I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
,
p
p
.
4
9
7
1
~
4
9
8
1
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
14
.i
6
.
p
p
4
9
7
1
-
4
9
8
1
4971
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Perf
o
rma
nce
eva
lua
tion o
f
pre
-
trai
ned dee
p learning
mo
del on
g
a
rba
g
e clas
sifica
tion wit
h data
au
g
menta
tion a
ppro
a
ch
I
K
o
m
a
ng
Ary
a
G
a
nd
a
Wig
un
a
1
,2
,
I
G
us
t
i Ma
de
Ng
ura
h
Desna
nja
y
a
3
,
I
K
a
dek
B
ud
i Sa
nd
ik
a
2
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
c
s
,
F
a
c
u
l
t
y
o
f
M
a
t
h
e
ma
t
i
c
s
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s
,
U
d
a
y
a
n
a
U
n
i
v
e
r
s
i
t
y
,
B
a
l
i
,
I
n
d
o
n
e
s
i
a
2
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
c
s
,
I
n
st
i
t
u
t
e
o
f
B
u
s
i
n
e
ss a
n
d
T
e
c
h
n
o
l
o
g
y
I
n
d
o
n
e
si
a
,
D
e
n
p
a
s
a
r
,
I
n
d
o
n
e
si
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
y
st
e
m
En
g
i
n
e
e
r
i
n
g
,
I
n
st
i
t
u
t
e
o
f
B
u
si
n
e
ss
a
n
d
T
e
c
h
n
o
l
o
g
y
I
n
d
o
n
e
si
a
,
D
e
n
p
a
s
a
r
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
1
0
,
2
0
2
4
R
ev
is
ed
Au
g
2
3
,
2
0
2
5
Acc
ep
ted
Sep
7
,
2
0
2
5
Was
te
c
las
sifica
ti
o
n
is
o
n
e
o
f
t
h
e
in
tere
stin
g
to
p
ics
f
o
r
c
las
sifi
c
a
ti
o
n
s
i
n
wh
ich
d
a
ta
c
a
n
b
e
v
e
ry
v
a
ried
a
n
d
c
o
m
p
lex
.
T
h
is
d
a
ta
d
i
v
e
rsity
is
a
c
h
a
ll
e
n
g
e
to
d
e
v
e
l
o
p
a
m
o
d
e
l
th
a
t
is ab
le t
o
c
las
sify
we
ll
.
T
h
e
p
u
rp
o
se
o
f
th
is
stu
d
y
is
t
o
a
n
a
ly
z
e
th
e
p
e
rfo
rm
a
n
c
e
o
f
th
e
p
re
-
train
e
d
d
e
e
p
lea
rn
in
g
m
o
d
e
l
u
sin
g
a
d
a
ta
a
u
g
m
e
n
tatio
n
a
p
p
r
o
a
c
h
.
Th
e
re
a
re
th
re
e
p
re
-
train
i
n
g
m
o
d
e
ls
u
se
d
in
th
is
stu
d
y
,
n
a
m
e
ly
r
e
sid
u
a
l
n
e
two
r
k
s
50
(
Re
sN
e
t5
0
)
,
v
is
u
a
l
g
e
o
m
e
tri
c
g
ro
u
p
with
1
6
la
y
e
rs
(
VGG
-
16
)
,
a
n
d
M
o
b
il
e
Ne
tV2
.
T
h
e
re
su
lt
s
sh
o
we
d
t
h
a
t
th
e
M
o
b
i
leN
e
t
V
2
m
o
d
e
l
re
c
e
iv
e
d
th
e
h
ig
h
e
st
a
c
c
u
ra
c
y
v
a
lu
e
,
re
a
c
h
in
g
8
4
.
4
5
%
fo
r
d
a
ta
with
o
u
t
a
u
g
m
e
n
tati
o
n
.
Wi
t
h
d
a
ta
a
u
g
m
e
n
tatio
n
th
e
re
is
a
d
e
c
re
a
s
e
o
f
2
.
7
3
%
.
Co
n
v
e
rse
ly
,
VG
G
-
1
6
sh
o
ws
p
e
rfo
rm
a
n
c
e
sta
b
il
it
y
with
a
n
i
n
c
re
a
se
in
a
c
c
u
ra
c
y
with
a
u
g
m
e
n
tati
o
n
d
a
ta,
re
a
c
h
in
g
7
5
.
8
4
%
.
W
h
il
e
Re
sN
e
t5
0
g
e
ts
t
h
e
lo
we
st
re
su
lt
s
c
o
m
p
a
re
d
to
b
o
t
h
m
o
d
e
ls.
Th
e
a
p
p
li
c
a
ti
o
n
o
f
d
a
ta
a
u
g
m
e
n
t
a
ti
o
n
tec
h
n
iq
u
e
s
with
th
e
a
im
o
f
in
c
re
a
sin
g
d
a
ta
v
a
riati
o
n
s
d
o
e
s
n
o
t
a
lwa
y
s
h
a
v
e
a
n
imp
a
c
t
o
n
in
c
re
a
sin
g
th
e
g
e
n
e
ra
li
z
a
ti
o
n
o
f
t
h
e
m
o
d
e
l
.
K
ey
w
o
r
d
s
:
C
las
s
if
icatio
n
Data
au
g
m
en
tatio
n
Dee
p
lear
n
in
g
Gar
b
ag
e
Pre
-
tr
ain
ed
m
o
d
el
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
Ko
m
an
g
Ar
y
a
Gan
d
a
W
ig
u
n
a
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics,
Facu
lty
o
f
Ma
th
e
m
atics a
n
d
Natu
r
al
Scien
ce
s
,
Ud
ay
an
a
U
n
iv
e
r
s
ity
B
ali,
I
n
d
o
n
esia
E
m
ail:
k
m
ar
y
ag
w
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
W
aste
m
an
ag
em
en
t
h
as
b
ec
o
m
e
an
in
cr
ea
s
in
g
ly
p
r
ess
in
g
g
lo
b
al
is
s
u
e,
with
s
ig
n
if
ican
t
im
p
ac
ts
o
n
th
e
en
v
ir
o
n
m
en
t,
h
u
m
an
h
e
alth
,
an
d
th
e
ec
o
n
o
m
y
.
Acc
o
r
d
in
g
t
o
a
W
o
r
ld
B
an
k
r
e
p
o
r
t,
g
lo
b
al
waste
p
r
o
d
u
ctio
n
r
ea
c
h
es
m
o
r
e
th
an
2
b
illi
o
n
to
n
s
p
er
y
ea
r
[
1
]
a
n
d
co
n
tin
u
es
to
in
cr
ea
s
e
alo
n
g
with
u
r
b
an
izatio
n
an
d
im
p
r
o
v
ed
liv
i
n
g
s
tan
d
a
r
d
s
.
T
h
is
ac
cu
m
u
latio
n
o
f
waste
ca
u
s
es
v
ar
io
u
s
en
v
ir
o
n
m
en
tal
p
r
o
b
lem
s
,
in
clu
d
in
g
wate
r
,
s
o
il,
an
d
ai
r
p
o
llu
tio
n
[
2
]
.
I
n
ad
d
itio
n
,
th
e
co
s
t
o
f
waste
m
an
ag
em
e
n
t,
wh
ich
in
clu
d
es
co
llectio
n
,
tr
an
s
p
o
r
tatio
n
,
a
n
d
f
in
al
d
is
p
o
s
al,
is
a
h
ea
v
y
b
u
r
d
en
f
o
r
m
a
n
y
co
u
n
tr
ies,
esp
ec
ially
d
ev
elo
p
in
g
co
u
n
tr
ies with
in
ad
e
q
u
ate
waste
m
an
ag
em
en
t i
n
f
r
astru
ctu
r
e.
O
n
e
o
f
t
h
e
m
a
i
n
s
o
l
u
t
i
o
n
s
p
r
o
p
o
s
e
d
t
o
r
e
d
u
c
e
t
h
e
n
e
g
a
t
i
v
e
i
m
p
a
c
t
o
f
w
a
s
t
e
is
e
f
f
e
c
t
i
v
e
s
eg
r
e
g
a
t
i
o
n
.
T
h
i
s
s
o
r
t
i
n
g
s
e
p
a
r
a
te
s
w
as
t
e
i
n
t
o
c
a
t
e
g
o
r
i
es
o
f
o
r
g
a
n
i
c
,
i
n
o
r
g
an
i
c
,
a
n
d
r
e
c
y
cl
a
b
l
e
m
a
t
e
r
i
al
s
[
3
]
,
t
h
e
r
e
b
y
r
e
d
u
c
i
n
g
t
h
e
a
m
o
u
n
t
o
f
w
as
t
e
e
n
t
e
r
i
n
g
la
n
d
f
i
l
l
s
a
n
d
i
n
c
r
e
as
i
n
g
t
h
e
e
f
f
ic
i
e
n
c
y
o
f
t
h
e
r
e
c
y
cl
i
n
g
p
r
o
c
e
s
s
.
H
o
w
e
v
e
r
,
m
a
n
u
a
l
s
o
r
t
i
n
g
r
e
q
u
i
r
e
s
h
i
g
h
a
w
a
r
e
n
ess
f
r
o
m
t
h
e
c
o
m
m
u
n
i
t
y
a
n
d
r
e
q
u
i
r
e
s
s
i
g
n
i
f
i
c
a
n
t
l
a
b
o
r
a
n
d
t
i
m
e.
T
o
o
v
er
co
m
e
th
is
o
b
s
tacle
,
te
ch
n
o
lo
g
y
p
lay
s
an
im
p
o
r
tan
t
r
o
le
in
d
ev
elo
p
in
g
a
m
o
r
e
ef
f
ic
ien
t
waste
m
an
ag
em
en
t
s
y
s
tem
.
On
e
p
r
o
m
is
in
g
s
o
lu
tio
n
is
th
e
d
ev
elo
p
m
en
t
o
f
s
m
ar
t
waste
b
in
s
eq
u
ip
p
ed
with
an
au
to
m
atic
s
o
r
tin
g
s
y
s
tem
b
ase
d
o
n
m
icr
o
c
o
n
tr
o
ller
tech
n
o
lo
g
y
[
4
]
.
T
h
is
s
m
ar
t
tr
ash
c
an
u
s
es
s
en
s
o
r
s
s
u
ch
as
p
r
o
x
im
ity
an
d
in
f
r
ar
ed
to
au
t
o
m
atica
lly
d
etec
t
t
h
e
ty
p
e
o
f
waste
[
5
]
.
W
h
ile
e
f
f
ec
tiv
e
in
b
asic
s
o
r
tin
g
,
th
ese
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
9
7
1
-
4
9
8
1
4972
s
y
s
tem
s
h
av
e
lim
itatio
n
s
wh
en
d
ea
lin
g
with
m
o
r
e
co
m
p
le
x
ty
p
es
o
f
waste,
wh
ich
o
f
te
n
lead
s
to
er
r
o
r
s
in
class
if
icatio
n
.
T
o
im
p
r
o
v
e
s
o
r
tin
g
ac
cu
r
ac
y
,
th
e
u
s
e
o
f
d
ig
ital
im
a
g
es
as
ad
d
itio
n
al
in
p
u
t
is
p
r
o
p
o
s
ed
.
C
o
m
p
u
ter
v
is
io
n
an
d
d
ee
p
lear
n
in
g
tech
n
o
lo
g
ies
o
f
f
er
s
o
lu
tio
n
s
to
i
m
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
waste
ty
p
e
class
if
icatio
n
[
6
]
.
C
o
m
p
u
ter
v
is
io
n
allo
ws
th
e
s
y
s
tem
to
p
r
o
ce
s
s
litt
er
im
ag
es
v
is
u
ally
,
wh
ile
d
ee
p
lear
n
in
g
is
u
s
ed
to
p
er
f
o
r
m
litt
er
d
etec
tio
n
an
d
class
if
icati
o
n
with
a
h
ig
h
d
e
g
r
ee
o
f
ac
c
u
r
ac
y
[
7
]
.
T
h
is
ap
p
r
o
ac
h
is
ex
p
ec
ted
to
o
v
er
co
m
e
s
en
s
o
r
lim
itatio
n
s
an
d
p
r
o
d
u
c
e
a
m
o
r
e
e
f
f
ec
tiv
e
an
d
ef
f
ici
en
t
litt
er
class
if
icatio
n
s
y
s
tem
[
8
]
.
Ho
we
v
er
,
th
e
m
ain
p
r
o
b
lem
i
n
im
ag
e
-
b
ased
litt
er
class
if
icatio
n
is
th
e
d
iv
er
s
ity
o
f
litt
er
s
h
a
p
es,
co
lo
r
s
,
an
d
co
n
d
itio
n
s
th
at
o
f
ten
d
if
f
e
r
f
r
o
m
o
n
e
im
ag
e
t
o
an
o
th
er
.
T
h
ese
v
ar
iatio
n
s
ca
n
m
ak
e
it
d
if
f
icu
lt
f
o
r
d
ee
p
lear
n
in
g
m
o
d
els
to
g
en
er
alize
th
e
d
ata,
esp
ec
ially
wh
en
f
ac
ed
with
n
ew
d
ata
o
r
en
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
th
a
t
ar
e
d
if
f
er
en
t
f
r
o
m
th
e
tr
ain
in
g
d
ata.
As
o
n
e
o
f
th
e
wid
ely
u
s
ed
a
p
p
r
o
ac
h
es
,
tr
an
s
f
er
lear
n
in
g
u
s
in
g
p
r
e
-
tr
ain
ed
d
ee
p
lear
n
i
n
g
m
o
d
els
p
r
o
v
id
es
s
ig
n
if
ican
t
ad
v
an
ta
g
es
in
ac
cu
r
ac
y
an
d
tr
ai
n
in
g
tim
e
ef
f
icien
cy
.
Pre
-
tr
ain
e
d
m
o
d
els
s
u
ch
as
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
[
9
]
,
v
is
u
al
g
eo
m
etr
ic
g
r
o
u
p
with
1
6
lay
e
r
s
(
VGG
-
16
)
[
1
0
]
,
r
esid
u
al
n
etwo
r
k
s
5
0
(
R
esNet5
0
)
[
1
1
]
,
an
d
Mo
b
ileNetV2
[
1
2
]
ar
e
ab
le
to
u
tili
ze
r
ich
an
d
ad
v
an
ce
d
v
is
u
al
r
ep
r
esen
tatio
n
f
ea
tu
r
es.
T
h
e
r
esu
ltin
g
ac
cu
r
ac
y
p
er
f
o
r
m
a
n
ce
o
f
th
ese
m
o
d
els
r
ea
ch
es
8
1
to
9
5
%
[
1
3
]
.
Ho
wev
er
,
ch
allen
g
es
in
m
o
d
e
l
g
en
er
aliza
tio
n
r
e
m
ain
,
esp
ec
ially
wh
en
th
e
m
o
d
el
is
u
s
ed
f
o
r
class
if
icatio
n
o
f
out
-
of
-
d
is
tr
ib
u
tio
n
d
ata,
i.e
.
d
a
ta
th
at
h
as a
d
if
f
er
e
n
t d
is
tr
ib
u
t
io
n
th
an
t
h
e
d
ata
tr
ain
e
d
o
n
[
1
4
]
.
On
e
ap
p
r
o
ac
h
th
at
ca
n
im
p
r
o
v
e
th
e
g
en
e
r
aliza
tio
n
ab
ili
ty
o
f
t
h
e
m
o
d
el
is
to
p
e
r
f
o
r
m
d
ata
au
g
m
en
tatio
n
.
Data
au
g
m
e
n
tatio
n
aim
s
to
ex
p
a
n
d
th
e
d
is
tr
ib
u
tio
n
o
f
tr
ain
in
g
d
ata
th
r
o
u
g
h
s
y
n
th
etic
v
ar
iatio
n
s
s
u
ch
as
r
o
tatio
n
,
f
lip
p
in
g
,
z
o
o
m
in
g
,
an
d
o
t
h
er
tr
an
s
f
o
r
m
atio
n
s
,
s
o
th
at
th
e
m
o
d
el
b
ec
o
m
e
s
m
o
r
e
ad
ap
tiv
e
to
d
if
f
er
en
t
im
ag
e
v
ar
iatio
n
s
[
9
]
.
Alth
o
u
g
h
d
ata
a
u
g
m
e
n
tatio
n
h
as
p
r
o
v
en
to
b
e
e
f
f
ec
tiv
e
in
s
o
m
e
c
ases
,
th
e
ef
f
ec
t
o
f
d
ata
au
g
m
e
n
tatio
n
o
n
g
en
er
aliza
tio
n
ab
ilit
y
f
o
r
v
ar
io
u
s
d
ee
p
lear
n
in
g
m
o
d
el
ar
ch
itectu
r
es
is
s
till
n
o
t
f
u
lly
u
n
d
er
s
to
o
d
[
1
5
]
.
T
h
is
r
esear
ch
f
o
c
u
s
es
o
n
d
ev
elo
p
in
g
a
waste
class
if
icatio
n
m
o
d
el
with
a
p
r
e
-
tr
ain
e
d
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
an
d
in
co
r
p
o
r
atin
g
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
to
im
p
r
o
v
e
g
e
n
er
aliza
tio
n
ca
p
ab
ilit
ies
[
1
6
]
.
T
h
e
m
ain
o
b
jectiv
e
is
to
ev
alu
ate
an
d
u
n
d
er
s
tan
d
th
e
ef
f
ec
t
o
f
a
u
g
m
en
tatio
n
tec
h
n
iq
u
es
o
n
v
a
r
io
u
s
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es,
n
am
el
y
R
esNet5
0
,
VGG
-
1
6
,
a
n
d
M
o
b
ileNetV2
,
in
t
h
e
f
ac
e
o
f
n
ew
d
ata
t
h
at
h
as
n
e
v
er
b
ee
n
s
ee
n
b
y
th
e
m
o
d
el
.
T
h
is
ap
p
r
o
ac
h
is
ex
p
ec
ted
to
c
o
n
tr
ib
u
t
e
to
th
e
v
er
if
icatio
n
o
f
m
o
d
e
l
g
en
er
aliza
tio
n
b
y
u
s
in
g
an
ex
te
r
n
al
d
ataset
th
at
i
s
co
m
p
letely
d
if
f
e
r
en
t f
r
o
m
th
e
tr
ain
in
g
d
ata
as a
v
alid
ity
test
.
2.
M
E
T
H
O
D
T
h
i
s
r
e
s
e
ar
c
h
p
r
o
p
o
s
e
s
th
e
co
n
s
t
r
u
c
t
i
o
n
o
f
a
w
a
s
t
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
m
o
d
e
l
u
s
i
n
g
t
h
r
e
e
a
r
ch
i
t
e
c
t
u
r
e
s
,
n
a
m
e
ly
R
e
s
N
e
t
5
0
,
V
G
G
-
1
6
,
a
n
d
M
o
b
i
l
e
N
e
t
V2
.
T
h
e
r
e
s
e
ar
c
h
s
t
a
g
e
c
o
n
s
i
s
t
s
o
f
s
ev
e
r
a
l
s
t
e
p
s
,
s
t
a
r
t
in
g
w
i
t
h
d
a
t
a
c
o
l
l
ec
t
i
o
n
(
d
a
ta
s
e
t
)
,
th
en
d
i
v
id
i
n
g
th
e
d
a
t
a
in
t
o
th
r
e
e
p
ar
t
s
:
tr
a
i
n
in
g
d
a
t
a
,
v
a
l
i
d
a
t
io
n
d
a
ta
,
a
n
d
t
e
s
t
i
n
g
d
a
t
a
[
1
7
]
,
[
1
8
]
.
T
h
e
t
r
a
in
i
n
g
d
a
ta
w
i
l
l
g
o
th
r
o
u
g
h
an
a
u
g
m
e
n
t
a
t
io
n
p
r
o
ce
s
s
t
o
en
r
i
ch
t
h
e
d
a
t
a
v
a
r
i
a
t
io
n
.
A
f
t
e
r
th
a
t
,
th
e
au
g
m
e
n
t
ed
d
a
ta
w
i
l
l
b
e
u
s
e
d
to
t
r
a
in
t
h
e
d
ee
p
l
e
a
r
n
i
n
g
m
o
d
e
l,
w
h
i
le
t
h
e
v
a
l
i
d
a
t
i
o
n
d
a
t
a
w
i
l
l
b
e
d
i
r
e
c
t
ly
u
s
e
d
i
n
t
h
e
tr
a
i
n
in
g
p
r
o
c
e
s
s
.
O
n
c
e
t
h
e
tr
a
i
n
in
g
i
s
co
m
p
l
e
t
e
,
t
h
e
m
o
d
e
l
w
il
l
b
e
e
v
a
l
u
a
t
ed
to
d
e
t
e
r
m
i
n
e
th
e
b
e
s
t
p
er
f
o
r
m
an
c
e
.
F
i
g
u
r
e
1
p
r
o
v
i
d
e
s
a
g
e
n
e
r
a
l
i
l
l
u
s
t
r
a
t
io
n
o
f
th
e
r
e
s
e
ar
ch
s
t
a
g
e
s
i
n
b
u
i
l
d
in
g
a
w
a
s
t
e
c
l
a
s
s
i
f
i
c
a
t
io
n
m
o
d
e
l
.
Fig
u
r
e
1
.
R
esear
ch
s
tag
es
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
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f p
r
e
-
tr
a
in
ed
d
ee
p
lea
r
n
in
g
mo
d
el
o
n
…
(
I
K
o
ma
n
g
A
r
ya
Ga
n
d
a
Wig
u
n
a
)
4973
2
.
1
.
Da
t
a
s
et
T
h
e
d
ataset
u
s
ed
in
th
e
r
esear
ch
is
an
o
p
e
n
d
ataset
f
r
o
m
th
e
Kag
g
le
we
b
s
ite.
R
ea
lW
aste
d
ataset
is
a
co
llectio
n
o
f
waste
class
if
icat
io
n
d
ata
g
r
o
u
p
e
d
in
to
9
ty
p
es
wh
er
e
th
e
d
ata
is
co
llected
f
r
o
m
lan
d
f
ills
[
1
9
]
.
Fig
u
r
e
2
s
h
o
ws
a
d
ataset
s
am
p
le,
p
r
o
v
id
in
g
a
v
is
u
al
r
ep
r
esen
tatio
n
o
f
t
h
e
v
ar
i
o
u
s
ty
p
es
o
f
waste
in
clu
d
ed
.
E
ac
h
im
ag
e
h
as
a
r
eso
lu
tio
n
o
f
2
2
4
×
2
2
4
in
o
r
d
er
to
d
is
tin
g
u
is
h
o
b
ject
f
ea
tu
r
es
m
o
r
e
cl
ea
r
ly
an
d
g
et
b
etter
class
if
icatio
n
ac
cu
r
ac
y
[
2
0
]
.
T
h
e
d
ata
is
d
iv
id
ed
in
to
3
p
a
r
ts
,
8
0
%
f
o
r
tr
ain
in
g
d
ata,
1
0
%
f
o
r
v
alid
atio
n
d
ata,
an
d
1
0
% a
s
test
in
g
d
ata
[
2
1
]
.
T
h
e
class
lab
els an
d
n
u
m
b
er
o
f
d
atasets
ar
e
s
h
o
wn
in
T
a
b
le
1
.
Fig
u
r
e
2
.
E
x
am
p
le
d
ataset
T
ab
le
1
.
L
a
b
els an
d
n
u
m
b
er
o
f
d
atasets
La
b
e
l
To
t
a
l
d
a
t
a
C
a
r
d
b
o
a
r
d
4
6
1
F
o
o
d
o
r
g
a
n
i
c
s
4
1
1
G
l
a
ss
4
2
0
M
e
t
a
l
7
9
0
M
i
s
c
e
l
l
a
n
e
o
u
s
t
r
a
s
h
4
9
5
P
a
p
e
r
5
0
0
P
l
a
st
i
c
9
2
1
Te
x
t
i
l
e
t
r
a
s
h
3
1
8
V
e
g
e
t
a
t
i
o
n
4
3
6
2
.
2
.
Aug
m
ent
a
t
io
n
da
t
a
Au
g
m
en
tatio
n
m
eth
o
d
s
ar
e
ap
p
lied
to
th
e
tr
ain
in
g
d
ataset
with
th
e
aim
o
f
g
ettin
g
b
etter
p
e
r
f
o
r
m
an
ce
.
T
h
e
au
g
m
e
n
tatio
n
m
et
h
o
d
s
a
p
p
lied
ar
e
f
lip
,
r
o
tatio
n
,
an
d
zo
o
m
[
2
2
]
.
T
ab
le
2
s
h
o
ws
th
e
d
etails
o
f
th
e
p
ar
am
eter
s
u
s
ed
.
T
h
e
d
eter
m
i
n
atio
n
o
f
au
g
m
en
tatio
n
tech
n
i
q
u
es is
d
o
n
e
r
an
d
o
m
ly
.
E
ac
h
au
g
m
en
tatio
n
r
esu
lts
in
d
if
f
er
e
n
t d
ata
v
a
r
iatio
n
s
f
o
r
ea
ch
s
am
p
le,
allo
win
g
th
e
m
o
d
el
to
s
ee
f
ea
tu
r
es in
v
ar
io
u
s
c
o
n
tex
ts
[
2
3
]
.
T
ab
le
2
.
Au
g
m
en
tatio
n
p
ar
am
eter
s
P
a
r
a
me
t
e
r
s
V
a
l
u
e
R
e
sc
a
l
e
1
.
/
2
5
5
R
o
t
a
t
i
o
n
_
r
a
n
g
e
30
S
h
e
a
r
_
r
a
n
g
e
0
.
1
Zo
o
m_
r
a
n
g
e
0
.
2
H
o
r
i
z
o
n
t
a
l
_
f
l
i
p
Tr
u
e
V
e
r
t
i
c
a
l
_
f
l
i
p
Tr
u
e
W
i
d
t
h
_
s
h
i
f
t
_
r
a
n
g
e
0
.
2
H
e
i
g
h
t
_
s
h
i
f
t
_
r
a
n
g
e
0
.
2
I
n
th
e
co
d
e,
th
e
`
I
m
ag
eDa
t
aGe
n
er
ato
r
`
o
b
ject
is
u
s
ed
f
o
r
im
ag
e
au
g
m
en
tatio
n
wit
h
v
ar
io
u
s
tr
an
s
f
o
r
m
atio
n
s
.
T
h
e
p
ar
am
et
er
`
r
escale=
1
.
/2
5
5
.
`
is
u
s
ed
t
o
n
o
r
m
alize
th
e
im
a
g
e
p
i
x
el
v
alu
es
b
y
d
iv
id
in
g
th
em
b
y
2
5
5
,
s
o
th
at
t
h
ey
ar
e
in
th
e
r
a
n
g
e
[
0
,
1
]
.
`
r
o
tatio
n
_
r
an
g
e=
3
0
`
allo
ws
r
o
tatio
n
o
f
t
h
e
im
ag
e
u
p
to
3
0
d
eg
r
ee
s
r
an
d
o
m
ly
.
`
s
h
ea
r
_
r
a
n
g
e=
0
.
1
`
g
iv
es
th
e
im
ag
e
a
s
h
ea
r
ef
f
ec
t
o
f
1
0
%.
`
zo
o
m
_
r
an
g
e=
0
.
2
`
allo
ws
zo
o
m
in
g
i
n
an
d
o
u
t
o
f
th
e
im
a
g
e
with
in
a
2
0
%
r
an
g
e.
`
h
o
r
iz
o
n
tal_
f
lip
=T
r
u
e`
an
d
`
v
e
r
tical_
f
lip
=T
r
u
e
`
en
ab
le
h
o
r
izo
n
tal
an
d
v
er
tical
f
lip
p
i
n
g
o
f
th
e
im
ag
e.
`
wid
t
h
_
s
h
if
t
_
r
an
g
e=
0
.
2
`
an
d
`
h
eig
h
t_
s
h
if
t
_
r
an
g
e=
0
.
2
`
en
a
b
le
h
o
r
izo
n
tal
a
n
d
v
er
tical
s
h
if
tin
g
o
f
th
e
im
ag
e
u
p
to
2
0
%
o
f
t
h
e
im
ag
e
wid
t
h
o
r
h
eig
h
t
r
esp
ec
tiv
ely
.
All
th
ese
p
ar
am
eter
s
ar
e
d
esig
n
e
d
to
e
n
r
ich
th
e
v
a
r
iety
o
f
tr
ai
n
in
g
d
ata
b
y
r
an
d
o
m
ly
m
o
d
if
y
in
g
th
e
im
ag
es,
th
u
s
im
p
r
o
v
in
g
th
e
g
en
er
aliza
tio
n
o
f
th
e
m
o
d
el.
Fig
u
r
e
3
s
h
o
ws
th
e
r
esu
lt
o
f
th
e
au
g
m
e
n
tatio
n
p
r
o
ce
s
s
ap
p
lied
to
th
e
d
ataset
s
am
p
le
s
h
o
wn
p
r
e
v
io
u
s
ly
in
Fig
u
r
e
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
9
7
1
-
4
9
8
1
4974
Fig
u
r
e
3
.
Data
au
g
m
en
tatio
n
r
esu
lt
2
.
3
.
M
o
bil
eNe
t
V2
Mo
b
ileNetV2
is
a
m
o
d
el
p
r
o
p
o
s
ed
b
y
Go
o
g
le
in
2
0
1
7
[
2
4
]
.
T
h
is
ar
ch
itectu
r
e
was
r
elea
s
ed
in
ea
r
ly
2
0
1
8
as
a
d
ev
elo
p
m
en
t
o
f
th
e
p
r
ev
io
u
s
m
o
d
el,
Mo
b
ileNetV1
,
wh
ich
was
p
r
o
p
o
s
ed
in
2
0
1
5
[
1
2
]
,
[
2
5
]
.
T
h
e
d
ev
elo
p
m
e
n
t
o
f
th
is
m
o
d
el
is
s
ee
n
in
th
e
em
er
g
en
ce
o
f
lin
ea
r
b
o
ttlen
ec
k
s
b
etwe
en
lay
er
s
a
n
d
f
ast
co
n
n
ec
tio
n
s
b
etwe
en
b
o
ttlen
ec
k
s
[
2
5
]
.
B
o
t
tlen
ec
k
s
in
M
o
b
ileNetV2
h
el
p
to
p
ac
k
ag
e
th
e
m
o
d
el
f
r
o
m
lo
w
-
lev
el
co
n
ce
p
ts
to
h
ig
h
-
lev
el
d
escr
ip
to
r
s
s
o
th
at
th
e
m
o
d
el
ca
n
b
e
tr
ain
ed
f
a
s
ter
an
d
p
r
o
d
u
ce
h
ig
h
ac
cu
r
a
cy
ev
en
th
o
u
g
h
th
e
m
o
d
el
u
s
es
f
ewe
r
p
ar
am
eter
s
[
2
6
]
.
Fig
u
r
e
4
illu
s
tr
ates
th
e
o
v
er
all
ar
ch
itectu
r
e
o
f
Mo
b
ile
NetV2
,
h
ig
h
lig
h
tin
g
th
e
u
s
e
o
f
in
v
e
r
ted
r
esid
u
al
b
l
o
ck
s
with
lin
ea
r
b
o
ttlen
ec
k
s
t
h
at
co
n
tr
ib
u
te
to
its
ef
f
icien
cy
an
d
p
er
f
o
r
m
an
ce
in
im
ag
e
class
if
icatio
n
task
s
.
Fig
u
r
e
4
.
Mo
b
ileNetV2
Ar
ch
it
ec
tu
r
e
2
.
4
.
Vis
ua
l g
eo
m
et
ric
g
r
o
up
wit
h 1
6
la
y
er
s
VGG
-
1
6
is
o
n
e
o
f
th
e
C
NN
a
r
ch
itectu
r
es
th
at
h
as
1
6
lay
er
s
in
th
e
f
o
r
m
o
f
1
3
co
n
v
o
l
u
tio
n
al
lay
er
s
an
d
3
f
u
lly
-
co
n
n
ec
ted
lay
er
s
[
2
7
]
,
[
2
8
]
.
T
h
e
ad
v
a
n
tag
e
o
f
th
is
ar
c
h
itectu
r
e
is
t
h
at
it
h
as
an
a
r
ch
itectu
r
e
co
n
s
is
tin
g
o
f
3
×
3
co
n
v
o
lu
tio
n
al
lay
er
s
an
d
2
×
2
p
o
o
lin
g
[
2
9
]
,
[
3
0
]
.
wh
ich
is
co
n
s
id
er
ed
m
o
r
e
ac
cu
r
ate
wh
en
co
m
p
ar
ed
to
p
r
ev
io
u
s
C
NN
ar
ch
itectu
r
es.
T
h
e
ac
cu
r
ac
y
o
f
t
h
is
m
o
d
el
is
tak
en
f
r
o
m
th
e
s
t
ate
-
of
-
th
e
-
a
r
t
C
NN
ar
ch
itectu
r
e,
n
a
m
ely
th
e
b
en
ch
m
ar
k
f
o
r
th
e
ac
c
u
r
ac
y
o
f
th
is
i
m
ag
e
is
tak
en
f
r
o
m
th
e
t
h
r
ee
m
o
d
els
[
3
1
]
.
VGG
-
1
6
also
h
as
g
o
o
d
g
en
e
r
aliza
tio
n
ca
p
ab
ilit
ies
an
d
is
o
f
ten
u
s
ed
as
a
m
o
d
el
b
ase
f
o
r
v
ar
i
o
u
s
im
ag
e
r
ec
o
g
n
itio
n
task
s
.
I
n
ad
d
itio
n
,
th
is
ar
ch
itectu
r
e
is
k
n
o
wn
to
b
e
ea
s
ier
to
o
p
tim
ize
d
u
e
to
its
s
im
p
le
y
et
ef
f
ec
tiv
e
d
esig
n
,
alth
o
u
g
h
it
r
eq
u
ir
es
c
o
n
s
id
er
ab
le
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
[
1
0
]
.
Fig
u
r
e
5
s
h
o
ws
th
e
ar
ch
itectu
r
al
s
tr
u
ctu
r
e
o
f
VGG
-
1
6
,
illu
s
tr
atin
g
its
s
eq
u
en
tial
an
d
u
n
i
f
o
r
m
lay
er
co
n
f
ig
u
r
atio
n
.
Fig
u
r
e
5
.
VGG
-
1
6
ar
ch
itectu
r
e
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
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f p
r
e
-
tr
a
in
ed
d
ee
p
lea
r
n
in
g
mo
d
el
o
n
…
(
I
K
o
ma
n
g
A
r
ya
Ga
n
d
a
Wig
u
n
a
)
4975
2
.
5
.
ResNet
5
0
R
esNet
s
ar
e
d
ee
p
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
wh
o
s
e
b
asic
id
e
a
is
to
b
y
p
ass
th
e
b
lo
ck
p
r
o
c
ess
o
f
th
e
co
n
v
o
l
u
tio
n
al
lay
e
r
b
y
u
s
in
g
s
h
o
r
tcu
t
co
n
n
ec
tio
n
s
.
R
esNet
h
as
a
b
asic
b
lo
c
k
ca
lled
b
o
ttl
en
ec
k
b
l
o
ck
wh
ic
h
h
as
two
r
u
les
in
its
s
im
p
le
d
es
ig
n
[
3
2
]
.
T
h
e
f
ir
s
t
r
u
le
is
th
at
f
o
r
th
e
s
am
e
o
u
tp
u
t
f
ea
tu
r
e
m
ap
s
ize,
ea
ch
lay
e
r
h
as
th
e
s
am
e
n
u
m
b
er
o
f
f
ilter
s
.
T
h
e
s
ec
o
n
d
r
u
le
is
th
at
if
th
e
s
ize
o
f
th
e
o
u
tp
u
t
f
ea
tu
r
e
m
ap
is
d
iv
id
ed
in
to
two
,
th
e
n
u
m
b
er
o
f
f
ilter
s
f
o
r
ea
ch
lay
er
is
d
o
u
b
led
.
Sam
p
li
n
g
is
d
o
n
e
d
i
r
ec
tly
b
y
th
e
co
n
v
o
lu
tio
n
lay
er
w
h
ich
h
as
th
e
s
tep
o
f
co
n
v
er
tin
g
ea
c
h
b
y
tep
lo
t
i
n
to
a
n
R
GB
im
ag
e
an
d
b
atch
n
o
r
m
aliza
tio
n
wh
ic
h
is
d
o
n
e
r
ig
h
t
af
ter
ea
ch
co
n
v
o
lu
tio
n
a
n
d
b
e
f
o
r
e
th
e
R
eL
U
ac
tiv
atio
n
p
r
o
ce
s
s
.
C
o
n
v
er
tin
g
ea
ch
b
y
tep
lo
t
i
n
to
an
R
GB
im
ag
e
r
esizes
it
to
2
2
4
×
2
2
4
d
im
en
s
io
n
s
an
d
s
u
b
tr
ac
ts
th
e
av
er
ag
e
R
GB
d
ataset
d
ata
f
r
o
m
ea
ch
p
ix
el
[
3
3
]
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
ar
ch
itectu
r
e
o
f
R
esNet5
0
,
h
ig
h
lig
h
tin
g
th
e
s
h
o
r
tcu
t
co
n
n
ec
tio
n
s
a
n
d
b
o
t
tlen
ec
k
d
esig
n
th
at
m
ak
e
th
e
n
etwo
r
k
d
ee
p
er
a
n
d
m
o
r
e
ef
f
icien
t.
Fig
u
r
e
6
.
R
esNet5
0
Ar
ch
itectu
r
e
3.
R
E
SU
L
T
S A
N
D
D
ISC
U
S
SIO
N
3
.
1
.
T
ra
ini
ng
re
s
ults
I
n
m
o
d
el
tr
ain
in
g
,
v
ar
io
u
s
im
p
o
r
tan
t
p
a
r
am
eter
s
ar
e
n
e
ed
ed
to
s
u
p
p
o
r
t
th
e
tr
ain
in
g
p
r
o
ce
s
s
o
p
tim
ally
.
T
h
e
d
etailed
co
n
f
i
g
u
r
atio
n
o
f
th
ese
p
ar
am
eter
s
is
s
u
m
m
ar
ized
in
T
a
b
le
3
,
wh
ich
p
r
esen
ts
th
e
v
alu
es
ap
p
lied
u
n
if
o
r
m
ly
ac
r
o
s
s
all
m
o
d
els
u
s
ed
in
th
is
s
tu
d
y
.
As
s
h
o
wn
in
T
a
b
le
4
,
R
esNet5
0
m
o
d
el
ex
h
ib
its
a
tr
ain
lo
s
s
o
f
1
.
3
3
9
7
an
d
a
v
alid
atio
n
lo
s
s
o
f
1
.
4
8
3
3
,
in
d
ic
atin
g
th
at
th
e
m
o
d
el
s
tr
u
g
g
les
to
ef
f
ec
tiv
ely
lear
n
f
r
o
m
th
e
d
ata
d
u
r
in
g
b
o
th
tr
a
in
in
g
an
d
v
alid
atio
n
.
T
h
is
is
f
u
r
th
er
s
u
p
p
o
r
ted
b
y
its
tr
ain
ac
cu
r
ac
y
o
f
0
.
5
1
0
1
an
d
v
ali
d
atio
n
ac
c
u
r
ac
y
o
f
0
.
4
5
6
8
,
s
u
g
g
esti
n
g
th
at
t
h
e
m
o
d
el
is
ex
p
er
ie
n
cin
g
u
n
d
er
f
itti
n
g
,
wh
e
r
e
it
f
ails
to
ca
p
tu
r
e
th
e
u
n
d
er
l
y
in
g
p
atter
n
s
in
th
e
tr
ain
in
g
d
ata
,
th
u
s
lea
d
in
g
to
p
o
o
r
g
en
e
r
aliza
tio
n
.
I
n
co
n
tr
ast,
VGG
-
1
6
,
as
p
r
esen
ted
i
n
T
a
b
le
4
,
d
em
o
n
s
tr
ates
im
p
r
o
v
ed
lea
r
n
in
g
c
ap
ab
ilit
y
with
a
lo
wer
tr
ain
lo
s
s
o
f
0
.
2
6
4
7
an
d
v
alid
atio
n
lo
s
s
o
f
0
.
7
2
1
1
.
I
ts
tr
ain
ac
cu
r
ac
y
o
f
0
.
9
1
4
6
r
ef
l
ec
ts
a
s
tr
o
n
g
ab
ilit
y
to
r
ec
o
g
n
ize
p
atter
n
s
in
th
e
tr
ain
in
g
s
et.
Ho
wev
er
,
th
e
g
ap
b
etwe
en
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
(
v
alid
atio
n
a
cc
u
r
ac
y
o
f
0
.
7
5
7
9
)
in
d
icate
s
a
ten
d
en
cy
to
war
d
o
v
er
f
itti
n
g
,
th
o
u
g
h
t
h
e
m
o
d
el
s
till
m
ain
tain
s
a
r
elativ
ely
g
o
o
d
lev
el
o
f
g
en
er
aliza
tio
n
.
Am
o
n
g
t
h
e
th
r
ee
m
o
d
els
in
T
ab
le
4
,
M
o
b
ile
NetV2
ac
h
iev
es
th
e
b
est
o
v
er
all
p
er
f
o
r
m
an
ce
.
I
t
r
ec
o
r
d
s
a
r
em
a
r
k
ab
l
y
lo
w
tr
ain
lo
s
s
o
f
0
.
0
0
1
5
a
n
d
v
alid
atio
n
lo
s
s
o
f
0
.
7
5
6
1
,
alo
n
g
s
id
e
a
p
e
r
f
ec
t tr
ain
ac
cu
r
ac
y
o
f
1
.
0
0
0
0
.
N
o
tab
ly
,
its
v
alid
a
tio
n
ac
cu
r
ac
y
r
ea
c
h
es
0
.
8
2
3
2
,
s
u
g
g
esti
n
g
e
x
ce
llen
t
g
e
n
er
ali
za
tio
n
ca
p
ab
ilit
ies
an
d
an
ef
f
ec
tiv
e
lear
n
in
g
p
r
o
c
ess
with
o
u
t
an
y
s
ig
n
i
f
ican
t
o
v
er
f
itti
n
g
.
T
h
ese
r
esu
lts
co
n
f
ir
m
th
at
Mo
b
ileNetV2
is
th
e
m
o
s
t e
f
f
icien
t m
o
d
el
i
n
l
ea
r
n
in
g
th
e
d
ata
in
th
e
ab
s
en
c
e
o
f
au
g
m
en
tatio
n
.
T
ab
le
3.
Hy
p
er
p
ar
a
m
eter
s
u
s
ed
f
o
r
all
m
o
d
els
P
a
r
a
me
t
e
r
s
V
a
l
u
e
I
n
p
u
t
i
m
a
g
e
2
2
4
×
2
2
4
O
p
t
i
mi
z
e
r
s
A
d
a
m
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
Ep
o
c
h
1
0
0
B
a
t
c
h
si
z
e
32
A
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
F
C
R
e
L
u
F
u
n
c
t
i
o
n
i
n
t
h
e
mo
d
e
l
A
vg
T
ab
le
4
.
T
r
ai
n
in
g
r
esu
lts
o
f
m
o
d
el
with
o
u
t a
u
g
m
en
tatio
n
D
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
Tr
a
i
n
l
o
ss
V
a
l
i
d
l
o
ss
Tr
a
i
n
a
c
c
u
r
a
c
y
V
a
l
i
d
a
c
c
u
r
a
c
y
R
e
sN
e
t
5
0
1
.
3
3
9
7
1
.
4
8
3
3
0
.
5
1
0
1
0
.
4
5
6
8
VGG
-
16
0
.
2
6
4
7
0
.
7
2
1
1
0
.
9
1
4
6
0
.
7
5
7
9
M
o
b
i
l
e
N
e
t
V
2
0
.
0
0
1
5
0
.
7
5
6
1
1
.
0
0
0
0
0
.
8
2
3
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
9
7
1
-
4
9
8
1
4976
T
ab
le
5
s
h
o
ws
th
e
r
esu
lts
o
f
t
esti
n
g
m
o
d
els
with
d
ata
au
g
m
en
tatio
n
.
I
t
ca
n
b
e
s
ee
n
th
at
n
o
n
e
o
f
th
e
th
r
ee
m
o
d
els
h
a
v
e
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
co
m
p
ar
e
d
to
wit
h
o
u
t
au
g
m
en
tatio
n
.
I
n
R
esNet5
0
,
th
e
tr
ai
n
lo
s
s
o
f
1
.
6
3
4
5
an
d
v
alid
lo
s
s
o
f
1
.
6
2
9
2
s
h
o
w
th
at
ev
en
th
o
u
g
h
th
i
s
m
o
d
el
u
s
es
au
g
m
en
tatio
n
,
its
p
er
f
o
r
m
an
ce
in
u
n
d
er
s
tan
d
i
n
g
p
atter
n
s
f
r
o
m
tr
ain
in
g
d
ata
r
em
ai
n
s
lo
w.
T
h
is
is
also
r
ef
lecte
d
in
th
e
tr
ain
ac
cu
r
ac
y
w
h
ich
o
n
ly
r
ea
ch
es
0
.
3
9
1
6
a
n
d
v
alid
ac
cu
r
ac
y
o
f
0
.
4
3
7
9
.
T
h
e
R
esNet5
0
m
o
d
el
s
till
h
as
d
if
f
icu
lty
i
n
g
en
er
alizin
g
p
atter
n
s
f
r
o
m
tr
ain
in
g
d
ata
to
v
alid
atio
n
d
ata,
an
d
its
p
er
f
o
r
m
an
ce
is
s
till
lo
w.
On
VGG
-
1
6
,
d
ata
au
g
m
en
tatio
n
h
el
p
ed
th
e
m
o
d
el
lear
n
b
etter
.
T
r
ai
n
lo
s
s
o
f
0
.
5
1
5
5
an
d
v
alid
lo
s
s
o
f
0
.
7
4
8
5
s
h
o
w
th
at
th
e
m
o
d
el
is
ab
le
to
r
ed
u
ce
er
r
o
r
s
in
b
o
th
tr
ain
i
n
g
d
ata
an
d
v
alid
atio
n
d
ata.
T
r
ain
ac
c
u
r
ac
y
o
f
0
.
8
1
6
7
a
n
d
v
alid
ac
cu
r
ac
y
o
f
0
.
7
4
1
1
s
h
o
w
th
at
th
e
m
o
d
el
is
m
o
r
e
s
tab
le
i
n
lear
n
in
g
d
ata
with
au
g
m
en
ta
tio
n
an
d
is
ab
le
to
m
ain
tain
g
o
o
d
g
en
e
r
aliza
tio
n
,
alth
o
u
g
h
th
er
e
is
a
s
lig
h
t
g
ap
b
etwe
en
th
e
p
er
f
o
r
m
an
ce
o
n
tr
ain
in
g
d
ata
an
d
v
alid
atio
n
d
ata.
Me
an
wh
ile,
Mo
b
ileNetV2
ag
ain
s
h
o
wed
th
e
b
est
p
er
f
o
r
m
an
ce
.
W
ith
v
er
y
lo
w
tr
ain
lo
s
s
(
0
.
0
9
7
2
)
a
n
d
v
alid
lo
s
s
(
0
.
6
2
7
9
)
,
th
e
m
o
d
el
s
h
o
wed
ex
ce
llen
t
le
ar
n
in
g
a
b
ilit
y
f
r
o
m
th
e
tr
ai
n
in
g
d
ata
an
d
was
ab
le
to
m
ain
tain
g
en
e
r
aliza
tio
n
o
n
th
e
v
alid
atio
n
d
ata.
T
r
ain
ac
cu
r
ac
y
o
f
0
.
9
6
9
0
an
d
v
alid
ac
cu
r
ac
y
o
f
0
.
8
1
8
9
s
h
o
w
th
at
with
au
g
m
en
tatio
n
th
er
e
is
a
d
ec
r
ea
s
e
in
p
er
f
o
r
m
an
ce
.
T
ab
le
5
.
T
r
ai
n
in
g
r
esu
lts
o
f
m
o
d
el
with
au
g
m
en
tatio
n
D
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
Tr
a
i
n
l
o
ss
V
a
l
i
d
l
o
ss
Tr
a
i
n
a
c
c
u
r
a
c
y
V
a
l
i
d
a
c
c
u
r
a
c
y
R
e
sN
e
t
5
0
1
.
6
3
4
5
1
.
6
2
9
2
0
.
3
9
1
6
0
.
4
3
7
9
VGG
-
16
0
.
5
1
5
5
0
.
7
4
8
5
0
.
8
1
6
7
0
.
7
4
1
1
M
o
b
i
l
e
N
e
t
V
2
0
.
0
9
7
2
0
.
6
2
7
9
0
.
9
6
9
0
0
.
8
1
8
9
3
.
2
.
T
esting
re
s
ults
T
h
e
test
r
esu
lts
u
s
in
g
test
d
ata
f
r
o
m
s
ev
er
al
m
o
d
els,
b
o
th
th
o
s
e
u
s
in
g
au
g
m
en
tatio
n
an
d
th
o
s
e
with
o
u
t
au
g
m
en
tatio
n
,
will
b
e
u
s
ed
to
s
elec
t
t
h
e
b
est
m
o
d
el
am
o
n
g
th
e
th
r
ee
m
o
d
els
ev
alu
ated
.
T
ab
le
6
p
r
esen
ts
a
co
m
p
ar
is
o
n
o
f
th
e
o
v
er
all
test
p
er
f
o
r
m
a
n
ce
.
A
c
o
m
p
ar
is
o
n
was
m
ad
e
to
s
ee
t
h
e
p
er
f
o
r
m
an
ce
o
f
th
r
ee
d
ee
p
lear
n
i
n
g
m
o
d
els,
n
am
ely
R
esNet5
0
,
VGG
-
1
6
,
an
d
Mo
b
ileNetV2
with
an
d
with
o
u
t
th
e
u
s
e
o
f
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es.
Dat
a
au
g
m
e
n
tatio
n
aim
s
to
im
p
r
o
v
e
th
e
g
en
er
aliza
tio
n
o
f
t
h
e
m
o
d
el
to
t
h
e
test
d
ata.
I
n
th
e
R
esNet5
0
m
o
d
el,
th
e
r
esu
lts
s
h
o
wed
th
at
wh
en
au
g
m
en
tatio
n
was
u
s
ed
,
r
ec
all
w
as
3
8
.
9
0
%,
p
r
ec
is
io
n
5
1
.
2
6
%,
F1
-
s
co
r
e
4
4
.
2
3
%,
an
d
ac
cu
r
ac
y
4
1
.
6
0
%.
Ho
we
v
er
,
wh
en
au
g
m
en
tatio
n
was
r
em
o
v
ed
,
th
er
e
was
an
in
cr
ea
s
e
in
r
ec
all
to
4
6
.
0
4
%
a
n
d
ac
cu
r
ac
y
t
o
4
6
.
4
3
%,
alth
o
u
g
h
p
r
ec
is
io
n
d
ec
r
ea
s
ed
s
lig
h
tly
to
4
8
.
2
1
%.
T
h
is
s
h
o
ws
th
at
R
es
Net5
0
g
ain
s
a
s
lig
h
t
ad
v
an
tag
e
with
o
u
t
au
g
m
en
tatio
n
,
b
u
t
its
o
v
er
all
p
e
r
f
o
r
m
an
ce
r
em
ain
s
lo
w
co
m
p
ar
ed
to
o
th
e
r
m
o
d
els.
M
ea
n
wh
ile,
th
e
VGG
-
1
6
m
o
d
el
s
h
o
ws
a
f
air
ly
s
tab
le
p
e
r
f
o
r
m
an
ce
b
o
th
with
an
d
with
o
u
t
au
g
m
e
n
tatio
n
.
W
ith
th
e
ap
p
licatio
n
o
f
au
g
m
en
tatio
n
,
th
e
ac
cu
r
ac
y
v
alu
e
in
c
r
ea
s
ed
b
y
1
.
8
9
%
.
T
h
is
s
h
o
ws
th
at
au
g
m
en
tatio
n
h
elp
s
in
m
ain
tain
in
g
p
er
f
o
r
m
an
ce
s
tab
ilit
y
,
alth
o
u
g
h
th
e
d
if
f
er
en
ce
is
n
o
t
v
er
y
s
ig
n
if
ican
t.
T
h
e
Mo
b
ileNetV2
m
o
d
el
g
av
e
th
e
b
est
r
esu
l
ts
am
o
n
g
th
e
th
r
ee
.
W
ith
au
g
m
en
tatio
n
,
r
ec
all
r
ea
ch
ed
8
1
.
8
9
%,
p
r
ec
is
io
n
8
0
.
4
4
%,
F1
-
s
co
r
e
8
1
.
1
6
%,
an
d
ac
cu
r
ac
y
8
1
.
7
2
%.
W
ith
o
u
t
au
g
m
en
tatio
n
,
th
e
p
er
f
o
r
m
an
ce
ac
t
u
ally
im
p
r
o
v
ed
,
with
r
ec
all
8
4
.
3
5
%,
p
r
ec
is
io
n
8
5
.
2
2
%,
F1
-
s
co
r
e
8
4
.
7
8
%,
an
d
ac
cu
r
ac
y
8
4
.
4
5
%.
T
h
is
s
h
o
ws
th
at
Mo
b
ileNetV2
h
as
an
ex
ce
llen
t
ab
ilit
y
to
g
en
er
alize
d
ata,
ev
en
with
o
u
t
au
g
m
en
tatio
n
.
Ov
er
all,
au
g
m
en
tatio
n
h
as
a
v
ar
y
i
n
g
im
p
ac
t
o
n
ea
c
h
m
o
d
el,
b
u
t
Mo
b
ileNetV2
s
till
ex
ce
ls
in
waste
ty
p
e
class
if
icatio
n
,
b
o
th
with
an
d
with
o
u
t
au
g
m
en
tatio
n
.
T
h
e
c
o
m
p
ar
is
o
n
o
f
th
e
test
ac
cu
r
ac
y
b
etwe
e
n
th
e
th
r
ee
m
o
d
els is
s
h
o
wn
in
Fig
u
r
e
7
.
I
n
m
an
y
ca
s
es,
d
ata
au
g
m
e
n
ta
tio
n
h
elp
s
d
ee
p
lear
n
in
g
m
o
d
e
ls
av
o
id
o
v
er
f
itti
n
g
o
n
tr
ain
in
g
d
ata
an
d
im
p
r
o
v
es
t
h
eir
p
er
f
o
r
m
an
ce
o
n
u
n
s
ee
n
test
d
ata.
B
y
i
n
tr
o
d
u
cin
g
co
n
tr
o
lled
v
ar
iatio
n
,
a
u
g
m
en
tatio
n
e
n
h
an
ce
s
th
e
m
o
d
el’
s
a
b
ilit
y
to
g
e
n
er
a
lize
to
n
ew
in
p
u
ts
.
Ho
wev
er
,
th
er
e
ar
e
s
itu
atio
n
s
wh
er
e
au
g
m
en
tatio
n
m
ay
r
ed
u
ce
ac
cu
r
ac
y
.
T
h
is
ty
p
ic
ally
o
cc
u
r
s
wh
en
th
e
v
ar
iat
io
n
s
in
tr
o
d
u
ce
d
d
o
n
o
t
alig
n
with
th
e
n
atu
r
al
ch
ar
ac
ter
is
tics
o
f
th
e
o
r
ig
in
a
l
d
ata.
Fo
r
in
s
tan
ce
,
in
s
o
m
e
class
if
icatio
n
task
s
,
tr
an
s
f
o
r
m
atio
n
s
s
u
ch
as
90
-
d
e
g
r
ee
r
o
tatio
n
s
o
r
f
lip
s
m
ay
g
en
e
r
ate
im
ag
es
th
at
ar
e
ir
r
elev
a
n
t
o
r
u
n
n
atu
r
al
f
o
r
th
e
class
b
ein
g
r
ec
o
g
n
ized
.
As
a
r
esu
lt,
th
e
m
o
d
el
m
a
y
lear
n
f
r
o
m
m
is
lead
in
g
p
atter
n
s
,
lea
d
in
g
to
i
n
co
r
r
ec
t
p
r
ed
ictio
n
s
.
Mo
r
eo
v
er
,
if
a
m
o
d
el
alr
ea
d
y
d
em
o
n
s
tr
ates
s
tr
o
n
g
p
atter
n
r
ec
o
g
n
itio
n
o
n
th
e
o
r
ig
in
al
d
ataset,
ad
d
itio
n
al
v
ar
iatio
n
ca
n
in
tr
o
d
u
ce
u
n
n
ec
ess
ar
y
n
o
is
e,
r
ed
u
cin
g
its
ab
ilit
y
to
f
o
cu
s
o
n
k
ey
f
ea
t
u
r
es.
T
h
is
p
h
en
o
m
en
o
n
was
o
b
s
er
v
ed
in
Mo
b
ileNetV2
,
wh
er
e
ac
cu
r
ac
y
im
p
r
o
v
ed
wh
en
au
g
m
en
tatio
n
was
n
o
t
ap
p
lied
.
R
esear
ch
co
m
p
ar
in
g
th
r
ee
m
o
d
els
—
R
e
s
Net5
0
,
VGG
-
1
6
,
an
d
M
o
b
il
eNe
tV2
—
f
u
r
th
er
em
p
h
asizes
th
at
th
e
ef
f
ec
t
o
f
au
g
m
en
tatio
n
d
ep
e
n
d
s
o
n
m
o
d
el
ar
c
h
itectu
r
e.
R
esNet5
0
p
er
f
o
r
m
e
d
p
o
o
r
ly
o
v
er
all,
with
o
r
with
o
u
t
au
g
m
en
tatio
n
,
wh
ile
VGG
-
1
6
s
h
o
wed
s
tab
le
b
u
t
m
o
d
er
ate
r
esu
lts
.
Mo
b
ileNetV2
co
n
s
is
t
en
tly
o
u
tp
er
f
o
r
m
e
d
b
o
th
,
e
v
en
with
o
u
t a
u
g
m
e
n
tatio
n
,
p
r
o
v
in
g
to
b
e
th
e
m
o
s
t e
f
f
ec
tiv
e
m
o
d
el
f
o
r
waste
ty
p
e
cl
ass
if
icatio
n
.
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
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f p
r
e
-
tr
a
in
ed
d
ee
p
lea
r
n
in
g
mo
d
el
o
n
…
(
I
K
o
ma
n
g
A
r
ya
Ga
n
d
a
Wig
u
n
a
)
4977
T
ab
le
6.
C
o
m
p
a
r
is
o
n
o
f
b
est
m
o
d
el
r
esu
lts
Pre
-
t
r
a
i
n
e
d
m
o
d
e
l
U
si
n
g
a
u
g
me
n
t
a
t
i
o
n
N
o
a
u
g
me
n
t
a
t
i
o
n
R
e
c
a
l
l
P
r
e
c
i
s
i
o
n
F1
-
sc
o
r
e
A
c
c
u
r
a
c
y
R
e
c
a
l
l
P
r
e
c
i
s
i
o
n
F1
-
sc
o
r
e
A
c
c
u
r
a
c
y
R
e
sN
e
t
5
0
3
8
.
9
0
5
1
.
2
6
4
4
.
2
3
4
1
.
6
0
4
6
.
0
4
4
8
.
2
1
4
7
.
1
0
4
6
.
4
3
VGG
-
16
7
3
.
7
5
7
8
.
2
0
7
5
.
9
1
7
5
.
8
4
7
3
.
6
5
7
6
.
8
5
7
5
.
2
1
7
3
.
9
5
M
o
b
i
l
e
N
e
t
V
2
8
1
.
8
9
8
0
.
4
4
8
1
.
1
6
8
1
.
7
2
8
4
.
3
5
8
5
.
2
2
8
4
.
7
8
8
4
.
4
5
Fig
u
r
e
7
.
C
o
m
p
a
r
is
o
n
o
f
test
ac
cu
r
ac
y
am
o
n
g
R
esNet5
0
,
VG
G
-
1
6
,
an
d
Mo
b
ileNetV2
with
an
d
with
o
u
t
d
ata
au
g
m
e
n
tatio
n
3
.
3
.
B
est
mo
del
T
h
e
m
o
d
el
lo
s
s
g
r
ap
h
d
ep
icts
th
e
lo
s
s
p
r
o
g
r
ess
io
n
d
u
r
in
g
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
f
o
r
b
o
th
tr
ain
in
g
an
d
v
alid
atio
n
d
ata.
Fig
u
r
e
8
illu
s
tr
ates
th
is
lo
s
s
b
eh
av
io
r
ac
r
o
s
s
th
e
th
r
ee
m
o
d
els
test
ed
,
p
r
o
v
id
i
n
g
a
v
is
u
al
co
m
p
ar
is
o
n
o
f
h
o
w
ea
ch
m
o
d
el
r
esp
o
n
d
s
o
v
er
th
e
co
u
r
s
e
o
f
tr
ain
in
g
.
Fro
m
th
e
th
r
ee
g
r
ap
h
s
s
h
o
wn
,
it
ca
n
b
e
s
ee
n
th
at
R
esNet5
0
(
Fig
u
r
e
8
(
a)
)
an
d
VGG
-
16
(
Fig
u
r
e
8
(
b
)
)
m
o
d
els
s
h
o
w
a
less
s
ig
n
if
ican
t
d
ec
r
ea
s
e
in
tr
ain
lo
s
s
.
I
n
ter
m
s
o
f
v
alid
atio
n
lo
s
s
.
I
n
co
n
tr
ast,
Mo
b
ileNetV2
m
o
d
el
(
Fig
u
r
e
8
(
c
)
)
s
h
o
ws
a
d
r
asti
c
d
ec
r
ea
s
e
in
tr
ain
lo
s
s
,
alm
o
s
t
ap
p
r
o
ac
h
i
n
g
ze
r
o
as
t
h
e
n
u
m
b
e
r
o
f
ep
o
ch
s
in
cr
ea
s
es,
in
d
icatin
g
t
h
at
it
is
ab
le
to
lear
n
th
e
p
atter
n
s
in
th
e
tr
ain
in
g
d
ata
w
ell
,
th
e
Mo
b
ileNetV2
m
o
d
el
d
ec
r
ea
s
ed
at
th
e
b
eg
in
n
in
g
,
b
u
t
th
en
s
tab
ilized
an
d
s
tar
ted
to
in
cr
ea
s
e
af
ter
ab
o
u
t f
iv
e
ep
o
ch
s
.
R
esNet5
0
m
o
d
el
s
h
o
wed
a
s
ig
n
if
ican
t in
cr
ea
s
e
in
v
alid
atio
n
lo
s
s
at
th
e
1
0
th
ep
o
c
h
,
w
h
ile
VGG
-
1
6
ex
p
e
r
ien
ce
d
an
i
n
cr
ea
s
e
at
t
h
e
2
0
t
h
ep
o
ch
,
wh
er
e
b
o
th
s
till
s
h
o
wed
in
s
tab
ilit
y
.
T
h
is
in
cr
ea
s
e
in
v
alid
atio
n
lo
s
s
is
an
in
d
icatio
n
o
f
o
v
er
f
itti
n
g
,
wh
er
e
th
e
m
o
d
el
f
its
th
e
tr
ain
in
g
d
ata
to
o
well
b
u
t
d
o
es
n
o
t
g
e
n
er
alize
well
t
o
th
e
v
alid
atio
n
d
ata.
T
h
is
s
h
o
ws
th
at
wh
ile
Mo
b
ileNetV2
m
o
d
el
p
er
f
o
r
m
s
v
er
y
well
o
n
th
e
tr
ain
in
g
d
ata,
it
ca
n
n
o
t
p
e
r
f
o
r
m
o
p
tim
ally
o
n
th
e
v
alid
atio
n
d
ata
.
T
h
e
m
o
d
el
m
em
o
r
izes
th
e
tr
ain
in
g
d
ata
r
at
h
er
th
an
ca
p
tu
r
in
g
m
o
r
e
g
en
e
r
al
p
atter
n
s
,
m
ak
in
g
it
less
ef
f
ec
tiv
e
wh
en
f
a
ce
d
with
n
ew
d
ata.
I
n
ter
m
s
o
f
ep
o
c
h
s
,
it
also
s
h
o
ws
th
at
th
e
Mo
b
ileNetV2
m
o
d
el
co
n
v
e
r
g
es
f
aster
,
r
e
q
u
ir
in
g
o
n
ly
2
5
e
p
o
ch
s
to
r
ea
ch
s
tab
ilit
y
.
T
h
is
is
d
u
e
to
Mo
b
ileNetV2
's
m
o
r
e
lig
h
tweig
h
t
an
d
p
ar
am
eter
-
ef
f
icien
t
a
r
ch
itectu
r
e,
an
d
is
d
esig
n
ed
to
p
er
f
o
r
m
o
p
tim
ally
in
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
.
(
a)
(
b
)
(
c)
Fig
u
r
e
8
.
L
o
s
s
g
r
ap
h
ic
f
o
r
(
a
)
R
esNet5
0
,
(
b
)
VGG
-
1
6
,
a
n
d
(
c)
Mo
b
ileNetV2
L
o
s
s
gr
ap
h
ic
Re
s
Ne
t
50
L
o
s
s
gr
ap
h
ic
VGG
-
16
L
o
s
s
gr
ap
h
ic
M
o
b
il
e
Ne
t
V2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
9
7
1
-
4
9
8
1
4978
T
h
e
g
r
ap
h
in
Fig
u
r
e
9
d
is
p
lay
s
th
e
ac
cu
r
ac
y
g
r
ap
h
s
f
o
r
th
e
th
r
ee
m
o
d
els.
T
h
e
R
esNet5
0
m
o
d
el
also
s
h
o
wed
v
ar
y
in
g
f
lu
ctu
atio
n
s
f
r
o
m
th
e
s
tar
t
o
f
tr
ain
in
g
u
n
til
it
r
ea
ch
ed
5
0
%
ac
cu
r
a
cy
(
Fi
g
u
r
e
9
(
a)
)
,
b
u
t
th
e
v
alid
atio
n
ac
cu
r
ac
y
s
h
o
wed
lo
wer
v
alu
es.
VGG
-
1
6
s
h
o
wed
a
s
tead
ier
in
cr
ea
s
e
in
ac
cu
r
ac
y
u
p
to
a
lift
o
f
9
0
%
(
Fig
u
r
e
9
(
b
)
)
.
T
h
e
v
alid
atio
n
ac
cu
r
ac
y
also
im
p
r
o
v
ed
m
o
r
e
s
ig
n
if
ican
tly
th
an
R
esNet5
0
,
in
d
icatin
g
b
etter
g
en
er
aliza
tio
n
ab
ilit
y
.
Mo
b
ile
NetV2
q
u
ick
ly
ac
h
iev
es
h
ig
h
ac
cu
r
ac
y
o
n
th
e
tr
ai
n
in
g
d
ata,
ap
p
r
o
ac
h
i
n
g
1
0
0
%
with
in
ab
o
u
t
th
e
f
i
r
s
t
7
ep
o
c
h
s
(
Fig
u
r
e
9
(
c)
)
.
Ho
wev
er
,
t
h
e
ac
cu
r
ac
y
o
n
th
e
v
alid
atio
n
d
ata
o
n
ly
r
ea
ch
ed
ab
o
u
t
8
2
%,
with
s
m
all
f
lu
ctu
atio
n
s
th
r
o
u
g
h
o
u
t
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
T
h
e
d
is
cr
ep
an
cy
b
et
wee
n
th
e
v
er
y
h
ig
h
tr
ain
in
g
ac
cu
r
ac
y
an
d
t
h
e
lo
wer
v
alid
atio
n
ac
cu
r
ac
y
in
d
ic
ates
th
at
th
e
m
o
d
el
m
ay
s
u
f
f
er
f
r
o
m
o
v
er
f
itti
n
g
,
wh
er
e
th
e
m
o
d
el
m
em
o
r
izes its
tr
ain
in
g
d
ata
to
o
m
u
c
h
an
d
d
o
es n
o
t g
en
e
r
alize
well
to
n
ew
d
ata.
(
a)
(
b
)
(
c)
Fig
u
r
e
9
.
Acc
u
r
ac
y
g
r
ap
h
o
f
tr
ain
in
g
an
d
v
alid
atio
n
r
esu
lts
ac
r
o
s
s
o
f
(
a)
R
esNet5
0
,
(
b
)
VG
G
-
1
6
,
an
d
(
c)
Mo
b
ileNetV2
T
h
e
co
n
f
u
s
io
n
m
atr
i
x
illu
s
tr
ates
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
Mo
b
ileNetV2
m
o
d
el
in
class
if
y
in
g
9
ca
teg
o
r
ies
o
f
waste,
in
clu
d
i
n
g
ca
r
d
b
o
a
r
d
,
f
o
o
d
o
r
g
an
ics,
g
lass
,
m
etal,
m
is
ce
llan
eo
u
s
tr
ash
,
p
ap
er
,
p
last
ic,
tex
tile
tr
ash
,
an
d
v
eg
etatio
n
.
Fro
m
th
e
m
atr
ix
in
Fig
u
r
e
1
0
,
it
ca
n
b
e
s
ee
n
th
at
th
e
m
o
d
el
h
as
a
f
air
ly
h
ig
h
ac
cu
r
ac
y
in
s
o
m
e
class
es,
s
u
c
h
as
m
etal
with
7
7
co
r
r
ec
t p
r
e
d
ictio
n
s
o
u
t o
f
a
to
tal
o
f
8
5
,
p
l
asti
c
with
8
0
co
r
r
ec
t
p
r
ed
ictio
n
s
o
u
t o
f
8
9
,
an
d
v
e
g
etatio
n
with
4
8
co
r
r
ec
t
p
r
ed
ict
io
n
s
o
u
t o
f
5
0
.
Ho
wev
er
,
s
o
m
e
class
es e
x
p
er
ien
ce
a
f
air
ly
h
ig
h
lev
el
o
f
co
n
f
u
s
io
n
,
esp
ec
ially
in
m
is
ce
llan
eo
u
s
tr
ash
a
n
d
f
o
o
d
o
r
g
a
n
ics,
wh
e
r
e
th
e
m
o
d
el
o
f
ten
m
is
class
if
ies
th
em
as
o
th
er
class
es.
T
h
i
s
s
h
o
ws
th
at
wh
ile
t
h
e
m
o
d
el
wo
r
k
s
well
o
n
s
o
m
e
ca
teg
o
r
ies,
th
er
e
is
s
till
r
o
o
m
f
o
r
im
p
r
o
v
em
en
t o
n
class
es th
at
ar
e
m
o
r
e
am
b
ig
u
o
u
s
o
r
d
i
f
f
icu
lt to
r
ec
o
g
n
ize.
Fig
u
r
e
1
0
.
Mo
b
ileNetV2
co
n
f
u
s
io
n
m
atr
ix
Ac
c
u
r
ac
y
gr
ap
h
Re
s
Ne
t
50
Ac
c
u
r
ac
y
gr
ap
h
VGG
-
16
Ac
c
u
r
ac
y
gr
ap
h
M
o
b
il
e
Ne
t
V2
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
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f p
r
e
-
tr
a
in
ed
d
ee
p
lea
r
n
in
g
mo
d
el
o
n
…
(
I
K
o
ma
n
g
A
r
ya
Ga
n
d
a
Wig
u
n
a
)
4979
4.
CO
NCLU
SI
O
N
B
ased
o
n
th
e
r
esu
lts
,
Mo
b
ileNetV2
s
h
o
wed
th
e
b
est
p
e
r
f
o
r
m
a
n
ce
in
waste
ty
p
e
cla
s
s
if
icatio
n
co
m
p
ar
ed
to
R
esNet5
0
an
d
VGG
-
1
6
.
W
ith
d
ata
a
u
g
m
en
tatio
n
tech
n
iq
u
es,
M
o
b
ile
NetV2
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
8
1
.
7
2
%,
wh
ile
with
o
u
t
au
g
m
e
n
tatio
n
th
e
ac
cu
r
a
cy
in
cr
ea
s
ed
to
8
4
.
4
5
%,
s
h
o
win
g
th
at
th
e
m
o
d
el
is
h
ig
h
ly
ef
f
icien
t
ev
en
with
o
u
t
e
x
ce
s
s
iv
e
au
g
m
e
n
tatio
n
.
VGG
-
1
6
also
s
h
o
ws
s
tab
le
p
er
f
o
r
m
a
n
ce
with
an
ac
cu
r
ac
y
o
f
7
5
.
8
4
%
wh
en
u
s
in
g
au
g
m
e
n
tatio
n
an
d
7
3
.
9
5
%
with
o
u
t
au
g
m
e
n
tatio
n
,
in
d
icatin
g
th
at
au
g
m
en
tatio
n
p
r
o
v
id
es
a
s
lig
h
t,
b
u
t
n
o
t
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t.
I
n
co
n
tr
ast,
R
esNet5
0
h
ad
t
h
e
lo
west
p
er
f
o
r
m
an
ce
with
an
ac
cu
r
ac
y
o
f
o
n
ly
4
1
.
6
0
%
u
s
in
g
a
u
g
m
e
n
tatio
n
an
d
s
lig
h
tly
im
p
r
o
v
ed
to
4
6
.
4
3
%
with
o
u
t
au
g
m
en
tatio
n
,
in
d
icatin
g
th
at
th
is
m
o
d
el
is
less
th
an
o
p
ti
m
al
f
o
r
litt
er
class
if
icatio
n
o
n
th
e
d
ataset
u
s
ed
.
Ov
er
all,
d
ata
au
g
m
e
n
tatio
n
h
a
s
a
d
if
f
er
en
t
ef
f
ec
t
o
n
ea
ch
m
o
d
el,
with
Mo
b
ileNetV2
p
r
o
v
in
g
to
b
e
th
e
m
o
s
t
ef
f
icien
t a
n
d
o
p
tim
al
m
o
d
el
in
th
is
task
,
b
o
th
with
an
d
with
o
u
t a
u
g
m
e
n
tatio
n
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
t
h
o
r
s
wo
u
ld
lik
e
to
th
a
n
k
th
e
Min
is
tr
y
o
f
E
d
u
ca
tio
n
,
C
u
ltu
r
e,
R
esear
ch
an
d
T
ec
h
n
o
lo
g
y
o
f
t
h
e
R
ep
u
b
lic
o
f
I
n
d
o
n
esia
(
Kem
en
d
ik
b
u
d
r
is
tek
)
f
o
r
th
e
f
u
ll
s
u
p
p
o
r
t
g
iv
e
n
in
th
e
im
p
lem
en
tat
io
n
o
f
t
h
is
r
esear
ch
with
n
u
m
b
e
r
1
1
0
/E5
/
PG.0
2
.
0
0
.
PL/2
0
2
4
.
Ass
is
tan
ce
in
th
e
f
o
r
m
o
f
f
u
n
d
in
g
th
r
o
u
g
h
th
e
f
u
n
d
am
en
tal
r
esear
c
h
g
r
an
t p
r
o
g
r
a
m
h
as c
o
n
tr
ib
u
ted
s
ig
n
if
ican
tly
to
th
e
ac
h
iev
em
e
n
t o
f
th
e
r
esu
lts
o
f
th
is
r
esear
c
h
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
r
ec
eiv
ed
f
u
n
d
in
g
f
r
o
m
th
e
Min
is
tr
y
o
f
E
d
u
ca
t
io
n
,
C
u
ltu
r
e,
R
esear
ch
,
a
n
d
T
ec
h
n
o
lo
g
y
o
f
th
e
R
ep
u
b
lic
o
f
I
n
d
o
n
esia (
Kem
en
d
ik
b
u
d
r
is
tek
)
u
n
d
er
g
r
a
n
t n
u
m
b
er
1
1
0
/E5
/PG.0
2
.
0
0
.
P
L
/2
0
2
4
.
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
Ko
m
an
g
Ar
y
a
Gan
d
a
W
ig
u
n
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
Gu
s
ti M
ad
e
Ng
u
r
ah
Desn
an
jay
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
Kad
ek
B
u
d
i San
d
i
k
a
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
d
ec
lar
e
n
o
co
n
f
lict o
f
in
ter
est.
I
NF
O
RM
E
D
CO
NS
E
N
T
W
e
h
av
e
o
b
tain
ed
in
f
o
r
m
ed
c
o
n
s
en
t f
r
o
m
all
in
d
iv
id
u
als in
c
lu
d
ed
in
t
h
is
s
tu
d
y
.
E
T
H
I
CAL AP
P
RO
V
AL
T
h
is
s
tu
d
y
d
o
es
n
o
t
in
v
o
lv
e
h
u
m
an
p
ar
ticip
a
n
ts
o
r
an
i
m
als,
th
er
ef
o
r
e
eth
ical
ap
p
r
o
v
al
is
n
o
t
r
eq
u
ir
ed
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
s
ec
o
n
d
ar
y
d
ata
u
s
ed
in
th
is
s
tu
d
y
ar
e
p
u
b
licly
a
v
ailab
le
o
n
K
ag
g
le
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/jo
eb
ea
ch
ca
p
ital/re
alwa
s
te/
d
ata.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
4
,
No
.
6
,
Dec
em
b
er
2
0
2
5
:
4
9
7
1
-
4
9
8
1
4980
RE
F
E
R
E
NC
E
S
[
1
]
H
.
K
a
u
r
a
n
d
P
.
K
a
u
r
,
“
F
a
c
t
o
r
s
d
e
t
e
r
mi
n
i
n
g
h
o
u
se
h
o
l
d
w
a
s
t
e
se
g
r
e
g
a
t
i
o
n
b
e
h
a
v
i
o
u
r
:
a
n
I
n
d
i
a
n
c
a
s
e
st
u
d
y
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Ex
p
e
r
i
m
e
n
t
a
l
Re
s
e
a
r
c
h
a
n
d
R
e
v
i
e
w
,
v
o
l
.
4
1
,
p
p
.
8
3
–
9
5
,
2
0
2
4
,
d
o
i
:
1
0
.
5
2
7
5
6
/
i
j
e
r
r
.
2
0
2
4
.
v
4
1
s
p
l
.
0
0
7
.
[
2
]
A
.
S
i
d
d
i
q
u
a
,
J.
N
.
H
a
h
l
a
d
a
k
i
s
,
a
n
d
W
.
A
.
K
.
A
.
A
l
-
A
t
t
i
y
a
,
“
A
n
o
v
e
r
v
i
e
w
o
f
t
h
e
e
n
v
i
r
o
n
m
e
n
t
a
l
p
o
l
l
u
t
i
o
n
a
n
d
h
e
a
l
t
h
e
f
f
e
c
t
s
a
sso
c
i
a
t
e
d
w
i
t
h
w
a
s
t
e
l
a
n
d
f
i
l
l
i
n
g
a
n
d
o
p
e
n
d
u
mp
i
n
g
,
”
E
n
v
i
ro
n
m
e
n
t
a
l
S
c
i
e
n
c
e
a
n
d
P
o
l
l
u
t
i
o
n
Re
se
a
r
c
h
,
v
o
l
.
2
9
,
n
o
.
3
9
,
p
p
.
5
8
5
1
4
–
5
8
5
3
6
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
3
5
6
-
0
2
2
-
2
1
5
7
8
-
z.
[
3
]
A
.
P
r
a
m
u
d
i
a
n
t
o
,
“
Th
e
r
o
l
e
o
f
i
n
t
e
r
n
a
t
i
o
n
a
l
l
a
w
a
n
d
n
a
t
i
o
n
a
l
l
a
w
i
n
h
a
n
d
l
i
n
g
mari
n
e
p
l
a
st
i
c
l
i
t
t
e
r
,
”
L
a
m
p
u
n
g
J
o
u
r
n
a
l
o
f
I
n
t
e
r
n
a
t
i
o
n
a
l
L
a
w
,
v
o
l
.
1
,
n
o
.
2
,
p
p
.
4
3
–
5
4
,
2
0
2
0
,
d
o
i
:
1
0
.
2
5
0
4
1
/
l
a
j
i
l
.
v
1
i
2
.
2
0
2
4
.
[
4
]
P
.
W
i
r
i
w
i
t
h
y
a
,
S
.
R
u
n
g
n
a
r
o
n
g
r
u
c
k
,
S
.
P
o
n
g
a
m
p
h
a
i
,
S
.
P
u
a
p
a
t
t
a
n
a
k
u
l
,
a
n
d
R
.
C
h
a
n
c
h
a
r
o
e
n
,
“
S
m
a
r
t
t
r
a
s
h
c
l
a
ssi
f
i
c
a
t
i
o
n
ma
c
h
i
n
e
,
”
i
n
2
0
2
3
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
Me
c
h
a
t
r
o
n
i
c
s
a
n
d
Ro
b
o
t
i
c
s
E
n
g
i
n
e
e
ri
n
g
,
2
0
2
3
,
p
p
.
1
4
6
–
150
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
M
R
E
5
6
7
8
9
.
2
0
2
3
.
1
0
1
0
6
6
0
3
.
[
5
]
V
.
T.
W
i
d
y
a
n
i
n
g
r
u
m,
A
.
S
.
R
o
ma
d
h
o
n
,
a
n
d
R
.
S
a
f
i
t
r
i
,
“
A
u
t
o
m
a
t
i
c
w
a
s
t
e
s
o
r
t
e
r
m
a
c
h
i
n
e
u
si
n
g
p
r
o
x
i
m
i
t
y
se
n
s
o
r
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s,
M
e
d
i
c
a
l
,
B
i
o
l
o
g
i
c
a
l
En
g
i
n
e
e
ri
n
g
,
a
n
d
P
h
a
rm
a
c
e
u
t
i
c
a
l
,
2
0
2
0
,
p
p
.
2
6
4
–
2
7
0
,
d
o
i
:
1
0
.
5
2
2
0
/
0
0
1
0
3
3
1
1
0
2
6
4
0
2
7
0
.
[
6
]
A
.
G
.
K
u
r
b
i
s
,
B
.
La
sc
h
o
w
sk
i
,
a
n
d
A
.
M
i
h
a
i
l
i
d
i
s,
“
S
t
a
i
r
r
e
c
o
g
n
i
t
i
o
n
f
o
r
r
o
b
o
t
i
c
e
x
o
sk
e
l
e
t
o
n
c
o
n
t
r
o
l
u
si
n
g
c
o
m
p
u
t
e
r
v
i
si
o
n
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
2
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Re
h
a
b
i
l
i
t
a
t
i
o
n
Ro
b
o
t
i
c
s
(
I
C
O
R
R)
,
2
0
2
2
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
O
R
R
5
5
3
6
9
.
2
0
2
2
.
9
8
9
6
5
0
1
.
[
7
]
H
.
A
b
d
u
a
n
d
M
.
H
.
M
.
N
o
o
r
,
“
A
s
u
r
v
e
y
o
n
w
a
s
t
e
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
1
2
8
1
5
1
–
1
2
8
1
6
5
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
2
2
6
6
8
2
.
[
8
]
N
.
B
u
d
u
m
a
,
N
.
B
u
d
u
m
a
,
a
n
d
J.
P
a
p
a
,
Fu
n
d
a
m
e
n
t
a
l
s
o
f
d
e
e
p
l
e
a
r
n
i
n
g
.
S
e
b
a
st
o
p
o
l
,
U
n
i
t
e
d
S
t
a
t
e
s
:
O
’
R
e
i
l
l
y
M
e
d
i
a
,
I
n
c
,
2
0
2
2
.
[
9
]
M
.
I
.
B
.
A
h
me
d
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
t
o
r
e
c
y
c
l
a
b
l
e
p
r
o
d
u
c
t
s
c
l
a
ssi
f
i
c
a
t
i
o
n
:
t
o
w
a
r
d
s
su
s
t
a
i
n
a
b
l
e
w
a
s
t
e
ma
n
a
g
e
m
e
n
t
,
”
S
u
s
t
a
i
n
a
b
i
l
i
t
y
,
v
o
l
.
1
5
,
n
o
.
1
4
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
su
1
5
1
4
1
1
1
3
8
.
[
1
0
]
R
.
P
u
sp
i
t
a
a
n
d
C
.
R
a
h
a
y
u
,
“
P
n
e
u
m
o
n
i
a
p
r
e
d
i
c
t
i
o
n
o
n
c
h
e
s
t
x
-
r
a
y
i
ma
g
e
s
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
I
A
ES
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
4
6
7
–
4
7
4
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
1
.
p
p
4
6
7
-
4
7
4
.
[
1
1
]
S
.
R
.
S
h
a
h
,
S
.
Q
a
d
r
i
,
H
.
B
i
b
i
,
S
.
M
.
W
.
S
h
a
h
,
M
.
I
.
S
h
a
r
i
f
,
a
n
d
F
.
M
a
r
i
n
e
l
l
o
,
“
C
o
m
p
a
r
i
n
g
i
n
c
e
p
t
i
o
n
V
3
,
V
G
G
1
6
,
V
G
G
1
9
,
C
N
N
,
a
n
d
R
e
sN
e
t
5
0
:
a
c
a
se
s
t
u
d
y
o
n
e
a
r
l
y
d
e
t
e
c
t
i
o
n
o
f
a
r
i
c
e
d
i
sea
se,
”
A
g
ro
n
o
m
y
,
v
o
l
.
1
3
,
n
o
.
6
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
g
r
o
n
o
m
y
1
3
0
6
1
6
3
3
.
[
1
2
]
L.
Y
o
n
g
,
L
.
M
a
,
D
.
S
u
n
,
a
n
d
L
.
D
u
,
“
A
p
p
l
i
c
a
t
i
o
n
o
f
M
o
b
i
l
e
N
e
t
V
2
t
o
w
a
s
t
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
PL
o
S
O
N
E
,
v
o
l
.
1
8
,
p
p
.
1
–
1
6
,
2
0
2
3
,
d
o
i
:
1
0
.
1
3
7
1
/
j
o
u
r
n
a
l
.
p
o
n
e
.
0
2
8
2
3
3
6
.
[
1
3
]
M
.
Ta
l
a
a
t
,
X
.
S
i
,
a
n
d
J.
X
i
,
“
M
u
l
t
i
-
l
e
v
e
l
t
r
a
i
n
i
n
g
a
n
d
t
e
s
t
i
n
g
o
f
C
N
N
mo
d
e
l
s
i
n
d
i
a
g
n
o
s
i
n
g
m
u
l
t
i
-
c
e
n
t
e
r
C
O
V
I
D
-
1
9
a
n
d
p
n
e
u
mo
n
i
a
X
-
r
a
y
i
ma
g
e
s,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
3
,
n
o
.
1
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
3
1
8
1
0
2
7
0
.
[
1
4
]
Z.
C
h
e
n
,
Z
.
D
i
n
g
,
X
.
Zh
a
n
g
,
X
.
Zh
a
n
g
,
a
n
d
T.
Q
i
n
,
“
I
mp
r
o
v
i
n
g
o
u
t
-
of
-
d
i
st
r
i
b
u
t
i
o
n
g
e
n
e
r
a
l
i
z
a
t
i
o
n
i
n
S
A
R
i
ma
g
e
s
c
e
n
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
w
i
t
h
l
i
m
i
t
e
d
t
r
a
i
n
i
n
g
sa
mp
l
e
s,”
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
5
,
n
o
.
2
4
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s1
5
2
4
5
7
6
1
.
[
1
5
]
S
.
Y
a
n
g
,
W
.
X
i
a
o
,
M
.
Z
h
a
n
g
,
S
.
G
u
o
,
J.
Zh
a
o
,
a
n
d
F
.
S
h
e
n
,
“
I
mag
e
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
f
o
r
d
e
e
p
l
e
a
r
n
i
n
g
:
a
s
u
r
v
e
y
,
”
a
rXi
v
:
2
2
0
4
.
0
8
6
1
0
,
2
0
2
2
.
[
1
6
]
M
.
S
a
t
v
i
l
k
a
r
a
n
d
P
.
N
.
C
o
s
g
r
a
v
e
,
“
I
mag
e
b
a
s
e
d
t
r
a
s
h
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms
f
o
r
r
e
c
y
c
l
a
b
i
l
i
t
y
st
a
t
u
s
,”
M.
S
c
.
Re
s
e
a
r
c
h
Pr
o
j
e
c
t
,
S
c
h
o
o
l
o
f
C
o
mp
u
t
i
n
g
,
N
a
t
i
o
n
a
l
C
o
l
l
e
g
e
o
f
I
r
e
l
a
n
d
,
D
u
b
l
i
n
,
I
r
e
l
a
n
d
,
2
0
1
8
.
[
1
7
]
C
.
A
l
b
o
n
,
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
w
i
t
h
P
y
t
h
o
n
c
o
o
k
b
o
o
k
:
p
r
a
c
t
i
c
a
l
s
o
l
u
t
i
o
n
s
f
r
o
m
p
re
p
r
o
c
e
ssi
n
g
t
o
d
e
e
p
l
e
a
r
n
i
n
g
.
S
e
b
a
st
o
p
o
l
,
U
n
i
t
e
d
S
t
a
t
e
s:
O
’
R
e
i
l
l
y
M
e
d
i
a
,
2
0
1
8
.
[
1
8
]
S
.
A
.
K
a
mr
a
n
e
t
a
l
.
,
“
S
A
N
S
-
C
N
N
:
a
n
a
u
t
o
ma
t
e
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
f
o
r
s
p
a
c
e
f
l
i
g
h
t
a
ss
o
c
i
a
t
e
d
n
e
u
r
o
-
o
c
u
l
a
r
sy
n
d
r
o
m
e
w
i
t
h
a
s
t
r
o
n
a
u
t
i
m
a
g
i
n
g
d
a
t
a
,
”
n
p
j
M
i
c
ro
g
r
a
v
i
t
y
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
1
–
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
2
6
-
0
2
4
-
0
0
3
6
4
-
w.
[
1
9
]
S
.
S
i
n
g
l
e
,
S
.
I
r
a
n
ma
n
e
s
h
,
a
n
d
R
.
R
a
a
d
,
“
R
e
a
l
W
a
st
e
:
a
n
o
v
e
l
r
e
a
l
-
l
i
f
e
d
a
t
a
se
t
f
o
r
l
a
n
d
f
i
l
l
w
a
s
t
e
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
I
n
f
o
rm
a
t
i
o
n
,
v
o
l
.
1
4
,
n
o
.
1
2
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
i
n
f
o
1
4
1
2
0
6
3
3
.
[
2
0
]
M
.
P
.
S
a
l
a
s
a
n
d
P
.
L
.
D
.
G
e
u
s
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
p
p
l
i
e
d
t
o
i
mb
a
l
a
n
c
e
d
mal
w
a
r
e
d
a
t
a
se
t
s
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
J
o
u
rn
a
l
o
f
I
n
t
e
rn
e
t
S
e
r
v
i
c
e
s
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
5
,
n
o
.
1
,
p
p
.
3
4
2
–
3
5
9
,
2
0
2
4
,
d
o
i
:
1
0
.
5
7
5
3
/
j
i
sa.
2
0
2
4
.
3
9
0
7
.
[
2
1
]
M
.
H
a
r
a
h
a
p
,
V
.
D
a
mar,
S
.
Y
e
k
,
M
.
M
i
c
h
a
e
l
,
a
n
d
M
.
R
.
P
u
t
r
a
,
“
S
t
a
t
i
c
a
n
d
d
y
n
a
m
i
c
h
u
ma
n
a
c
t
i
v
i
t
y
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
V
G
G
-
1
6
p
r
e
-
t
r
a
i
n
e
d
C
N
N
m
o
d
e
l
,
”
J
u
r
n
a
l
I
n
f
o
t
e
l
,
v
o
l
.
1
5
,
n
o
.
2
,
p
p
.
4
4
–
4
8
,
2
0
2
3
,
d
o
i
:
1
0
.
2
0
8
9
5
/
i
n
f
o
t
e
l
.
v
1
5
i
2
.
9
1
6
.
[
2
2
]
E.
A
y
a
n
a
n
d
H
.
M
.
Ü
n
v
e
r
,
“
D
a
t
a
a
u
g
men
t
a
t
i
o
n
i
m
p
o
r
t
a
n
c
e
f
o
r
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
s
k
i
n
l
e
s
i
o
n
s
v
i
a
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
1
8
El
e
c
t
ri
c
El
e
c
t
r
o
n
i
c
s
,
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
B
i
o
m
e
d
i
c
a
l
En
g
i
n
e
e
r
i
n
g
s’
M
e
e
t
i
n
g
,
2
0
1
8
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
E
B
B
T.
2
0
1
8
.
8
3
9
1
4
6
9
.
[
2
3
]
E.
Ta
s
c
i
,
C
.
U
l
u
t
u
r
k
,
a
n
d
A
.
U
g
u
r
,
“
A
v
o
t
i
n
g
-
b
a
s
e
d
e
n
sem
b
l
e
d
e
e
p
l
e
a
r
n
i
n
g
me
t
h
o
d
f
o
c
u
si
n
g
o
n
i
m
a
g
e
a
u
g
m
e
n
t
a
t
i
o
n
a
n
d
p
r
e
p
r
o
c
e
ssi
n
g
v
a
r
i
a
t
i
o
n
s fo
r
t
u
b
e
r
c
u
l
o
si
s d
e
t
e
c
t
i
o
n
,
”
N
e
u
r
a
l
C
o
m
p
u
t
i
n
g
a
n
d
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
3
,
n
o
.
2
2
,
p
p
.
1
5
5
4
1
–
1
5
5
5
5
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
2
1
-
021
-
0
6
1
7
7
-
2.
[
2
4
]
M
.
S
a
n
d
l
e
r
,
A
.
H
o
w
a
r
d
,
M
.
Z
h
u
,
A
.
Z
h
m
o
g
i
n
o
v
,
a
n
d
L
.
-
C
.
C
h
e
n
,
“
M
o
b
i
l
e
N
et
V
2
:
i
n
v
e
r
t
e
d
r
e
s
i
d
u
a
l
s
a
n
d
l
i
n
e
a
r
b
o
t
t
l
e
n
e
c
k
s
,
”
i
n
2
0
1
8
I
E
E
E
/
C
V
F
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
s
i
o
n
a
n
d
P
a
t
t
e
r
n
R
e
c
o
g
n
i
t
i
o
n
,
2
0
1
8
,
p
p
.
4
5
1
0
–
4
5
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
8
.
0
0
4
7
4
.
[
2
5
]
M
.
A
k
a
y
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
c
l
a
s
s
i
f
i
c
a
t
i
o
n
o
f
sy
s
t
e
m
i
c
sc
l
e
r
o
si
s
sk
i
n
u
s
i
n
g
t
h
e
M
o
b
i
l
e
N
e
t
V
2
m
o
d
e
l
,
”
I
EEE
O
p
e
n
J
o
u
r
n
a
l
o
f
En
g
i
n
e
e
ri
n
g
i
n
Me
d
i
c
i
n
e
a
n
d
B
i
o
l
o
g
y
,
v
o
l
.
2
,
p
p
.
1
0
4
–
1
1
0
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
O
JE
M
B
.
2
0
2
1
.
3
0
6
6
0
9
7
.
[
2
6
]
K
.
D
o
n
g
,
C
.
Zh
o
u
,
Y
.
R
u
a
n
,
a
n
d
Y
.
Li
,
“
M
o
b
i
l
e
N
e
t
V
2
m
o
d
e
l
f
o
r
i
m
a
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
2
0
2
0
2
n
d
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
C
o
m
p
u
t
e
r A
p
p
l
i
c
a
t
i
o
n
,
D
e
c
.
2
0
2
0
,
p
p
.
4
7
6
–
480
,
d
o
i
:
1
0
.
1
1
0
9
/
I
TC
A
5
2
1
1
3
.
2
0
2
0
.
0
0
1
0
6
.
[
2
7
]
A
.
V
.
S
a
mso
n
o
v
i
c
h
,
“
U
n
i
v
e
r
sa
l
l
e
a
r
n
e
r
a
s
a
n
e
m
b
r
y
o
o
f
c
o
m
p
u
t
a
t
i
o
n
a
l
c
o
n
sc
i
o
u
s
n
e
ss
,
”
i
n
AA
AI
F
a
l
l
S
y
m
p
o
si
u
m
-
T
e
c
h
n
i
c
a
l
Re
p
o
r
t
,
2
0
0
7
,
p
p
.
1
2
9
–
1
3
4
.
[
2
8
]
S
.
A
.
A
l
t
h
u
b
i
t
i
,
F
.
A
l
e
n
e
z
i
,
S
.
S
h
i
t
h
a
r
t
h
,
S
.
K
.
,
a
n
d
C
.
V
.
S
.
R
e
d
d
y
,
“
C
i
r
c
u
i
t
m
a
n
u
f
a
c
t
u
r
i
n
g
d
e
f
e
c
t
d
e
t
e
c
t
i
o
n
u
s
i
n
g
V
G
G
1
6
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
Wi
re
l
e
ss
C
o
m
m
u
n
i
c
a
t
i
o
n
s
a
n
d
Mo
b
i
l
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
0
2
2
,
n
o
.
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
2
/
1
0
7
0
4
0
5
.
[
2
9
]
O
.
N
.
B
e
l
a
i
d
a
n
d
M
.
Lo
u
d
i
n
i
,
“
C
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
b
r
a
i
n
t
u
m
o
r
b
y
c
o
mb
i
n
a
t
i
o
n
o
f
p
r
e
-
t
r
a
i
n
e
d
V
G
G
1
6
C
N
N
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
Ma
n
a
g
e
m
e
n
t
,
v
o
l
.
1
2
,
n
o
.
2
,
p
p
.
1
3
–
2
5
,
2
0
2
0
,
d
o
i
:
1
0
.
2
2
0
5
9
/
JI
TM
.
2
0
2
0
.
7
5
7
8
8
.
[
3
0
]
D
.
Th
e
c
k
e
d
a
t
h
a
n
d
R
.
R
.
S
e
d
a
m
k
a
r
,
“
D
e
t
e
c
t
i
n
g
a
f
f
e
c
t
s
t
a
t
e
s u
si
n
g
V
G
G
1
6
,
R
e
sN
e
t
5
0
a
n
d
S
E
-
R
e
sN
e
t
5
0
n
e
t
w
o
r
k
s
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
,
n
o
.
2
,
p
p
.
1
–
7
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s4
2
9
7
9
-
0
2
0
-
0
1
1
4
-
9.
[
3
1
]
P
.
B
h
a
t
e
t
a
l
.
,
“
B
r
a
i
n
t
u
mo
r
d
e
t
e
c
t
i
o
n
u
si
n
g
C
N
N
,
”
i
n
C
o
m
m
u
n
i
c
a
t
i
o
n
s
i
n
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
,
2
0
2
4
,
p
p
.
1
8
–
28
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
031
-
7
0
0
0
1
-
9
_
2
.
[
3
2
]
N
.
B
e
h
a
r
a
n
d
M
.
S
h
r
i
v
a
s
t
a
v
a
,
“
R
e
s
N
e
t
5
0
-
b
a
se
d
e
f
f
e
c
t
i
v
e
m
o
d
e
l
f
o
r
b
r
e
a
st
c
a
n
c
e
r
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
h
i
s
t
o
p
a
t
h
o
l
o
g
y
i
ma
g
e
s
,
”
C
ME
S
-
C
o
m
p
u
t
e
r
Mo
d
e
l
i
n
g
i
n
E
n
g
i
n
e
e
ri
n
g
&
S
c
i
e
n
c
e
s
,
v
o
l
.
1
3
0
,
n
o
.
2
,
2
0
2
2
.
[
3
3
]
E.
S
i
t
o
m
p
u
l
,
V
.
L.
S
e
t
i
a
w
a
n
,
H
.
J.
T
a
r
i
g
a
n
,
a
n
d
M
.
G
a
l
i
n
a
,
“
I
mag
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
f
a
b
r
i
c
d
e
f
e
c
t
s
u
s
i
n
g
R
e
sN
e
t
5
0
d
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
i
n
F
a
st
A
I
,
”
B
u
l
l
e
t
i
n
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
ri
n
g
a
n
d
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
3
,
n
o
.
5
,
p
p
.
3
2
5
5
–
3
2
6
7
,
2
0
2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.