c
o
m
p
a
ra
ti
v
e
e
v
a
l
u
a
ti
o
n
o
f
YO
LO
a
n
d
F
a
ste
r
R
-
CNN
t
o
a
n
a
ly
z
e
t
h
e
ir
p
e
rfo
rm
a
n
c
e
in
a
u
t
o
n
o
m
o
u
s
d
riv
i
n
g
e
n
v
i
ro
n
m
e
n
ts.
T
h
e
e
x
p
e
rime
n
ts
we
re
c
o
n
d
u
c
ted
u
si
n
g
t
h
e
KITT
I
o
p
e
n
-
s
o
u
rc
e
d
a
tas
e
t,
wh
ic
h
is
wi
d
e
ly
u
se
d
f
o
r
b
e
n
c
h
m
a
rk
in
g
o
b
jec
t
d
e
tec
ti
o
n
m
o
d
e
ls.
All
e
x
p
e
rime
n
ts
we
re
p
e
rfo
rm
e
d
o
n
a
n
NV
IDIA
R
TX
A5
0
0
0
G
P
U
to
e
n
su
re
e
fficie
n
t
c
o
m
p
u
tati
o
n
,
with
imp
lem
e
n
tati
o
n
s
d
e
v
e
lo
p
e
d
u
si
n
g
P
y
th
o
n
v
e
rsi
o
n
3
.
9
.
1
3
.
Th
e
e
x
p
e
rime
n
tal
fin
d
i
n
g
s
i
n
d
ica
te
th
a
t
YO
LO
su
rp
a
ss
e
s
F
a
ste
r
R
-
CNN
in
p
e
rfo
rm
a
n
c
e
,
a
tt
a
in
in
g
a
n
a
c
c
u
ra
c
y
ra
t
e
o
f
9
0
%
.
Th
e
se
fin
d
i
n
g
s
h
i
g
h
li
g
h
t
th
e
e
ff
e
c
ti
v
e
n
e
ss
o
f
YO
LO
fo
r
re
a
l
-
ti
m
e
o
b
jec
t
d
e
tec
ti
o
n
tas
k
s,
m
a
k
in
g
it
a
su
it
a
b
le
a
n
d
p
re
fe
rre
d
c
h
o
ice
fo
r
ti
m
e
-
se
n
siti
v
e
a
p
p
li
c
a
ti
o
n
s su
c
h
a
s a
u
to
n
o
m
o
u
s
d
riv
i
n
g
sy
ste
m
s.
K
ey
w
o
r
d
s
:
Au
to
n
o
m
o
u
s
d
r
iv
in
g
C
o
m
p
u
ter
v
is
io
n
Dee
p
lear
n
in
g
Ma
ch
in
e
lear
n
in
g
Ob
ject
d
etec
tio
n
Op
tim
izer
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
:
Ma
d
h
u
r
a
M.
B
h
o
s
ale
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
T
elec
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
in
g
J
SP
M
's R
ajar
s
h
i Sh
ah
u
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
Pu
n
e
4
1
1
0
3
3
,
I
n
d
ia
E
m
ail:
Ma
d
h
u
r
ab
h
o
s
ale4
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Ob
ject
d
etec
tio
n
is
a
co
r
e
tas
k
in
co
m
p
u
ter
v
is
io
n
th
at
in
v
o
lv
es
id
en
tify
in
g
an
d
lo
ca
lizin
g
o
b
jects
with
in
im
ag
es
o
r
v
id
eo
f
r
am
es
u
s
in
g
b
o
u
n
d
in
g
b
o
x
es.
Un
lik
e
im
ag
e
class
if
icatio
n
,
wh
i
ch
o
n
ly
d
eter
m
i
n
es
o
b
ject
p
r
esen
ce
,
o
b
ject
d
etec
t
io
n
p
r
o
v
id
es
p
r
ec
is
e
s
p
atial
l
o
ca
lizatio
n
,
en
a
b
lin
g
ap
p
licati
o
n
s
in
a
u
to
n
o
m
o
u
s
d
r
iv
in
g
,
v
id
eo
s
u
r
v
eillan
ce
,
m
ed
ical
im
ag
in
g
,
an
d
in
d
u
s
tr
ial
au
to
m
atio
n
.
E
a
r
ly
o
b
ject
d
etec
tio
n
ap
p
r
o
ac
h
es
r
elied
o
n
h
an
d
cr
af
ted
f
ea
tu
r
e
s
an
d
tr
ad
itio
n
al
m
ac
h
in
e
lea
r
n
in
g
tech
n
i
q
u
es,
s
u
ch
as
Vio
la
an
d
J
o
n
es
[
1
]
,
h
is
to
g
r
am
s
o
f
o
r
ie
n
ted
g
r
a
d
ie
n
ts
(
HOG)
[
2
]
,
d
ef
o
r
m
ab
le
p
ar
t
m
o
d
els
(
DPM)
[
3
]
,
s
elec
t
iv
e
s
ea
r
ch
[
4
]
,
a
n
d
s
ca
le
-
in
v
ar
ian
t
f
ea
t
u
r
e
d
escr
ip
to
r
s
[
5
]
.
Alth
o
u
g
h
e
f
f
ec
tiv
e
f
o
r
co
n
s
tr
ain
ed
s
ce
n
ar
io
s
,
t
h
ese
m
eth
o
d
s
lac
k
ed
r
o
b
u
s
tn
ess
an
d
s
ca
lab
ilit
y
f
o
r
co
m
p
lex
r
ea
l
-
wo
r
ld
en
v
ir
o
n
m
en
ts
.
T
h
e
in
tr
o
d
u
ctio
n
o
f
d
ee
p
lear
n
i
n
g
an
d
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN
s
)
m
ar
k
ed
a
s
ig
n
if
ic
an
t
s
h
if
t
in
o
b
ject
d
etec
tio
n
r
esear
ch
,
en
ab
lin
g
au
to
m
atic
f
ea
tu
r
e
lear
n
in
g
an
d
im
p
r
o
v
ed
g
en
er
aliza
tio
n
[
6
]
.
T
h
is
led
to
t
h
e
d
ev
el
o
p
m
e
n
t
o
f
r
e
g
io
n
-
b
ased
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
R
-
C
NN)
f
r
am
ewo
r
k
s
,
in
clu
d
i
n
g
Fas
t
R
-
C
N
N
[
7
]
an
d
Fas
ter
R
-
C
NN
[
8
]
,
wh
er
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
8
1
-
3
5
9
0
3582
r
eg
io
n
p
r
o
p
o
s
al
n
etwo
r
k
s
(
R
PN)
en
ab
led
en
d
-
to
-
e
n
d
tr
ai
n
in
g
an
d
s
ig
n
if
ican
tly
im
p
r
o
v
ed
d
etec
tio
n
ac
cu
r
ac
y
.
Fas
ter
R
-
C
N
N
an
d
its
ex
ten
s
io
n
s
,
s
u
ch
as
C
ascad
e
R
-
C
NN,
ar
e
ca
teg
o
r
ized
as
two
-
s
tag
e
d
etec
to
r
s
an
d
ar
e
k
n
o
wn
f
o
r
h
ig
h
lo
ca
lizatio
n
p
r
ec
is
io
n
,
with
in
cr
ea
s
ed
co
m
p
u
tatio
n
al
co
s
t
[
8
]
,
[
9
]
.
I
n
p
ar
allel,
o
n
e
-
s
tag
e
d
etec
to
r
s
em
er
g
ed
to
ad
d
r
ess
r
ea
l
-
tim
e
r
eq
u
ir
em
en
ts
b
y
u
n
if
y
in
g
lo
ca
lizatio
n
an
d
class
if
icatio
n
in
to
a
s
in
g
le
n
etwo
r
k
.
Y
o
u
o
n
l
y
lo
o
k
o
n
ce
(
YOL
O)
f
r
am
ewo
r
k
in
tr
o
d
u
ce
d
r
eg
r
ess
io
n
-
b
ase
d
f
o
r
m
u
latio
n
th
at
p
r
o
ce
s
s
es
th
e
en
tire
im
ag
e
i
n
o
n
e
f
o
r
wa
r
d
p
ass
,
en
ab
lin
g
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
[
9
]
.
Su
b
s
eq
u
en
t
d
ev
elo
p
m
en
ts
,
in
clu
d
in
g
an
c
h
o
r
-
f
r
ee
an
d
lig
h
t
weig
h
t
v
ar
ian
ts
,
f
u
r
th
e
r
im
p
r
o
v
ed
ef
f
icien
c
y
an
d
p
er
f
o
r
m
an
ce
ac
r
o
s
s
d
iv
er
s
e
ap
p
licatio
n
s
.
YOL
O
-
b
ased
a
p
p
r
o
ac
h
es
h
av
e
d
em
o
n
s
tr
ated
s
tr
o
n
g
r
esu
lts
in
au
to
n
o
m
o
u
s
d
r
iv
i
n
g
,
e
d
g
e
c
o
m
p
u
tin
g
,
s
en
s
o
r
f
u
s
io
n
,
an
d
ae
r
ial
im
ag
e
an
aly
s
is
[
1
0
]
–
[
1
7
]
.
Sev
er
al
s
tu
d
ies
h
av
e
co
n
d
u
ct
ed
co
m
p
a
r
ativ
e
ev
alu
atio
n
s
o
f
m
o
d
er
n
o
b
ject
d
etec
tio
n
a
lg
o
r
ith
m
s
,
h
ig
h
lig
h
tin
g
th
e
tr
ad
e
-
o
f
f
s
b
etwe
en
ac
cu
r
ac
y
an
d
i
n
f
er
en
ce
s
p
ee
d
am
o
n
g
YOL
O,
s
in
g
le
s
h
o
t
m
u
ltib
o
x
d
etec
to
r
(
SSD
)
,
an
d
Fas
ter
R
-
C
NN
–
b
ased
m
o
d
els
[
1
8
]
–
[
2
3
]
.
T
h
ese
f
in
d
i
n
g
s
em
p
h
asize
p
r
ac
tical
im
p
o
r
tan
ce
o
f
s
elec
tin
g
d
etec
tio
n
f
r
a
m
e
wo
r
k
s
b
ased
o
n
a
p
p
licatio
n
c
o
n
s
tr
ain
ts
.
Mo
tiv
ated
b
y
th
es
e
o
b
s
er
v
atio
n
s
,
th
is
p
ap
er
p
r
esen
ts
a
co
m
p
ar
ativ
e
s
tu
d
y
o
f
YOL
O
an
d
Fas
t
er
R
-
C
NN
u
s
in
g
th
e
KI
T
T
I
d
ataset
[
2
4
]
–
[
2
6
]
.
E
x
p
er
im
en
ts
ar
e
co
n
d
u
cted
o
n
h
ig
h
-
p
er
f
o
r
m
an
ce
h
a
r
d
war
e
eq
u
ip
p
ed
with
an
NVI
DI
A
R
T
X
A5
0
0
0
GPU
to
ev
alu
ate
d
etec
tio
n
ac
cu
r
a
cy
,
in
f
er
en
ce
s
p
ee
d
,
a
n
d
co
m
p
u
t
atio
n
al
ef
f
icien
cy
.
T
h
e
o
b
jec
tiv
e
is
to
p
r
o
v
id
e
p
r
ac
tical
in
s
ig
h
ts
in
to
th
e
s
u
itab
ilit
y
o
f
o
n
e
-
s
tag
e
an
d
two
-
s
tag
e
d
etec
to
r
s
f
o
r
r
ea
l
-
tim
e
an
d
ac
cu
r
ac
y
-
cr
itical
co
m
p
u
ter
v
is
io
n
ap
p
licatio
n
s
.
2.
O
B
J
E
CT
D
E
T
E
C
T
I
O
N
F
R
AM
E
WO
RK
S
2
.
1
.
F
a
s
t
er
re
g
io
n
-
ba
s
ed
co
n
v
o
lutio
na
l neura
l net
wo
rk
Fas
ter
R
-
C
NN
,
in
tr
o
d
u
ce
d
b
y
R
en
et
a
l
.
[
8
]
in
2
0
1
5
,
r
ep
r
esen
ted
a
m
ajo
r
s
tep
f
o
r
war
d
in
o
b
ject
d
etec
tio
n
b
y
p
r
o
p
o
s
in
g
a
u
n
if
i
ed
f
r
a
m
ewo
r
k
th
at
ca
n
b
e
tr
ai
n
ed
in
an
en
d
-
to
-
en
d
m
a
n
n
er
.
I
n
co
n
tr
ast
to
ea
r
lier
ap
p
r
o
ac
h
es
s
u
ch
as
R
-
C
NN
an
d
Fas
t
R
-
C
NN,
wh
ich
d
ep
en
d
ed
o
n
s
ep
ar
ate
e
x
ter
n
a
l
r
eg
io
n
p
r
o
p
o
s
al
tech
n
iq
u
es,
Fas
ter
R
-
C
NN
i
n
teg
r
ates
th
e
R
PN
d
ir
ec
tly
in
to
its
o
v
er
all
ar
ch
itectu
r
e.
T
h
is
in
te
g
r
atio
n
elim
in
ated
th
e
b
o
ttlen
ec
k
o
f
g
en
er
atin
g
r
e
g
io
n
p
r
o
p
o
s
als
s
ep
ar
ately
,
lead
in
g
to
a
s
u
b
s
tan
t
ial
im
p
r
o
v
em
e
n
t
in
s
p
ee
d
an
d
e
f
f
icien
cy
.
Fig
u
r
e
1
illu
s
tr
ates th
e
ar
ch
itectu
r
e
o
f
Fas
ter
R
-
C
NN.
Fig
u
r
e
1
.
C
o
m
p
lete
Fas
ter
R
-
C
NN
ar
ch
itectu
r
e
[
5
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
ea
l
-
time
o
b
ject
d
etec
tio
n
f
o
r
a
u
to
n
o
m
o
u
s
d
r
ivin
g
:
a
co
mp
a
r
a
tive
s
tu
d
y
o
f
…
(
Ma
d
h
u
r
a
M.
B
h
o
s
a
le
)
3583
T
h
e
Fas
ter
R
-
C
NN
ar
ch
itectu
r
e
is
co
m
p
o
s
ed
o
f
th
r
ee
m
ain
co
m
p
o
n
e
n
ts
:
a
co
n
v
o
lu
tio
n
al
b
ac
k
b
o
n
e
f
o
r
e
x
tr
ac
tin
g
f
ea
tu
r
e
m
a
p
s
,
a
R
PN
f
o
r
g
e
n
er
atin
g
ca
n
d
id
ate
o
b
ject
r
eg
io
n
s
,
an
d
a
d
etec
tio
n
h
ea
d
t
h
at
p
er
f
o
r
m
s
o
b
ject
class
if
icatio
n
alo
n
g
with
b
o
u
n
d
i
n
g
b
o
x
r
e
g
r
ess
io
n
.
T
h
e
R
PN
lev
er
ag
es
s
h
ar
ed
co
n
v
o
lu
tio
n
al
f
ea
tu
r
es
to
ef
f
icien
tly
g
e
n
er
at
e
h
ig
h
-
q
u
ality
r
e
g
io
n
p
r
o
p
o
s
al
s
,
s
ig
n
if
ican
tly
r
ed
u
cin
g
c
o
m
p
u
tatio
n
al
o
v
e
r
h
ea
d
.
Ad
d
itio
n
ally
,
th
e
r
eg
io
n
o
f
in
t
er
est
(
R
OI
)
p
o
o
lin
g
lay
er
en
s
u
r
es
th
at
f
ea
tu
r
e
m
ap
s
co
r
r
esp
o
n
d
in
g
to
p
r
o
p
o
s
ed
r
eg
io
n
s
ar
e
r
esized
f
o
r
th
e
d
et
ec
tio
n
s
tag
e,
f
ac
ilit
atin
g
ac
cu
r
ate
class
if
icatio
n
an
d
lo
ca
lizatio
n
.
C
o
m
p
ar
e
d
to
ea
r
lier
m
o
d
els,
Fas
ter
R
-
C
N
N
ac
h
iev
es
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
b
y
tig
h
tly
c
o
u
p
lin
g
th
e
r
eg
io
n
p
r
o
p
o
s
al
an
d
d
etec
tio
n
task
s
with
in
a
s
in
g
le
n
etwo
r
k
.
Fo
r
ex
am
p
le,
it
is
s
i
g
n
if
ican
tly
f
aster
th
an
R
-
C
NN
an
d
Fas
t
R
-
C
N
N
wh
ile
m
ain
tain
in
g
h
i
g
h
ac
c
u
r
ac
y
,
th
is
m
ak
es
it
well
-
s
u
ited
f
o
r
task
s
th
at
d
em
a
n
d
h
ig
h
a
cc
u
r
ac
y
,
in
cl
u
d
in
g
m
ed
ical
im
ag
e
a
n
aly
s
is
,
v
id
eo
p
r
o
ce
s
s
in
g
,
an
d
au
to
n
o
m
o
u
s
d
r
iv
in
g
ap
p
licatio
n
s
.
S
in
ce
its
in
tr
o
d
u
ctio
n
,
Fas
ter
R
-
C
NN
h
as
b
ec
o
m
e
a
f
o
u
n
d
atio
n
al
m
o
d
el
in
o
b
j
ec
t
d
etec
tio
n
r
esear
ch
.
I
t
h
as
b
ee
n
ad
a
p
ted
a
n
d
im
p
r
o
v
e
d
in
v
ar
io
u
s
co
n
tex
ts
,
s
u
ch
as r
ea
l
-
tim
e
d
etec
tio
n
[
8
]
an
d
m
u
lti
-
s
ca
le
d
etec
tio
n
task
s
[
8
]
.
I
ts
ca
p
ab
ilit
y
to
ef
f
ec
tiv
ely
tr
ad
e
o
f
f
b
etwe
en
co
m
p
u
tatio
n
al
s
p
ee
d
an
d
d
e
tectio
n
ac
cu
r
ac
y
h
as
m
ad
e
it
a
s
tan
d
ar
d
r
ef
er
e
n
ce
f
o
r
ass
ess
in
g
co
n
tem
p
o
r
ar
y
o
b
ject
d
etec
tio
n
m
eth
o
d
s
.
2
.
2
.
Yo
u
o
nly
lo
o
k
o
nce
Ob
ject
d
etec
tio
n
is
a
c
o
r
e
p
r
o
b
lem
in
co
m
p
u
ter
v
is
io
n
,
w
id
ely
u
s
ed
in
ar
ea
s
s
u
ch
as
s
elf
-
d
r
iv
in
g
v
eh
icles
an
d
s
ec
u
r
ity
s
u
r
v
eillan
ce
s
y
s
tem
s
.
Am
o
n
g
th
e
m
a
n
y
alg
o
r
ith
m
s
d
ev
elo
p
ed
,
YO
L
O
h
as
em
er
g
ed
as
o
n
e
o
f
th
e
m
o
s
t
wid
ely
ad
o
p
ted
an
d
im
p
ac
t
f
u
l
f
r
am
ewo
r
k
s
f
o
r
r
ea
l
-
tim
e
o
b
ject
d
ete
ctio
n
.
YOL
O
[
1
7
]
r
ev
o
lu
tio
n
ize
d
th
e
f
ield
b
y
p
r
esen
tin
g
a
u
n
if
ied
ar
ch
itectu
r
e
ca
p
ab
le
o
f
d
etec
tin
g
m
u
ltip
l
e
o
b
jects
in
im
ag
es
with
r
em
ar
k
ab
le
s
p
ee
d
an
d
a
cc
u
r
ac
y
.
U
n
lik
e
tr
ad
itio
n
al
o
b
ject
d
etec
tio
n
m
eth
o
d
s
,
wh
i
ch
ap
p
ly
a
s
lid
in
g
win
d
o
w
o
r
R
PN
,
YOL
O
p
er
f
o
r
m
s
d
etec
tio
n
in
a
s
in
g
le
f
o
r
war
d
p
ass
th
r
o
u
g
h
a
d
ee
p
n
eu
r
al
n
etwo
r
k
,
m
ak
in
g
it si
g
n
if
ican
tly
f
aster
an
d
s
u
ita
b
le
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
.
Fig
u
r
e
2
illu
s
tr
ates th
e
YOL
O
ar
ch
itectu
r
e.
Fig
u
r
e
2
.
YOL
O
ar
c
h
itectu
r
e
[
9
]
YOL
O
was
f
ir
s
t
in
tr
o
d
u
ce
d
b
y
R
ed
m
o
n
et
a
l.
[
9
]
in
2
0
1
6
with
th
e
r
elea
s
e
o
f
YOL
Ov
1
.
T
h
e
co
r
e
id
ea
was
to
tr
ea
t
o
b
ject
d
et
ec
tio
n
as
a
s
in
g
le
r
eg
r
ess
io
n
p
r
o
b
lem
,
wh
er
e
th
e
n
etwo
r
k
d
ir
ec
tly
p
r
ed
icts
b
o
u
n
d
in
g
b
o
x
es
an
d
class
p
r
o
b
ab
ilit
ies
f
r
o
m
a
n
im
ag
e.
T
h
is
ap
p
r
o
ac
h
r
esu
lted
in
a
s
ig
n
if
ican
t
s
p
ee
d
ad
v
an
tag
e
o
v
e
r
o
t
h
er
d
etec
tio
n
alg
o
r
ith
m
s
lik
e
R
-
C
NN,
wh
ich
r
eq
u
ir
ed
m
u
lti
-
s
tag
e
p
ip
elin
es
an
d
wer
e
co
m
p
u
tatio
n
ally
ex
p
e
n
s
iv
e.
Su
b
s
eq
u
en
t
v
er
s
io
n
s
o
f
YOL
O,
s
u
ch
as
YOL
Ov
2
(
2
0
1
7
)
an
d
YOL
Ov
3
(
2
0
1
8
)
,
in
tr
o
d
u
ce
d
s
ev
er
al
im
p
r
o
v
em
en
ts
.
YOL
Ov
2
in
co
r
p
o
r
ated
b
atch
n
o
r
m
aliza
tio
n
an
d
an
c
h
o
r
b
o
x
es,
im
p
r
o
v
in
g
ac
cu
r
ac
y
an
d
g
en
er
aliza
tio
n
.
YOL
Ov
3
f
u
r
th
er
en
h
an
ce
d
p
er
f
o
r
m
a
n
ce
b
y
u
s
in
g
a
m
o
r
e
p
o
wer
f
u
l
b
ac
k
b
o
n
e,
Dar
k
n
et
-
5
3
,
an
d
in
tr
o
d
u
cin
g
a
m
u
lti
-
s
ca
le
d
etec
tio
n
a
p
p
r
o
ac
h
,
en
a
b
lin
g
th
e
d
etec
tio
n
o
f
s
m
aller
o
b
jects.
T
h
ese
im
p
r
o
v
em
e
n
ts
allo
wed
YOL
O
to
ac
h
iev
e
s
tate
-
of
-
t
h
e
-
ar
t
ac
cu
r
ac
y
wh
ile
m
ain
ta
in
in
g
its
r
ea
l
-
tim
e
p
er
f
o
r
m
an
ce
,
wh
ich
is
cr
u
cial
f
o
r
m
an
y
r
ea
l
-
wo
r
ld
a
p
p
lic
atio
n
s
.
T
h
e
ev
o
lu
tio
n
o
f
YO
L
O
co
n
tin
u
ed
with
YOL
Ov
4
an
d
YOL
Ov
5
,
wh
er
e
th
e
f
o
cu
s
s
h
if
ted
to
war
d
s
o
p
tim
izin
g
th
e
m
o
d
el
f
o
r
d
if
f
er
en
t
h
ar
d
war
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
8
1
-
3
5
9
0
3584
en
v
ir
o
n
m
en
ts
an
d
in
c
r
ea
s
in
g
r
o
b
u
s
tn
ess
.
YOL
Ov
4
in
co
r
p
o
r
a
ted
s
ev
er
al
en
h
an
ce
m
en
ts
,
in
cl
u
d
in
g
Mo
s
aic
d
ata
au
g
m
en
tatio
n
,
s
elf
-
ad
v
er
s
ar
ial
tr
ain
in
g
,
an
d
c
o
m
p
lete
in
ter
s
ec
tio
n
o
v
er
u
n
io
n
(
C
I
o
U
)
lo
s
s
,
to
b
o
o
s
t
d
etec
tio
n
ac
cu
r
ac
y
.
YOL
Ov
5
,
alth
o
u
g
h
n
o
t
o
f
f
icially
p
a
r
t
o
f
th
e
o
r
ig
in
al
YOL
O
s
er
ies,
b
r
o
u
g
h
t
f
u
r
th
er
o
p
tim
izatio
n
s
with
a
f
o
cu
s
o
n
ea
s
e
o
f
u
s
e
an
d
d
ep
lo
y
m
en
t,
o
f
te
n
b
ein
g
r
eg
ar
d
e
d
f
o
r
its
p
er
f
o
r
m
an
ce
in
ed
g
e
co
m
p
u
tin
g
ap
p
licatio
n
s
.
T
h
e
j
o
u
r
n
ey
o
f
YOL
O
d
id
n
o
t
s
to
p
with
YOL
Ov
5
.
I
n
2
0
2
2
,
YOL
Ov
6
was
i
n
tr
o
d
u
ce
d
,
aim
in
g
at
o
p
tim
izin
g
th
e
f
r
am
ewo
r
k
f
o
r
b
o
t
h
p
er
f
o
r
m
a
n
ce
an
d
ef
f
icien
cy
.
YOL
Ov
6
f
o
cu
s
ed
o
n
im
p
r
o
v
in
g
th
e
b
ac
k
b
o
n
e
ar
c
h
itectu
r
e
an
d
in
c
o
r
p
o
r
ati
n
g
n
ewe
r
tech
n
i
q
u
es
i
n
o
b
ject
d
etec
tio
n
,
s
u
ch
as
th
e
u
s
e
o
f
tr
a
n
s
f
o
r
m
e
r
lay
er
s
f
o
r
b
etter
co
n
tex
t
u
al
u
n
d
er
s
tan
d
in
g
a
n
d
atten
tio
n
m
ec
h
an
is
m
s
.
T
h
e
r
elea
s
e
o
f
YOL
Ov
7
f
u
r
th
er
r
ef
in
ed
t
h
ese
ad
v
an
ce
s
,
im
p
r
o
v
in
g
ac
c
u
r
ac
y
o
n
a
v
ar
iety
o
f
d
atasets
,
en
h
an
cin
g
its
r
o
b
u
s
tn
ess
in
d
etec
tin
g
s
m
all
o
b
jects,
an
d
o
p
ti
m
izin
g
in
f
er
e
n
ce
tim
e.
YOL
Ov
7
also
in
teg
r
ated
im
p
r
o
v
e
d
tr
ain
in
g
s
tr
ateg
ies,
s
u
ch
as
au
g
m
e
n
ted
l
o
s
s
f
u
n
c
tio
n
s
,
to
en
s
u
r
e
b
etter
g
en
e
r
a
lizatio
n
to
d
i
v
er
s
e
en
v
ir
o
n
m
en
ts
.
T
h
e
latest
iter
atio
n
,
YOL
Ov
8
,
is
a
s
u
b
s
tan
tial
leap
f
o
r
war
d
,
o
f
f
er
i
n
g
s
tate
-
of
-
t
h
e
-
ar
t
p
er
f
o
r
m
an
ce
i
n
ter
m
s
o
f
s
p
ee
d
an
d
ac
cu
r
ac
y
.
YOL
Ov
8
f
u
r
t
h
er
o
p
tim
izes
th
e
b
ac
k
b
o
n
e
an
d
f
ea
tu
r
e
ex
tr
ac
to
r
,
p
r
o
v
id
i
n
g
s
ig
n
if
ica
n
t
im
p
r
o
v
e
m
en
ts
in
s
m
all
o
b
ject
d
etec
tio
n
,
m
u
lti
-
class
d
etec
tio
n
,
an
d
m
o
d
el
ad
ap
tab
ilit
y
ac
r
o
s
s
d
if
f
er
e
n
t
d
ev
ices
an
d
u
s
e
ca
s
es.
W
ith
a
f
o
cu
s
o
n
d
ep
l
o
y
in
g
in
ed
g
e
d
ev
ices,
YOL
Ov
8
ad
d
r
ess
es
th
e
ch
allen
g
es
o
f
lo
w
-
r
eso
u
r
ce
en
v
ir
o
n
m
en
ts
,
o
f
f
er
i
n
g
b
o
th
h
ig
h
e
f
f
icien
cy
an
d
r
ea
l
-
tim
e
d
etec
tio
n
ca
p
ab
ilit
ies.
YOL
Ov
8
also
s
u
p
p
o
r
ts
ad
v
an
ce
d
f
ea
tu
r
es
lik
e
b
etter
m
u
lti
-
o
b
ject
tr
ac
k
in
g
an
d
en
h
an
ce
d
p
o
s
t
-
p
r
o
ce
s
s
in
g
tech
n
iq
u
es.
3.
M
E
T
H
O
D
3
.
1
.
O
bje
c
t
det
ec
t
io
n us
ing
y
o
u o
nly
lo
o
k
o
nce
Fig
u
r
e
3
p
r
esen
ts
co
m
p
r
eh
e
n
s
iv
e
wo
r
k
f
lo
w
o
f
th
e
o
b
ject
d
etec
tio
n
f
r
am
ewo
r
k
s
h
o
win
g
b
o
th
th
e
tr
ain
in
g
an
d
test
in
g
s
tag
es.
Fig
u
r
e
3
(
a)
s
h
o
ws
s
ch
em
atic
d
iag
r
am
o
f
t
h
e
m
o
d
el
tr
ain
in
g
p
r
o
ce
d
u
r
e,
wh
ile
Fig
u
r
e
3
(
b
)
s
h
o
ws
s
ch
em
atic
d
iag
r
am
o
f
th
e
m
o
d
el
test
in
g
p
r
o
ce
d
u
r
e.
T
h
e
in
p
u
t
d
atas
et
is
d
iv
id
ed
in
to
tr
ain
in
g
a
n
d
v
alid
atio
n
s
u
b
s
et
s
.
T
h
e
r
esu
ltin
g
d
atasets
ar
e
p
r
o
v
id
ed
as
in
p
u
t
to
t
h
e
o
b
ject
d
etec
tio
n
m
o
d
el.
I
n
th
is
s
tu
d
y
,
YOL
Ov
8
,
o
n
e
o
f
th
e
m
o
s
t
r
ec
e
n
t
ad
v
an
ce
m
e
n
ts
in
o
b
ject
d
etec
tio
n
,
was
em
p
lo
y
ed
.
Pro
p
e
r
h
y
p
er
p
ar
am
eter
c
o
n
f
i
g
u
r
atio
n
p
lay
s
a
v
ital
r
o
le
in
ac
h
ie
v
in
g
ef
f
icien
t
m
o
d
el
tr
ain
in
g
.
T
h
e
m
o
d
el
was
tr
ain
ed
u
s
in
g
a
b
atch
s
ize
o
f
1
6
,
an
in
p
u
t
im
ag
e
r
eso
lu
tio
n
o
f
6
4
0
,
an
d
1
5
0
tr
ain
in
g
ep
o
c
h
s
.
T
h
e
s
elec
ted
YOL
O
ar
ch
itectu
r
e
was
YOL
Ov
8
-
x
,
wh
ile
th
e
Ad
am
o
p
tim
izer
was
ap
p
lied
with
m
o
m
en
tu
m
p
ar
am
ete
r
s
s
et
t
o
0
.
2
2
2
,
0
.
5
5
5
,
an
d
0
.
9
9
9
,
r
esp
ec
tiv
ely
.
Af
ter
th
e
m
o
d
el
is
tr
ai
n
ed
u
s
in
g
th
e
tr
ain
in
g
d
ataset
an
d
v
alid
ated
with
th
e
v
alid
atio
n
d
ataset,
u
n
s
ee
n
d
ata
is
s
u
p
p
lied
to
ev
alu
ate
its
p
r
ed
ictio
n
ca
p
a
b
ilit
y
.
Fig
u
r
e
3
(
b
)
illu
s
tr
ates
th
e
test
in
g
wo
r
k
f
lo
w
o
f
th
e
o
b
ject
d
etec
tio
n
s
y
s
tem
.
(
a)
(
b
)
Fig
u
r
e
3
.
Sch
em
atic
r
ep
r
esen
t
atio
n
o
f
th
e
YOL
O
-
b
ased
o
b
je
ct
d
etec
tio
n
m
o
d
el:
(
a)
tr
ain
i
n
g
p
r
o
ce
s
s
an
d
(
b
)
test
in
g
p
r
o
ce
s
s
In
p
u
t
d
a
tas
e
t
(8
0
%
train
in
g
a
n
d
2
0
%
v
a
li
d
a
ti
o
n
)
YO
LO
-
V8
m
o
d
e
l
Ba
tch
siz
e
=
1
6
Im
a
g
e
siz
e
=
6
4
0
Ep
o
c
h
s
=
1
5
0
Yo
lo
v
8
-
x
Op
ti
m
ize
r
=
Ad
a
m
M
o
m
e
n
tu
m
=
0
.
2
2
2
,
0
.
5
5
5
,
0
.
9
9
9
Train
e
d
m
o
d
e
l
Tes
t
d
a
ta i
m
a
g
e
s/v
id
e
o
s
Train
e
d
m
o
d
e
l
Ob
jec
t
d
e
tec
ti
o
n
re
su
lt
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
ea
l
-
time
o
b
ject
d
etec
tio
n
f
o
r
a
u
to
n
o
m
o
u
s
d
r
ivin
g
:
a
co
mp
a
r
a
tive
s
tu
d
y
o
f
…
(
Ma
d
h
u
r
a
M.
B
h
o
s
a
le
)
3585
3
.
2
.
O
bje
c
t
det
ec
t
io
n us
ing
F
a
s
t
er
re
g
io
n
-
ba
s
ed
co
nv
o
lutio
na
l neura
l net
wo
rk
Fig
u
r
e
4
p
r
esen
ts
th
e
co
m
p
let
e
wo
r
k
f
l
o
w
o
f
th
e
o
b
ject
d
etec
tio
n
f
r
am
ew
o
r
k
,
co
v
e
r
in
g
b
o
th
tr
ain
in
g
an
d
test
in
g
p
h
ases
.
Fig
u
r
e
4
(
a)
s
h
o
ws
th
e
s
ch
e
m
atic
r
ep
r
esen
tatio
n
o
f
t
h
e
m
o
d
el
tr
ain
in
g
p
i
p
elin
e,
wh
ile
Fig
u
r
e
4
(
b
)
s
h
o
ws
th
e
s
ch
em
atic
r
ep
r
esen
tatio
n
o
f
th
e
m
o
d
el
test
in
g
p
ip
elin
e.
T
h
e
i
n
p
u
t
d
ataset
is
s
p
lit
in
to
tr
ain
in
g
an
d
v
alid
atio
n
s
ets,
wh
ich
ar
e
th
en
u
s
ed
to
tr
ain
th
e
o
b
ject
d
etec
tio
n
m
o
d
el.
I
n
th
is
wo
r
k
,
Fas
ter
R
-
C
NN,
a
s
tate
-
of
-
th
e
-
ar
t o
b
ject
d
etec
tio
n
ar
ch
itectu
r
e,
is
u
tili
ze
d
.
Pro
p
er
s
elec
tio
n
o
f
h
y
p
e
r
p
ar
am
ete
r
s
is
cr
itical
f
o
r
e
f
f
ec
tiv
e
m
o
d
el
tr
ain
in
g
.
I
n
th
is
s
tu
d
y
,
th
e
m
o
d
el
is
tr
ain
ed
with
a
b
atch
s
ize
o
f
1
6
,
an
i
n
p
u
t
im
ag
e
r
eso
lu
tio
n
o
f
6
4
0
,
an
d
1
5
0
ep
o
ch
s
,
u
s
in
g
th
e
Ad
am
o
p
tim
izer
with
m
o
m
en
t
u
m
p
a
r
am
eter
s
o
f
0
.
2
2
2
,
0
.
5
5
5
,
an
d
0
.
9
9
9
.
Fu
r
t
h
er
m
o
r
e,
ex
p
er
im
e
n
ts
ar
e
p
er
f
o
r
m
e
d
u
s
in
g
two
b
ac
k
b
o
n
e
n
etwo
r
k
s
,
R
esNet5
0
an
d
R
esNet1
0
1
,
to
ev
alu
ate
th
eir
i
m
p
ac
t
o
n
p
er
f
o
r
m
an
ce
.
A
f
ter
tr
ain
in
g
an
d
v
alid
atio
n
,
th
e
m
o
d
el
is
test
ed
u
s
in
g
u
n
s
ee
n
d
ata
to
ass
ess
its
p
r
ed
ictiv
e
ca
p
ab
ilit
y
.
Fig
u
r
e
4
(
b
)
illu
s
tr
ates
th
e
te
s
tin
g
wo
r
k
f
lo
w
o
f
th
e
o
b
ject
d
etec
tio
n
s
y
s
tem
.
(
a)
(
b
)
Fig
u
r
e
4
.
Sch
em
atic
d
iag
r
am
o
f
th
e
Fas
ter
R
-
C
NN
-
b
ased
o
b
ject
d
etec
tio
n
m
o
d
el:
(
a)
tr
ain
i
n
g
wo
r
k
f
lo
w
an
d
(
b
)
test
in
g
wo
r
k
f
lo
w
4.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
e
ex
p
e
r
im
en
ts
in
th
is
s
tu
d
y
wer
e
ca
r
r
ied
o
u
t
u
s
in
g
Py
th
o
n
3
.
9
.
1
3
o
n
a
W
in
d
o
ws
1
0
Pr
o
s
y
s
tem
,
with
3
2
GB
R
AM
an
d
a
1
2
th
-
g
en
er
atio
n
I
n
tel(
R
)
C
o
r
e
(
T
M)
i9
-
1
2
9
0
0
p
r
o
ce
s
s
o
r
r
u
n
n
i
n
g
at
2
.
4
0
GHz
.
Fo
r
ac
ce
ler
ated
co
m
p
u
tatio
n
,
an
NVI
DI
A
R
T
X
A5
0
0
0
GPU
w
ith
1
2
GB
V
R
AM
wa
s
u
s
ed
,
alo
n
g
with
Py
T
o
r
c
h
2
.
0
.
1
an
d
C
UDA
1
1
.
7
.
T
h
e
d
ataset
em
p
lo
y
ed
is
th
e
p
u
b
lic
ly
av
ailab
le
KI
T
T
I
d
ataset
[
2
6
]
,
wh
ich
co
n
tain
s
m
u
ltip
le
o
b
ject
ca
teg
o
r
ies in
cl
u
d
in
g
ca
r
s
,
p
e
d
estrian
s
,
v
an
s
,
cy
clis
ts
,
tr
u
ck
s
,
m
is
ce
llan
eo
u
s
o
b
jects,
tr
am
s
,
an
d
s
ea
ted
p
er
s
o
n
s
.
I
n
to
tal,
th
e
d
ataset
co
n
s
is
ts
o
f
ap
p
r
o
x
im
a
tely
5
,
9
8
4
tr
ain
in
g
im
ag
es
an
d
1
,
4
0
0
v
alid
atio
n
im
ag
es.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
YOL
Ov
8
was
an
aly
ze
d
b
y
m
o
d
if
y
in
g
th
e
m
o
m
en
tu
m
p
ar
a
m
eter
o
f
th
e
Ad
am
o
p
tim
izer
wh
ile
k
ee
p
in
g
o
th
er
s
ettin
g
s
u
n
ch
an
g
ed
.
T
h
e
t
e
s
t
ed
m
o
m
e
n
t
u
m
v
a
l
u
e
s
w
e
r
e
0
.
0
0
2
,
0
.
0
0
5
,
a
n
d
0
.
9
9
9
,
w
i
t
h
f
i
x
e
d
c
o
n
f
i
g
u
r
a
ti
o
n
s
o
f
1
5
0
e
p
o
c
h
s
,
b
a
t
c
h
s
i
z
e
o
f
1
6
,
im
a
g
e
r
e
s
o
l
u
ti
o
n
o
f
6
4
0
,
a
n
d
th
e
Y
O
L
O
v
8
-
x
[
9
]
m
o
d
el
v
a
r
ian
t.
As
s
h
o
wn
i
n
Fig
u
r
e
5
,
w
h
en
th
e
m
o
m
e
n
tu
m
was
s
et
to
0
.
2
2
2
as
s
h
o
wn
in
Fig
u
r
e
5
(
a)
,
th
e
m
AP@
5
0
an
d
m
AP@
5
0
–
9
5
s
co
r
es
r
em
ain
e
d
b
elo
w
9
0
%
an
d
7
0
%,
r
esp
ec
tiv
ely
.
W
ith
a
m
o
m
en
tu
m
v
alu
e
o
f
0
.
5
5
5
as
s
h
o
wn
in
Fig
u
r
e
5
(
b
)
,
th
e
m
o
d
el
ac
h
iev
e
d
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
,
with
m
AP@
5
0
ex
ce
ed
in
g
9
0
%
an
d
m
AP@
5
0
–
9
5
s
u
r
p
ass
in
g
7
0
%.
Ho
wev
er
,
wh
e
n
th
e
m
o
m
en
tu
m
was
in
cr
e
ased
to
0
.
9
9
9
as
s
h
o
wn
in
Fig
u
r
e
5
(
c)
,
p
er
f
o
r
m
a
n
ce
d
ec
lin
ed
,
with
m
AP@
5
0
an
d
m
AP@
5
0
–
9
5
f
allin
g
b
elo
w
8
0
%
a
n
d
6
0
%
,
r
esp
ec
tiv
ely
.
Fo
r
Fas
ter
R
-
C
NN
[
8
]
with
a
R
esNet5
0
b
ac
k
b
o
n
e,
th
e
ac
cu
r
ac
ies
ac
h
ie
v
ed
with
m
o
m
en
tu
m
v
alu
es
o
f
0
.
2
2
2
,
0
.
5
5
5
,
an
d
0
.
9
9
9
wer
e
1
8
.
3
3
%,
2
6
.
5
7
%,
an
d
3
8
.
3
1
%,
r
esp
ec
tiv
el
y
.
Similar
ly
,
with
R
esNet1
0
1
b
ac
k
b
o
n
e,
th
e
co
r
r
esp
o
n
d
i
n
g
ac
cu
r
ac
ies
wer
e
2
0
.
8
6
%,
2
8
.
0
3
%,
an
d
3
8
.
6
4
%.
Fig
u
r
es
6
to
1
0
in
d
ic
at
ed
F
aste
r
R
-
C
NN
m
AP
co
m
p
a
r
a
ti
v
e
p
l
o
t
o
f
R
e
s
n
et
5
0
a
n
d
R
es
n
et
1
0
1
b
ac
k
b
o
n
e,
m
AP
v
e
r
s
u
s
m
o
m
e
n
t
u
m
v
al
u
e
p
l
o
t
o
f
Fas
te
r
R
-
C
NN
,
b
ac
k
o
n
e
v
e
r
s
u
s
m
o
m
e
n
t
u
m
v
a
lu
e
b
o
x
p
lo
t o
f
Fas
t
er
R
-
C
N
N
,
b
a
ck
o
n
e
v
e
r
s
u
s
m
o
m
en
tu
m
v
al
u
e
d
o
t
p
l
o
t
o
f
Fas
t
e
r
R
-
C
NN
,
m
o
m
e
n
t
u
m
a
n
d
Ma
p
5
0
-
9
5
co
m
p
a
r
i
s
o
n
b
y
b
a
c
k
b
o
n
e
o
f
Fas
te
r
R
-
C
NN
.
T
a
b
l
e
1
i
n
d
ic
at
ed
c
o
m
p
ar
ati
v
e
p
e
r
f
o
r
m
a
n
c
e
a
n
al
y
s
is
b
et
we
en
YO
L
O
a
n
d
F
a
s
ter
R
-
C
N
N.
In
p
u
t
d
a
tas
e
t
(8
0
%
train
in
g
a
n
d
2
0
%
v
a
li
d
a
ti
o
n
)
F
a
ste
r
R
-
CNN
m
o
d
e
l
Ba
tch
siz
e
=
1
6
Im
a
g
e
siz
e
=
6
4
0
Ep
o
c
h
s
=
1
5
0
Yo
lo
v
8
-
x
Op
ti
m
ize
r
=
Ad
a
m
M
o
m
e
n
tu
m
=
0
.
2
2
2
,
0
.
5
5
5
,
0
.
9
9
9
Ba
c
k
b
o
n
e
=
Re
sn
e
t
5
0
,
Re
sn
e
t1
0
1
Train
e
d
m
o
d
e
l
Tes
t
d
a
ta i
m
a
g
e
s/v
id
e
o
s
Train
e
d
m
o
d
e
l
Ob
jec
t
d
e
tec
ti
o
n
re
su
lt
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
8
1
-
3
5
9
0
3586
(
a)
(
b
)
(
c)
Fig
u
r
e
5
.
Per
f
o
r
m
an
c
e
an
aly
s
i
s
with
(
a)
0
.
2
2
2
m
o
m
en
tu
m
,
(
b
)
0
.
5
5
5
m
o
m
en
t
u
m
,
an
d
(
c)
0
.
9
9
9
m
o
m
en
t
u
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
ea
l
-
time
o
b
ject
d
etec
tio
n
f
o
r
a
u
to
n
o
m
o
u
s
d
r
ivin
g
:
a
co
mp
a
r
a
tive
s
tu
d
y
o
f
…
(
Ma
d
h
u
r
a
M.
B
h
o
s
a
le
)
3587
Fig
u
r
e
6
.
Fas
ter
R
-
C
NN
m
AP
co
m
p
ar
ativ
e
p
lo
t o
f
R
esn
et5
0
an
d
R
esn
et1
0
1
b
ac
k
b
o
n
e
Fig
u
r
e
7
.
B
ac
k
b
o
n
e
v
er
s
u
s
m
o
m
en
tu
m
v
al
u
e
p
lo
t
o
f
Fas
ter
R
-
C
NN
Fig
u
r
e
8
.
B
ac
k
o
n
e
v
er
s
u
s
m
o
m
en
tu
m
v
al
u
e
b
o
x
p
lo
t
o
f
Fas
ter
R
-
C
NN
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
8
1
-
3
5
9
0
3588
Fig
u
r
e
9
.
B
ac
k
o
n
e
v
er
s
u
s
m
o
m
en
tu
m
v
al
u
e
d
o
t
p
lo
t o
f
Fas
ter
R
-
C
NN
Fig
u
r
e
1
0
.
Mo
m
en
tu
m
an
d
m
AP
50
-
9
5
c
o
m
p
ar
is
o
n
b
y
b
ac
k
b
o
n
e
o
f
Fas
ter
R
-
C
NN
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
b
etwe
en
YOL
O
an
d
Fas
ter
R
-
C
NN
S
r
.
n
o
Y
O
LO
v
e
r
s
i
o
n
Y
O
LO
v
a
r
i
a
n
t
M
o
me
n
t
u
m
v
a
l
u
e
mA
P
5
0
(
%)
Tr
a
i
n
i
n
g
t
i
me
1
Y
O
LO
-
V8
Y
O
LO
V
8
-
x
0
.
2
2
2
90
3
.
4
8
2
h
o
u
r
s
2
Y
O
LO
-
V8
Y
O
LO
V
8
-
x
0
.
5
5
5
90
3
.
2
7
1
h
o
u
r
s
3
Y
O
LO
-
V8
Y
O
LO
V
8
-
x
0
.
9
9
9
80
3
.
3
7
5
h
o
u
r
s
4
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
5
0
b
a
c
k
b
o
n
e
0
.
2
2
2
1
8
.
3
3
1
3
.
0
8
m
i
n
u
t
e
s
5
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
5
0
b
a
c
k
b
o
n
e
0
.
5
5
5
2
6
.
5
7
1
3
.
1
6
m
i
n
u
t
e
s
6
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
5
0
b
a
c
k
b
o
n
e
0
.
9
9
9
3
8
.
3
1
1
3
.
0
9
m
i
n
u
t
e
s
7
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
1
0
1
b
a
c
k
b
o
n
e
0
.
2
2
2
2
0
.
8
6
1
7
.
2
3
m
i
n
u
t
e
s
8
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
1
0
1
b
a
c
k
b
o
n
e
0
.
5
5
5
2
8
.
0
3
6
1
7
.
2
5
m
i
n
u
t
e
s
9
F
a
st
e
r
R
-
C
N
N
R
e
s
n
e
t
1
0
1
b
a
c
k
b
o
n
e
0
.
9
9
9
3
8
.
6
6
1
7
.
2
6
m
i
n
u
t
e
s
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
esen
ts
a
c
o
m
p
ar
ativ
e
ev
alu
atio
n
o
f
YOL
Ov
8
an
d
Fas
ter
R
-
C
NN
o
n
th
e
KI
T
T
I
d
ataset,
f
o
cu
s
in
g
o
n
h
o
w
d
if
f
e
r
en
t
Ad
am
o
p
tim
izer
m
o
m
e
n
tu
m
s
ettin
g
s
in
f
lu
en
ce
m
o
d
el
p
er
f
o
r
m
an
ce
.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
in
d
icate
th
at
YOL
Ov
8
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
s
Fas
ter
R
-
C
NN
ac
r
o
s
s
all
te
s
ted
co
n
f
ig
u
r
atio
n
s
.
Fo
r
YOL
Ov
8
,
th
e
b
est
r
esu
lts
wer
e
o
b
tain
ed
at
a
m
o
m
en
tu
m
v
al
u
e
o
f
0
.
5
5
5
,
wh
e
r
e
m
AP@
5
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
R
ea
l
-
time
o
b
ject
d
etec
tio
n
f
o
r
a
u
to
n
o
m
o
u
s
d
r
ivin
g
:
a
co
mp
a
r
a
tive
s
tu
d
y
o
f
…
(
Ma
d
h
u
r
a
M.
B
h
o
s
a
le
)
3589
ex
ce
ed
ed
9
0
%
an
d
m
AP@
5
0
–
9
5
was
ab
o
v
e
7
0
%.
I
n
co
m
p
ar
is
o
n
,
u
s
in
g
lo
wer
(
0
.
2
2
2
)
an
d
h
ig
h
e
r
(
0
.
9
9
9
)
m
o
m
en
tu
m
v
al
u
es
led
to
a
n
o
ticea
b
le
d
r
o
p
in
ac
cu
r
ac
y
,
s
h
o
win
g
th
at
YOL
Ov
8
p
er
f
o
r
m
an
ce
is
af
f
ec
ted
b
y
ch
an
g
es
in
t
h
is
p
ar
am
eter
.
Fo
r
Fas
ter
R
-
C
NN,
p
er
f
o
r
m
an
ce
d
ep
e
n
d
ed
o
n
th
e
b
ac
k
b
o
n
e
a
r
ch
itectu
r
e.
T
h
e
R
esNet1
0
1
b
ac
k
b
o
n
e
g
en
er
ally
p
r
o
d
u
ce
d
b
etter
r
esu
lts
th
an
R
esNet5
0
.
Ho
wev
er
,
ev
e
n
with
R
esNet1
0
1
an
d
th
e
b
est
-
p
er
f
o
r
m
in
g
m
o
m
en
tu
m
s
ettin
g
,
Fas
ter
R
-
C
NN
s
till
ac
h
iev
ed
lo
wer
ac
cu
r
ac
y
t
h
an
YOL
Ov
8
i
n
all
ca
s
es.
Ov
er
all,
th
e
f
in
d
in
g
s
em
p
h
asize
YOL
Ov
8
’
s
s
u
p
er
i
o
r
ity
in
ter
m
s
o
f
ac
c
u
r
ac
y
an
d
s
u
itab
ilit
y
f
o
r
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
task
s
co
m
p
ar
ed
to
Fas
ter
R
-
C
NN.
Fu
tu
r
e
r
esear
ch
co
u
ld
f
u
r
th
er
ex
p
lo
r
e
ad
v
an
ce
m
e
n
ts
in
m
o
d
el
ar
ch
it
ec
tu
r
es
an
d
ev
alu
ate
th
eir
ef
f
ec
tiv
en
ess
ac
r
o
s
s
d
iv
er
s
e
d
atasets
an
d
co
n
d
itio
n
s
to
ex
p
an
d
th
e
a
p
p
licab
ilit
y
o
f
th
ese
f
in
d
in
g
s
.
ACK
NO
WL
E
DG
M
E
N
T
S
W
e
ac
k
n
o
wled
g
e
J
SP
M'
s
R
aj
ar
s
h
i
Sh
ah
u
C
o
lleg
e
o
f
E
n
g
i
n
ee
r
in
g
,
Pu
n
e,
I
n
d
ia,
f
o
r
p
r
o
v
id
in
g
th
e
in
f
r
astru
ctu
r
e,
la
b
o
r
ato
r
y
f
ac
il
ities
,
an
d
ac
ad
em
ic
en
v
ir
o
n
m
e
n
t n
ec
ess
ar
y
f
o
r
co
n
d
u
ctin
g
t
h
is
r
esear
ch
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ma
d
h
u
r
a
M.
B
h
o
s
ale
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Yo
g
esh
S.
An
g
al
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
h
av
e
n
o
co
n
f
lict
o
f
in
ter
est
to
d
ec
lar
e.
All
c
o
-
au
th
o
r
s
h
av
e
s
ee
n
a
n
d
a
g
r
e
e
with
th
e
co
n
ten
t
o
f
m
an
u
s
cr
ip
t.
W
e
c
o
n
f
ir
m
e
d
th
at
s
u
b
m
is
s
io
n
is
o
r
ig
in
al
w
o
r
k
is
n
o
t
u
n
d
er
r
ev
iew
at
a
n
y
o
th
e
r
p
u
b
licatio
n
.
au
th
o
r
s
d
o
n
’
t
h
av
e
an
y
f
in
an
cial
an
d
n
o
n
-
f
in
an
cial
in
ter
est
in
th
e
s
u
b
ject
m
atter
o
r
m
ater
ial
d
is
cu
s
s
ed
in
th
is
m
an
u
s
cr
ip
t.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
is
KI
T
T
I
d
ataset.
T
h
is
d
ataset
o
p
en
ly
av
ailab
le
in
Kag
g
le
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/k
lem
en
k
o
/k
itti
-
d
ataset.
RE
F
E
R
E
NC
E
S
[
1
]
P
.
V
i
o
l
a
a
n
d
M
.
J
o
n
e
s,
“
R
a
p
i
d
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
u
s
i
n
g
a
b
o
o
s
t
e
d
c
a
sc
a
d
e
o
f
si
m
p
l
e
f
e
a
t
u
r
e
s
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
0
1
I
E
EE
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
.
C
VP
R
2
0
0
1
,
2
0
0
1
,
p
p
.
1
-
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
0
1
.
9
9
0
5
1
7
.
[
2
]
N
.
D
a
l
a
l
a
n
d
B
.
Tr
i
g
g
s,
“
H
i
st
o
g
r
a
ms
o
f
o
r
i
e
n
t
e
d
g
r
a
d
i
e
n
t
s
f
o
r
h
u
ma
n
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
0
5
I
EE
E
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
(
C
VP
R’
0
5
)
,
2
0
0
5
,
p
p
.
8
8
6
–
8
9
3
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
0
5
.
1
7
7
.
[
3
]
P
.
F
e
l
z
e
n
szwa
l
b
,
D
.
M
c
A
l
l
e
s
t
e
r
,
a
n
d
D
.
R
a
m
a
n
a
n
,
“
A
d
i
s
c
r
i
m
i
n
a
t
i
v
e
l
y
t
r
a
i
n
e
d
,
m
u
l
t
i
s
c
a
l
e
,
d
e
f
o
r
ma
b
l
e
p
a
r
t
mo
d
e
l
,
”
i
n
2
0
0
8
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
0
8
,
p
p
.
1
–
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
0
8
.
4
5
8
7
5
9
7
.
[
4
]
J.
R
.
R
.
U
i
j
l
i
n
g
s
,
K
.
E
.
A
.
V
.
D
.
S
a
n
d
e
,
T
.
G
e
v
e
r
s,
a
n
d
A
.
W
.
M
.
S
meu
l
d
e
r
s,
“
S
e
l
e
c
t
i
v
e
s
e
a
r
c
h
f
o
r
o
b
j
e
c
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
m
p
u
t
e
r
Vi
si
o
n
,
v
o
l
.
1
0
4
,
n
o
.
2
,
p
p
.
1
5
4
–
1
7
1
,
S
e
p
.
2
0
1
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
3
-
0
1
3
-
0
6
2
0
-
5.
[
5
]
D
.
G
.
L
o
w
e
,
“
O
b
j
e
c
t
r
e
c
o
g
n
i
t
i
o
n
f
r
o
m
l
o
c
a
l
s
c
a
l
e
-
i
n
v
a
r
i
a
n
t
f
e
a
t
u
r
e
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
S
e
v
e
n
t
h
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
N
o
v
.
1
9
9
9
,
p
p
.
1
1
5
0
–
1
1
5
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
1
9
9
9
.
7
9
0
4
1
0
.
[
6
]
C
.
S
z
e
g
e
d
y
e
t
a
l
.
,
“
G
o
i
n
g
d
e
e
p
e
r
w
i
t
h
c
o
n
v
o
l
u
t
i
o
n
s,”
i
n
2
0
1
5
I
E
EE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
Ju
n
.
2
0
1
5
,
p
p
.
1
–
9
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
5
.
7
2
9
8
5
9
4
.
[
7
]
R
.
G
i
r
s
h
i
c
k
,
“
F
a
st
R
-
C
N
N
,
”
i
n
2
0
1
5
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
(
I
C
C
V)
,
D
e
c
.
2
0
1
5
,
p
p
.
1
4
4
0
–
1
4
4
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
1
5
.
1
6
9
.
[
8
]
S
.
R
e
n
,
K
.
H
e
,
R
.
G
i
r
sh
i
c
k
,
a
n
d
J
.
S
u
n
,
“
F
a
s
t
e
r
R
-
C
N
N
:
t
o
w
a
r
d
s
r
e
a
l
-
t
i
m
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
w
i
t
h
r
e
g
i
o
n
p
r
o
p
o
sal
n
e
t
w
o
r
k
s
,
”
i
n
2
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
N
e
u
ra
l
I
n
f
o
rm
a
t
i
o
n
Pro
c
e
ss
i
n
g
S
y
s
t
e
m
s
,
2
0
1
5
,
p
p
.
9
1
–
9
9
,
d
o
i
:
1
1
0
9
/
TPA
M
I
.
2
0
1
6
.
2
5
7
7
0
3
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
5
8
1
-
3
5
9
0
3590
[
9
]
J.
R
e
d
m
o
n
,
S
.
D
i
v
v
a
l
a
,
R
.
G
i
r
sh
i
c
k
,
a
n
d
A
.
F
a
r
h
a
d
i
,
“
Y
o
u
o
n
l
y
l
o
o
k
o
n
c
e
:
u
n
i
f
i
e
d
,
r
e
a
l
-
t
i
me
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
6
I
E
E
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
(
C
VP
R)
,
J
u
n
.
2
0
1
6
,
p
p
.
7
7
9
–
7
8
8
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
6
.
9
1
.
[
1
0
]
P
.
D
a
i
,
S
.
Y
a
o
,
Z.
Li
,
S
.
Z
h
a
n
g
,
a
n
d
X
.
C
a
o
,
“
A
C
E:
a
n
c
h
o
r
-
f
r
e
e
c
o
r
n
e
r
e
v
o
l
u
t
i
o
n
f
o
r
r
e
a
l
-
t
i
me
a
r
b
i
t
r
a
r
i
l
y
o
r
i
e
n
t
e
d
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
m
a
g
e
Pro
c
e
ss
i
n
g
,
v
o
l
.
3
1
,
p
p
.
4
0
7
6
–
4
0
8
9
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
P
.
2
0
2
2
.
3
1
6
7
9
1
9
.
[
1
1
]
N
.
M
.
A
.
A
.
D
a
z
l
e
e
,
S
.
A
.
K
h
a
l
i
l
,
S
.
A
.
-
R
a
h
ma
n
,
a
n
d
S
.
M
u
t
a
l
i
b
,
“
O
b
j
e
c
t
d
e
t
e
c
t
i
o
n
f
o
r
a
u
t
o
n
o
m
o
u
s
v
e
h
i
c
l
e
s
w
i
t
h
se
n
s
o
r
-
b
a
s
e
d
t
e
c
h
n
o
l
o
g
y
u
si
n
g
Y
O
LO
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
i
n
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
1
2
9
–
1
3
4
,
M
a
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
8
2
0
1
/
i
j
i
sae
.
2
0
2
2
.
2
7
6
.
[
1
2
]
S
.
Li
a
n
g
e
t
a
l
.
,
“
E
d
g
e
Y
O
LO
:
r
e
a
l
-
t
i
m
e
i
n
t
e
l
l
i
g
e
n
t
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
sy
s
t
e
m
b
a
se
d
o
n
e
d
g
e
-
c
l
o
u
d
c
o
o
p
e
r
a
t
i
o
n
i
n
a
u
t
o
n
o
m
o
u
s
v
e
h
i
c
l
e
s,”
I
EE
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
I
n
t
e
l
l
i
g
e
n
t
T
ra
n
s
p
o
r
t
a
t
i
o
n
S
y
st
e
m
s
,
v
o
l
.
2
3
,
n
o
.
1
2
,
p
p
.
2
5
3
4
5
–
2
5
3
6
0
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
TS.2
0
2
2
.
3
1
5
8
2
5
3
.
[
1
3
]
M
.
B
.
B
l
a
sc
h
k
o
a
n
d
C
.
H
.
La
mp
e
r
t
,
“
Le
a
r
n
i
n
g
t
o
l
o
c
a
l
i
z
e
o
b
j
e
c
t
s
w
i
t
h
s
t
r
u
c
t
u
r
e
d
o
u
t
p
u
t
r
e
g
r
e
ssi
o
n
,
”
i
n
C
o
m
p
u
t
e
r
Vi
si
o
n
–
EC
C
V
2
0
0
8
(
E
C
C
V
2
0
0
8
)
,
B
e
r
l
i
n
,
H
e
i
d
e
l
b
e
r
g
:
S
p
r
i
n
g
e
r
,
2
0
0
8
,
p
p
.
2
–
1
5
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
540
-
8
8
6
8
2
-
2
_
2
.
[
1
4
]
X
.
Zo
u
,
“
A
r
e
v
i
e
w
o
f
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
e
s,”
i
n
2
0
1
9
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
S
m
a
r
t
G
ri
d
a
n
d
El
e
c
t
ri
c
a
l
A
u
t
o
m
a
t
i
o
n
(
I
C
S
G
EA)
,
A
u
g
.
2
0
1
9
,
p
p
.
2
5
1
–
2
5
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
S
G
EA
.
2
0
1
9
.
0
0
0
6
5
.
[
1
5
]
J.
Te
r
v
e
n
,
D
.
-
M
.
C
.
-
Es
p
a
r
z
a
,
a
n
d
J.
-
A
.
R
.
-
G
o
n
z
á
l
e
z
,
“
A
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
Y
O
LO
a
r
c
h
i
t
e
c
t
u
r
e
s i
n
c
o
m
p
u
t
e
r
v
i
s
i
o
n
:
f
r
o
m
Y
O
LO
v
1
t
o
Y
O
LO
v
8
a
n
d
Y
O
LO
-
N
A
S
,
”
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
a
n
d
K
n
o
w
l
e
d
g
e
Ex
t
r
a
c
t
i
o
n
,
v
o
l
.
5
,
n
o
.
4
,
p
p
.
1
6
8
0
–
1
7
1
6
,
N
o
v
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
ma
k
e
5
0
4
0
0
8
3
.
[
1
6
]
J.
H
a
n
,
Y
.
L
i
a
o
,
J
.
Z
h
a
n
g
,
S
.
W
a
n
g
,
a
n
d
S
.
Li
,
“
T
a
r
g
e
t
f
u
si
o
n
d
e
t
e
c
t
i
o
n
o
f
Li
D
A
R
a
n
d
c
a
m
e
r
a
b
a
s
e
d
o
n
t
h
e
i
mp
r
o
v
e
d
Y
O
LO
a
l
g
o
r
i
t
h
m,
”
M
a
t
h
e
m
a
t
i
c
s
,
v
o
l
.
6
,
n
o
.
1
0
,
O
c
t
.
2
0
1
8
,
d
o
i
:
1
0
.
3
3
9
0
/
mat
h
6
1
0
0
2
1
3
.
[
1
7
]
Q
.
H
e
,
A
.
X
u
,
Z.
Y
e
,
W
.
Zh
o
u
,
a
n
d
T.
C
a
i
,
“
O
b
j
e
c
t
d
e
t
e
c
t
i
o
n
b
a
se
d
o
n
Li
g
h
t
W
e
i
g
h
t
Y
O
LO
X
f
o
r
a
u
t
o
n
o
m
o
u
s
d
r
i
v
i
n
g
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
3
,
n
o
.
1
7
,
S
e
p
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
3
1
7
7
5
9
6
.
[
1
8
]
A
.
A
mm
a
r
,
A
.
K
o
u
b
a
a
,
M
.
A
h
me
d
,
a
n
d
A
.
S
a
a
d
,
“
A
e
r
i
a
l
i
ma
g
e
s
p
r
o
c
e
ss
i
n
g
f
o
r
c
a
r
d
e
t
e
c
t
i
o
n
u
si
n
g
C
N
N
s
:
c
o
m
p
a
r
i
s
o
n
b
e
t
w
e
e
n
F
a
st
e
r
R
-
C
N
N
a
n
d
Y
O
LO
v
3
,
”
T
e
c
h
n
i
c
a
l
R
e
p
o
rt
C
I
S
T
ER
-
TR
-
2
0
0
1
0
7
,
p
p
.
1
–
1
2
,
2
0
1
9
.
[
1
9
]
S
.
S
r
i
v
a
s
t
a
v
a
,
A
.
V
.
D
i
v
e
k
a
r
,
C
.
A
n
i
l
k
u
m
a
r
,
I
.
N
a
i
k
,
V
.
K
u
l
k
a
r
n
i
,
a
n
d
V
.
P
a
t
t
a
b
i
r
a
m
a
n
,
“
C
o
mp
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
d
e
e
p
l
e
a
r
n
i
n
g
i
ma
g
e
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
ms
,
”
J
o
u
r
n
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
8
,
n
o
.
1
,
D
e
c
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
021
-
0
0
4
3
4
-
w.
[
2
0
]
S
.
M
.
A
l
k
e
n
t
a
r
,
B
.
A
l
sa
h
w
a
,
A
.
A
ssa
l
e
m,
a
n
d
D
.
K
a
r
a
k
o
l
l
a
,
“
P
r
a
c
t
i
c
a
l
c
o
m
p
a
r
a
t
i
o
n
o
f
t
h
e
a
c
c
u
r
a
c
y
a
n
d
s
p
e
e
d
o
f
Y
O
LO
,
S
S
D
a
n
d
F
a
st
e
r
R
C
N
N
f
o
r
d
r
o
n
e
d
e
t
e
c
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
2
7
,
n
o
.
8
,
p
p
.
1
9
–
3
1
,
A
u
g
.
2
0
2
1
,
d
o
i
:
1
0
.
3
1
0
2
6
/
j
.
e
n
g
.
2
0
2
1
.
0
8
.
0
2
.
[
2
1
]
D
.
D
.
A
b
o
y
o
m
i
a
n
d
C
.
D
a
n
i
e
l
,
“
A
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
si
s
o
f
m
o
d
e
r
n
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
ms:
Y
O
LO
v
s
S
S
D
v
s
F
a
st
e
r
R
-
C
N
N
,
”
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
En
g
i
n
e
e
ri
n
g
J
o
u
r
n
a
l
s
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
9
6
–
1
0
6
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
2
4
2
3
5
/
i
t
e
j
.
v
8
i
2
.
1
2
3
.
[
2
2
]
W
.
L
i
u
e
t
a
l
.
,
“
S
S
D
:
s
i
n
g
l
e
s
h
o
t
m
u
l
t
i
b
o
x
d
e
t
e
c
t
o
r
,
”
i
n
C
o
m
p
u
t
e
r
V
i
si
o
n
–
E
C
C
V
2
0
1
6
(
EC
C
V
2
0
1
6
)
,
C
h
a
m
,
S
w
i
t
z
e
r
l
a
n
d
:
S
p
r
i
n
g
e
r
,
2
0
1
6
,
p
p
.
2
1
–
3
7
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
3
1
9
-
4
6
4
4
8
-
0
_
2
.
[
2
3
]
T.
-
Y
.
L
i
n
,
P
.
G
o
y
a
l
,
R
.
G
i
r
sh
i
c
k
,
K
.
H
e
,
a
n
d
P
.
D
o
l
l
a
r
,
“
F
o
c
a
l
l
o
ss
f
o
r
d
e
n
s
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
An
a
l
y
si
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
4
2
,
n
o
.
2
,
p
p
.
3
1
8
–
3
2
7
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TPA
M
I
.
2
0
1
8
.
2
8
5
8
8
2
6
.
[
2
4
]
A
.
G
e
i
g
e
r
,
P
.
Le
n
z
,
C
.
S
t
i
l
l
e
r
,
a
n
d
R
.
U
r
t
a
su
n
,
“
V
i
si
o
n
me
e
t
s
r
o
b
o
t
i
c
s
:
T
h
e
K
I
TTI
d
a
t
a
se
t
,
”
T
h
e
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
R
o
b
o
t
i
c
s
Re
se
a
rc
h
,
v
o
l
.
3
2
,
n
o
.
1
1
,
p
p
.
1
2
3
1
–
1
2
3
7
,
S
e
p
.
2
0
1
3
,
d
o
i
:
1
0
.
1
1
7
7
/
0
2
7
8
3
6
4
9
1
3
4
9
1
2
9
7
.
[
2
5
]
Z.
C
a
i
a
n
d
N
.
V
a
s
c
o
n
c
e
l
o
s,
“
C
a
s
c
a
d
e
R
-
C
N
N
:
d
e
l
v
i
n
g
i
n
t
o
h
i
g
h
q
u
a
l
i
t
y
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
,
”
i
n
2
0
1
8
I
E
EE/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
8
,
p
p
.
6
1
5
4
–
6
1
6
2
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
8
.
0
0
6
4
4
.
[
2
6
]
A
.
K
l
e
me
n
k
o
,
“
k
i
t
t
i
d
a
t
a
se
t
:
k
i
t
t
i
d
a
t
a
set
2
0
1
2
/
2
0
1
5
st
e
r
e
o
i
ma
g
e
s
f
r
o
m
c
a
mera,
”
k
a
g
g
l
e
.
c
o
m
.
2
0
2
0
.
A
c
c
e
sse
d
:
J
u
n
.
1
8
,
2
0
2
4
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
set
s/
k
l
e
me
n
k
o
/
k
i
t
t
i
-
d
a
t
a
s
e
t
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ma
d
h
u
r
a
M.
B
h
o
sa
le
re
c
e
iv
e
d
h
e
r
M
.
Tec
h
.
d
e
g
re
e
in
sig
n
a
l
p
ro
c
e
ss
in
g
a
n
d
is
c
u
rre
n
tl
y
p
u
rsu
in
g
a
P
h
.
D.
fro
m
S
a
v
it
rib
a
i
P
h
u
le
P
u
n
e
Un
i
v
e
rsity
,
P
u
n
e
,
In
d
ia,
in
t
h
e
d
o
m
a
in
o
f
a
rti
ficia
l
in
tell
ig
e
n
c
e
(AI)
a
n
d
m
a
c
h
in
e
lea
rn
in
g
.
He
r
d
o
c
to
ra
l
re
se
a
rc
h
fo
c
u
se
s
o
n
o
b
jec
t
d
e
tec
ti
o
n
a
n
d
trac
k
i
n
g
f
ro
m
i
m
a
g
e
a
n
d
v
i
d
e
o
d
a
ta,
with
a
p
p
li
c
a
ti
o
n
s
in
in
tell
ig
e
n
t
tran
sp
o
rtati
o
n
sy
ste
m
s.
S
h
e
is
a
n
R&
D
e
n
g
i
n
e
e
r
with
o
v
e
r
5
+
y
e
a
rs
o
f
p
r
o
fe
ss
io
n
a
l
e
x
p
e
rien
c
e
in
AI
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
c
o
m
p
u
ter
v
isi
o
n
.
He
r
wo
rk
p
rima
ril
y
in
v
o
l
v
e
s
th
e
d
e
v
e
lo
p
m
e
n
t
o
f
AI
-
b
a
se
d
v
i
d
e
o
a
n
a
ly
ti
c
s
s
o
lu
t
io
n
s,
in
c
l
u
d
i
n
g
a
u
t
o
m
a
ti
c
n
u
m
b
e
r
p
late
re
c
o
g
n
i
ti
o
n
(AN
P
R),
v
id
e
o
i
n
c
id
e
n
t
d
e
tec
ti
o
n
s
y
ste
m
s
(VID
S
),
v
e
h
icle
a
c
tu
a
ted
sp
e
e
d
d
e
tec
ti
o
n
(VA
S
D),
a
n
d
P
TZ
-
b
a
se
d
a
n
a
ly
ti
c
s.
S
h
e
h
a
s
e
x
ten
si
v
e
e
x
p
e
rien
c
e
in
train
i
n
g
d
e
e
p
lea
r
n
in
g
m
o
d
e
ls
su
c
h
a
s
YO
LO
v
a
rian
ts,
wo
r
k
in
g
with
OCR
fr
a
m
e
wo
rk
s,
a
n
d
d
e
p
l
o
y
i
n
g
s
o
lu
t
i
o
n
s
o
n
G
P
U
a
n
d
e
d
g
e
AI
p
latfo
rm
s.
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
e
e
p
lea
rn
in
g
,
c
o
m
p
u
ter
v
isi
o
n
,
sig
n
a
l
a
n
d
ima
g
e
p
ro
c
e
ss
in
g
,
in
telli
g
e
n
t
v
id
e
o
a
n
a
ly
ti
c
s,
a
n
d
e
m
b
e
d
d
e
d
AI
sy
ste
m
s.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
m
a
d
h
u
ra
b
h
o
sa
le4
@g
m
a
il
.
c
o
m
.
Yo
g
e
sh
S
.
An
g
a
l
is
h
e
a
d
t
h
e
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ics
a
n
d
T
e
lec
o
m
m
u
n
ica
ti
o
n
En
g
i
n
e
e
rin
g
,
JSP
M
'
s
Bh
iv
a
ra
b
a
i
S
a
wa
n
t
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
a
n
d
Re
se
a
rc
h
,
P
u
n
e
,
In
d
ia
.
He
h
a
s
o
v
e
r
2
0
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rie
n
c
e
a
n
d
h
a
s
m
a
d
e
sig
n
ifi
c
a
n
t
c
o
n
tri
b
u
t
io
n
s
t
o
a
c
a
d
e
m
ics
a
n
d
re
se
a
rc
h
.
He
h
a
s
p
u
b
li
s
h
e
d
se
v
e
ra
l
re
se
a
rc
h
p
a
p
e
rs
in
re
p
u
ted
n
a
ti
o
n
a
l
a
n
d
in
tern
a
ti
o
n
a
l
j
o
u
r
n
a
ls
a
n
d
c
o
n
fe
re
n
c
e
s.
His
a
re
a
s
o
f
in
tere
st
in
c
lu
d
e
e
lec
tro
n
ics
a
n
d
c
o
m
m
u
n
ica
ti
o
n
e
n
g
in
e
e
rin
g
d
o
m
a
i
ns
.
H
e
is
a
c
ti
v
e
ly
i
n
v
o
lv
e
d
i
n
g
u
id
i
n
g
stu
d
e
n
ts
a
n
d
fa
c
u
lt
y
to
wa
rd
re
se
a
rc
h
,
in
n
o
v
a
ti
o
n
,
a
n
d
a
c
a
d
e
m
ic
e
x
c
e
ll
e
n
c
e
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
y
sa
n
g
a
l_
e
n
tc@
jsp
m
b
sio
tr.
e
d
u
.
i
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.