I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2020
,
p
p
.
1
0
1
7
~
1
0
2
6
I
SS
N:
2088
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
10
i
1
.
pp
1
0
1
7
-
1
0
2
6
1017
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m/in
d
ex
.
p
h
p
/I
JE
C
E
Res
S
eg
:
R
esidua
l
enco
der
-
de
co
der
co
nv
o
lutiona
l neural
netw
o
rk
f
o
r f
o
o
d
seg
m
en
tatio
n
J
a
v
ier
O
.
P
inzó
n
-
Are
na
s
,
Ro
bin
s
o
n J
i
m
é
nez
-
M
o
re
no
,
Cesa
r
G
.
P
a
chó
n
-
Su
esc
ún
F
a
c
u
lt
y
o
f
En
g
in
e
e
rin
g
,
Nu
e
v
a
G
r
a
n
a
d
a
M
il
it
a
ry
Un
iv
e
rsit
y
,
Co
lo
m
b
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
2
,
2
0
1
9
R
ev
i
s
ed
Sep
2
,
2
0
1
9
A
cc
ep
ted
Sep
2
7
,
2
0
1
9
T
h
is
p
a
p
e
r
p
re
se
n
ts
th
e
i
m
p
le
m
e
n
tatio
n
a
n
d
e
v
a
lu
a
ti
o
n
o
f
d
iff
e
re
n
t
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
a
rc
h
it
e
c
tu
re
s
f
o
c
u
se
d
o
n
f
o
o
d
se
g
m
e
n
tatio
n
.
T
o
p
e
rf
o
r
m
th
is
tas
k
,
it
is
p
ro
p
o
se
d
th
e
re
c
o
g
n
it
i
o
n
o
f
6
c
a
teg
o
ri
e
s,
a
m
o
n
g
w
h
ich
a
re
th
e
m
a
in
f
o
o
d
g
ro
u
p
s
(p
ro
tein
,
g
ra
in
s,
f
ru
it
,
a
n
d
v
e
g
e
t
a
b
les
)
a
n
d
tw
o
a
d
d
it
io
n
a
l
g
ro
u
p
s,
rice
a
n
d
d
rin
k
o
r
ju
ice
.
In
a
d
d
it
i
o
n
,
to
m
a
k
e
th
e
re
c
o
g
n
it
i
o
n
m
o
re
c
o
m
p
lex
,
it
is
d
e
c
id
e
d
t
o
tes
t
th
e
n
e
tw
o
rk
s
w
it
h
f
o
o
d
d
ish
e
s
a
lrea
d
y
sta
rted
,
i.
e
.
d
u
rin
g
d
if
fe
re
n
t
m
o
m
e
n
ts,
f
ro
m
it
s
se
r
v
in
g
to
it
s
f
in
ish
in
g
,
in
o
rd
e
r
t
o
v
e
rify
th
e
c
a
p
a
b
il
it
y
to
se
e
w
h
e
n
th
e
re
is
n
o
m
o
re
f
o
o
d
o
n
t
h
e
p
late
.
F
i
n
a
ll
y
,
a
c
o
m
p
a
ris
o
n
is
m
a
d
e
b
e
twe
e
n
th
e
tw
o
b
e
s
t
re
su
lt
in
g
n
e
tw
o
rk
s,
a
S
e
g
N
e
t
w
it
h
a
rc
h
it
e
c
tu
re
V
GG
-
1
6
a
n
d
a
n
e
tw
o
rk
p
ro
p
o
se
d
in
th
is
w
o
rk
,
c
a
ll
e
d
Re
sid
u
a
l
S
e
g
m
e
n
tatio
n
Co
n
v
o
lu
t
io
n
a
l
Ne
u
ra
l
N
e
tw
o
rk
o
r
Re
sSeg
,
w
it
h
w
h
ich
a
c
c
u
ra
c
ies
g
r
e
a
ter
th
a
n
9
0
%
a
n
d
in
terc
e
p
ti
o
n
-
o
v
e
r
-
u
n
io
n
g
re
a
ter
th
a
n
7
5
%
w
e
re
o
b
tain
e
d
.
T
h
is
d
e
m
o
n
stra
tes
th
e
a
b
il
it
y
,
n
o
t
o
n
ly
o
f
S
e
g
Ne
t
a
rc
h
it
e
c
tu
re
s
f
o
r
f
o
o
d
se
g
m
e
n
tatio
n
,
b
u
t
t
h
e
u
se
o
f
re
sid
u
a
l
lay
e
r
s
to
im
p
ro
v
e
th
e
c
o
n
to
u
r
o
f
th
e
se
g
m
e
n
tatio
n
a
n
d
se
g
m
e
n
tatio
n
o
f
c
o
m
p
le
x
d
istri
b
u
ti
o
n
o
r
in
it
iate
d
o
f
f
o
o
d
d
ish
e
s,
o
p
e
n
in
g
th
e
f
ield
o
f
a
p
p
li
c
a
ti
o
n
o
f
th
is
ty
p
e
o
f
n
e
tw
o
rk
s
to
b
e
im
p
le
m
e
n
ted
in
f
e
e
d
in
g
a
ss
ista
n
ts
o
r
i
n
a
u
to
m
a
ted
re
sta
u
ra
n
ts,
in
c
l
u
d
i
n
g
a
lso
f
o
r
d
ieta
ry
c
o
n
tro
l
f
o
r
th
e
a
m
o
u
n
t
o
f
f
o
o
d
c
o
n
s
u
m
e
d
.
K
ey
w
o
r
d
s
:
E
n
co
d
er
-
d
ec
o
d
er
C
NN
Fo
o
d
r
ec
o
g
n
itio
n
R
esid
u
al
la
y
er
s
Seg
Net
Se
m
a
n
tic
s
e
g
m
e
n
tatio
n
Co
p
y
rig
h
t
©
2
0
2
0
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
J
av
ier
Or
lan
d
o
P
in
zó
n
A
r
e
n
as
,
Me
ch
atr
o
n
ic
s
E
n
g
in
ee
r
i
n
g
P
r
o
g
r
a
m
,
Fac
u
lt
y
o
f
E
n
g
i
n
ee
r
i
n
g
,
Nu
e
v
a
Gr
an
ad
a
Militar
y
,
U
n
i
v
er
s
it
y
,
C
ar
r
er
a
1
1
#
1
0
1
-
8
0
,
B
o
g
o
tá
D
.
C
.
,
C
o
lo
m
b
ia.
E
m
ail:
u
3
9
0
0
2
3
1
@
u
n
i
m
ilit
ar
.
ed
u
.
co
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ec
o
g
n
itio
n
o
f
p
atter
n
s
a
p
p
lied
to
f
o
o
d
is
a
to
p
ic
th
at
h
as
b
eg
u
n
to
ta
k
e
i
m
p
o
r
tan
ce
w
it
h
i
n
m
ac
h
in
e
v
is
io
n
s
y
s
te
m
s
,
m
ai
n
l
y
f
o
c
u
s
ed
o
n
ca
lo
r
ie
co
n
tr
o
l
[
1
]
o
r
d
iet
co
n
tr
o
l
[
2
]
.
Ho
w
ev
er
,
th
at
ap
p
licatio
n
h
as
b
ee
n
a
ch
alle
n
g
e
b
ec
au
s
e
th
e
d
is
h
e
s
d
o
n
o
t
co
n
tain
ch
ar
ac
ter
is
tic
s
h
ap
e
s
o
r
th
e
in
g
r
ed
ie
n
ts
u
s
ed
ca
n
d
if
f
er
b
y
d
r
asti
ca
ll
y
c
h
a
n
g
i
n
g
t
h
e
v
is
u
al
ch
ar
ac
ter
i
s
tics
o
f
a
t
y
p
e
o
f
f
o
o
d
.
Fo
r
th
e
e
x
ec
u
tio
n
o
f
t
h
i
s
tas
k
,
s
ev
er
a
l
d
ev
elo
p
m
en
t
s
h
a
v
e
b
ee
n
i
m
p
l
e
m
en
te
d
,
s
u
ch
as
th
o
s
e
p
r
ese
n
ted
i
n
[
3
-
5
]
,
b
u
t
t
h
e
m
ai
n
p
r
o
b
lem
is
t
h
at
t
h
ese
s
y
s
te
m
s
d
ep
en
d
o
n
th
e
d
is
h
co
n
tai
n
in
g
o
n
l
y
o
n
e
t
y
p
e
o
f
f
o
o
d
,
r
ec
o
g
n
izin
g
a
s
p
ec
if
ic
d
is
h
w
i
th
o
u
t
d
is
cr
i
m
i
n
ati
n
g
t
h
e
i
n
g
r
ed
ie
n
ts
[
6
]
o
r
b
ein
g
in
a
co
n
tr
o
lled
o
r
k
n
o
w
n
en
v
ir
o
n
m
e
n
t [
7
,
8
].
T
o
in
cr
ea
s
e
th
e
r
o
b
u
s
tn
e
s
s
o
f
t
h
e
f
o
o
d
r
ec
o
g
n
itio
n
s
y
s
te
m
s
,
o
t
h
er
tec
h
n
iq
u
es
h
a
v
e
b
eg
u
n
to
b
e
ap
p
lied
in
a
w
a
y
t
h
at
allo
w
s
r
ec
o
g
n
izi
n
g
d
i
f
f
er
e
n
t
t
y
p
es
o
f
f
o
o
d
in
a
s
i
n
g
le
d
is
h
.
So
m
e
d
ev
elo
p
ed
ex
a
m
p
le
s
ar
e
p
r
esen
ted
in
[
9
]
an
d
[
10
]
.
T
h
e
f
ir
s
t
o
n
e
m
a
k
es
u
s
e
o
f
p
air
w
i
s
e
s
tati
s
tic
s
to
r
ec
o
g
n
ize
d
if
f
er
e
n
t
ca
teg
o
r
ies
o
f
f
o
o
d
o
n
a
p
late
b
y
m
ea
n
s
o
f
th
e
ex
p
lo
itatio
n
o
f
lo
ca
l
f
ea
tu
r
es,
b
u
t
it
d
ep
en
d
s
o
n
b
ein
g
in
a
co
n
tr
o
lled
en
v
ir
o
n
m
e
n
t
w
it
h
a
w
h
ite
b
ac
k
g
r
o
u
n
d
.
I
n
th
e
s
ec
o
n
d
o
n
e,
a
co
m
b
in
a
tio
n
o
f
s
e
g
m
e
n
tati
o
n
o
f
o
b
j
ec
ts
w
i
t
h
p
er
ce
p
tu
al
s
i
m
ilar
i
ties
is
u
s
ed
,
in
s
u
c
h
a
w
a
y
t
h
at
th
e
al
g
o
r
ith
m
r
ec
o
g
n
ize
s
an
d
s
eg
m
e
n
ts
f
o
o
d
t
y
p
es
w
it
h
r
esp
ec
t
to
th
eir
ch
ar
ac
ter
is
t
ics,
ac
h
ie
v
i
n
g
a
n
ac
cu
r
ac
y
o
f
4
4
%
in
t
h
e
s
e
g
m
e
n
tat
io
n
o
f
3
2
m
ea
ls
.
A
lt
h
o
u
g
h
t
h
ese
tec
h
n
iq
u
es
alr
ea
d
y
d
is
cr
i
m
i
n
ate
p
ar
ts
o
f
t
h
e
s
a
m
e
d
is
h
,
t
h
e
p
er
f
o
r
m
a
n
ce
is
v
er
y
lo
w
,
w
h
ich
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
0
:
1
0
1
7
-
1
0
2
6
1018
w
h
y
r
esear
ch
er
s
b
eg
a
n
to
i
n
clu
d
e
ar
ti
f
icia
l
in
telli
g
en
ce
in
t
h
i
s
ar
ea
,
m
o
r
e
s
p
ec
i
f
ic
all
y
Dee
p
L
ea
r
n
in
g
(
DL
)
tech
n
iq
u
e
s
[
11
]
.
W
ith
in
D
L
,
t
h
er
e
i
s
a
tech
n
iq
u
e
ca
lled
C
o
n
v
o
lu
tio
n
al
Neu
r
al
Net
w
o
r
k
(
C
NN)
[
1
2
,
1
3
]
,
w
h
ic
h
h
as
h
ad
an
e
x
p
o
n
e
n
tial
e
v
o
lu
tio
n
,
n
o
t
o
n
l
y
i
n
it
s
p
er
f
o
r
m
a
n
ce
i
n
t
h
e
r
ec
o
g
n
itio
n
o
f
p
atter
n
s
i
n
i
m
ag
e
s
b
u
t
i
n
it
s
v
ar
iatio
n
s
an
d
ap
p
licatio
n
.
I
n
th
e
w
o
r
k
d
escr
ib
ed
in
[1
4
]
,
a
co
m
p
ar
is
o
n
is
m
ad
e
b
et
w
ee
n
r
eg
io
n
-
b
ased
C
N
N
an
d
a
Dee
p
C
N
N,
to
s
eg
m
e
n
t
an
d
lo
ca
te
f
o
o
d
in
a
p
h
o
to
,
wh
ile
in
[1
5
]
,
a
Dee
p
C
NN
i
s
u
s
ed
w
it
h
a
v
ar
iat
io
n
in
t
h
e
s
tr
u
ct
u
r
e
o
f
th
e
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
to
s
e
g
m
en
t
th
e
f
o
o
d
alo
n
g
w
it
h
a
m
u
lti
-
s
ca
le
C
NN
to
e
s
ti
m
ate
it
s
d
ep
th
,
ac
h
iev
i
n
g
r
esu
lts
w
i
th
r
an
g
e
s
b
et
w
ee
n
7
0
% a
n
d
7
5
% a
cc
u
r
ac
y
i
n
th
e
e
s
ti
m
atio
n
o
f
f
o
o
d
q
u
an
tit
y
.
On
t
h
e
o
th
er
h
a
n
d
,
o
th
er
a
r
ch
itect
u
r
es
h
av
e
b
ee
n
d
ev
e
lo
p
ed
to
ac
h
iev
e
an
i
m
p
r
o
v
e
m
en
t
i
n
th
e
s
e
g
m
e
n
tat
io
n
o
f
o
b
j
ec
ts
,
c
alled
E
n
co
d
er
-
Dec
o
d
er
C
NNs,
o
r
c
o
m
m
o
n
l
y
k
n
o
w
n
a
s
Seg
Net
[1
6
]
,
w
h
ic
h
i
s
a
C
NN
t
h
at
co
n
s
is
ts
o
f
t
wo
s
tag
e
s
.
T
h
e
f
ir
s
t
s
ta
g
e
co
n
s
i
s
ts
o
f
a
n
e
n
co
d
er
in
ch
a
r
g
e
o
f
g
e
n
er
ati
n
g
th
e
r
ec
o
g
n
itio
n
o
f
t
h
e
o
b
j
ec
t,
h
o
w
ev
er
,
it
d
o
es
n
o
t
co
n
tai
n
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
,
i.e
.
i
ts
las
t
co
n
v
o
lu
tio
n
la
y
er
is
co
n
n
ec
ted
d
ir
ec
tl
y
to
th
e
s
ec
o
n
d
s
tag
e,
w
h
ic
h
i
s
a
m
ir
r
o
r
o
f
t
h
e
e
n
co
d
er
,
ca
lled
d
ec
o
d
er
,
ad
d
in
g
a
d
ir
ec
t
co
n
n
ec
tio
n
b
et
w
ee
n
ea
ch
s
ec
tio
n
o
f
t
h
e
en
co
d
er
's
d
o
w
n
s
a
m
p
li
n
g
w
i
th
it
s
r
esp
ec
tiv
e
p
ar
t
o
f
th
e
d
ec
o
d
er
'
s
u
p
s
a
m
p
li
n
g
,
allo
w
i
n
g
h
av
i
n
g
a
b
etter
ch
ar
ac
ter
izatio
n
o
f
th
e
i
m
ag
e
i
n
th
e
last
la
y
er
s
.
T
h
is
t
y
p
e
o
f
n
et
w
o
r
k
h
as
h
ad
a
g
r
ea
t
p
er
f
o
r
m
a
n
ce
i
n
ta
s
k
s
r
elate
d
to
t
h
e
s
e
g
m
en
tatio
n
o
f
o
b
j
ec
ts
[
1
7
-
1
9
]
,
ev
en
in
th
e
s
eg
m
e
n
tatio
n
o
f
m
ed
ical
i
m
ag
es
[
20
-
2
3
]
,
w
h
ich
h
av
e
t
h
e
ch
ar
ac
ter
is
tic
o
f
n
o
t
h
a
v
i
n
g
s
ec
t
i
o
n
s
w
it
h
a
s
p
ec
if
ic
s
h
ap
e
o
r
to
tall
y
a
m
o
r
p
h
o
u
s
.
Ho
w
e
v
er
,
th
i
s
n
et
w
o
r
k
h
a
s
n
o
t
b
ee
n
w
id
el
y
u
s
ed
i
n
th
e
ta
s
k
o
f
f
o
o
d
s
eg
m
e
n
tatio
n
,
f
o
r
t
h
is
r
ea
s
o
n
,
t
h
is
w
o
r
k
e
x
p
lo
r
es
th
e
p
o
s
s
ib
ilit
y
o
f
b
ei
n
g
u
s
ed
a
n
d
d
e
m
o
n
s
tr
ate
i
ts
p
er
f
o
r
m
a
n
ce
i
n
th
i
s
task
.
A
lt
h
o
u
g
h
t
h
e
f
o
o
d
s
eg
m
en
tatio
n
d
ev
elo
p
m
e
n
t
s
ar
e
m
a
in
l
y
f
o
c
u
s
ed
o
n
th
e
d
ietar
y
co
n
tr
o
l,
th
i
s
w
o
r
k
e
x
p
an
d
s
th
e
u
s
e
o
f
t
h
ese
s
y
s
te
m
s
to
k
n
o
w
th
e
ex
i
s
te
n
ce
o
r
n
o
t
o
f
f
o
o
d
o
n
a
d
is
h
,
s
o
t
h
at
it
ca
n
b
e
ap
p
lied
in
f
u
t
u
r
e
w
o
r
k
to
d
ev
elo
p
m
e
n
ts
t
h
at
r
eq
u
ir
e
k
n
o
w
i
n
g
th
e
p
er
ce
n
ta
g
e
o
f
cu
r
r
en
t
f
o
o
d
o
r
au
to
n
o
m
o
u
s
s
y
s
te
m
s
o
f
ass
is
t
ed
f
ee
d
in
g
.
L
i
k
e
w
i
s
e,
d
if
f
er
en
t
ar
ch
itect
u
r
es
ar
e
i
m
p
le
m
e
n
te
d
an
d
ev
alu
ated
to
an
al
y
ze
t
h
eir
r
es
u
lt
s
w
it
h
r
es
p
ec
t
to
an
ar
ch
itectu
r
e
p
r
o
p
o
s
ed
in
th
e
s
tate
o
f
t
h
e
ar
t,
e
x
p
lo
r
in
g
t
h
e
u
s
e
o
f
r
esid
u
al
la
y
er
s
in
co
n
j
u
n
ctio
n
w
it
h
t
h
e
Se
g
Net,
w
h
ich
i
s
ca
ll
ed
in
t
h
i
s
w
o
r
k
R
e
s
Se
g
.
T
h
e
w
o
r
k
i
s
d
iv
id
ed
in
to
4
s
ec
tio
n
s
,
i
n
cl
u
d
in
g
t
h
e
p
r
ese
n
t
in
tr
o
d
u
ctio
n
.
I
n
s
ec
tio
n
2
,
t
h
e
d
atab
ase
p
r
ep
ar
e
d
alo
n
g
w
i
th
t
h
e
p
r
o
p
o
s
ed
ar
ch
itect
u
r
es
i
s
p
r
esen
te
d
.
Sectio
n
3
d
escr
ib
es
th
e
r
es
u
lts
o
b
tain
ed
f
r
o
m
th
e
tr
ain
i
n
g
an
d
test
i
n
g
o
f
t
h
e
n
et
w
o
r
k
s
,
tak
i
n
g
in
to
ac
co
u
n
t
th
e
u
s
e
an
d
n
o
n
-
u
s
e
o
f
th
e
b
ac
k
g
r
o
u
n
d
lab
el.
Fin
all
y
,
i
n
s
ec
tio
n
4
,
th
e
co
n
cl
u
s
io
n
s
r
ea
ch
ed
ar
e
g
i
v
en
.
2.
M
E
T
H
O
DO
L
O
G
Y
T
h
e
w
o
r
k
d
o
n
e
f
o
cu
s
es
o
n
th
e
s
eg
m
e
n
tatio
n
o
f
6
f
o
o
d
g
r
o
u
p
s
,
f
o
r
t
h
is
ca
s
e,
f
r
o
m
t
h
e
l
u
n
ch
m
ea
l,
w
it
h
i
n
w
h
ich
ar
e
th
e
4
m
ai
n
g
r
o
u
p
s
o
f
f
o
o
d
s
(
P
r
o
tein
,
Veg
etab
les,
Fru
i
ts
,
an
d
Gr
ain
s
)
p
lu
s
t
w
o
s
u
b
g
r
o
u
p
s
,
w
h
er
e
J
u
ice
a
n
d
R
ice
ar
e
lo
ca
ted
.
T
h
is
is
d
o
n
e
s
i
n
ce
,
i
n
th
e
ca
s
e
o
f
r
ice,
i
n
t
h
e
co
m
m
o
n
f
o
o
d
m
e
n
u
o
f
th
e
C
o
lo
m
b
ian
r
e
g
io
n
i
t
is
f
o
u
n
d
in
m
o
s
t
d
i
s
h
e
s
,
an
d
f
o
r
j
u
ice,
b
ec
au
s
e
it
i
s
th
e
liq
u
id
p
ar
t
o
f
lu
n
c
h
.
I
t
s
h
o
u
ld
b
e
n
o
ted
th
at
th
e
s
o
u
p
,
w
h
ic
h
is
also
an
ess
e
n
tial
ele
m
e
n
t
i
n
th
e
f
o
o
d
o
f
th
e
r
eg
io
n
,
is
n
o
t
tak
en
i
n
to
ac
co
u
n
t.
Fo
r
th
is
,
t
h
e
co
n
s
tr
u
c
tio
n
o
f
o
u
r
o
w
n
d
ataset
is
m
ad
e,
as
well
as
t
h
e
p
r
o
p
o
s
al
o
f
d
if
f
er
en
t
ar
ch
itect
u
r
es
a
n
d
th
eir
co
m
p
ar
is
o
n
w
i
th
t
h
e
VG
G
-
1
6
f
o
r
s
e
m
a
n
tic
s
eg
m
e
n
tati
o
n
.
Nex
t,
t
h
e
d
ev
elo
p
m
e
n
t o
f
t
h
e
w
o
r
k
is
e
x
p
o
s
ed
.
2
.
1
.
Da
t
a
ba
s
e
T
h
e
elab
o
r
ated
d
ataset
co
n
s
is
t
s
o
f
i
m
a
g
es
o
f
b
asic
f
o
o
d
d
is
h
es,
o
r
co
m
m
o
n
l
y
ca
lled
"
E
x
ec
u
ti
v
es"
i
n
th
e
r
eg
io
n
.
T
h
is
d
is
h
co
n
s
i
s
t
s
o
f
a
p
o
r
tio
n
o
f
r
ice,
a
t
y
p
e
o
f
g
r
ain
,
a
p
r
o
tein
,
a
p
o
r
tio
n
o
f
f
r
u
it,
a
g
las
s
o
f
j
u
ice,
an
d
m
o
s
tl
y
a
p
o
r
tio
n
o
f
s
alad
,
an
d
ar
e
ta
k
en
o
n
b
ac
k
g
r
o
u
n
d
s
w
it
h
s
i
m
p
le
an
d
co
m
p
le
x
te
x
t
u
r
es.
A
d
d
itio
n
a
ll
y
,
n
o
t
o
n
l
y
i
m
a
g
es
o
f
f
r
es
h
l
y
p
r
ep
ar
ed
d
is
h
es
ar
e
tak
e
n
,
b
u
t
a
s
th
e
p
er
s
o
n
co
n
s
u
m
es
it,
p
ict
u
r
es
o
f
it
ar
e
tak
e
n
,
in
cr
ea
s
i
n
g
its
co
m
p
le
x
it
y
f
o
r
r
ec
o
g
n
itio
n
,
b
ec
a
u
s
e
i
n
th
e
d
is
h
,
t
h
er
e
ar
e
r
esid
u
es
o
f
s
a
u
ce
s
a
n
d
th
e
f
o
o
d
p
o
r
tio
n
s
b
eg
in
to
b
e
s
ep
ar
at
ed
o
r
m
i
x
ed
.
T
h
e
p
ictu
r
es
ar
e
tak
en
i
n
t
w
o
r
esta
u
r
an
t
s
,
w
i
t
h
o
u
t
co
n
tr
o
l
o
f
th
e
li
g
h
ti
n
g
a
n
d
w
it
h
o
u
t
a
f
i
x
e
d
d
is
tan
ce
b
et
w
ee
n
th
e
p
late
a
n
d
th
e
ca
m
er
a,
b
u
t
m
o
s
t
l
y
tak
en
ar
o
u
n
d
5
5
c
m
o
f
d
is
tan
ce
.
T
h
ese
i
m
a
g
es
ar
e
ad
j
u
s
ted
to
a
s
ta
n
d
ar
d
s
ize
o
f
4
8
0
x
3
6
0
p
ix
els,
to
a
v
o
id
a
h
i
g
h
co
m
p
u
tatio
n
al
co
s
t
in
tr
ai
n
i
n
g
i
f
h
ig
h
e
r
r
eso
l
u
tio
n
s
ar
e
u
s
ed
.
E
ac
h
p
h
o
to
is
m
an
u
al
l
y
lab
eled
,
o
b
tain
in
g
a
t
o
tal
o
f
2
3
6
i
m
a
g
es,
w
h
er
e
2
0
0
ar
e
u
s
ed
f
o
r
tr
ain
in
g
an
d
3
6
f
o
r
v
alid
atio
n
o
f
n
e
t
w
o
r
k
s
.
A
n
e
x
a
m
p
le
o
f
t
h
e
d
ataset
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
1
alo
n
g
w
it
h
it
s
lab
elin
g
.
I
t sh
o
u
ld
b
e
n
o
ted
th
at
n
o
d
ata
au
g
m
e
n
tatio
n
h
a
s
b
ee
n
p
er
f
o
r
m
ed
.
2
.
2
.
P
ro
po
s
ed
a
rc
hite
ct
ures
I
n
o
r
d
er
to
ev
al
u
ate
t
h
e
f
o
o
d
s
eg
m
e
n
tat
io
n
ca
p
ac
it
y
w
ith
a
s
m
all
d
atab
ase
as
th
e
o
n
e
elab
o
r
ated
h
er
e,
d
if
f
er
e
n
t
ar
ch
itectu
r
e
s
o
f
C
N
Ns
w
er
e
p
r
o
p
o
s
ed
,
u
s
in
g
co
n
f
i
g
u
r
atio
n
s
s
u
c
h
as
Seg
Net
o
r
E
n
co
d
er
-
Dec
o
d
er
w
it
h
v
ar
iatio
n
in
d
ep
t
h
,
an
d
v
ar
ian
t
s
o
f
th
i
s
ar
ch
itect
u
r
e,
w
h
ic
h
co
n
tai
n
r
esid
u
al
la
y
er
s
.
T
h
is
w
as
d
o
n
e
in
o
r
d
er
to
o
b
s
er
v
e
th
eir
p
er
f
o
r
m
an
ce
a
n
d
q
u
alit
y
o
f
th
e
s
e
g
m
e
n
tat
io
n
o
f
ea
ch
c
ateg
o
r
y
.
E
ac
h
o
f
th
e
ar
ch
itect
u
r
es i
s
s
h
o
w
n
i
n
F
ig
u
r
e
2
an
d
ar
e
d
escr
ib
ed
b
elo
w
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
R
esS
eg
:
R
esid
u
a
l e
n
co
d
er
-
d
ec
o
d
er c
o
n
vo
l
u
tio
n
a
l
n
eu
r
a
l n
et
w
o
r
k
…
(
Ja
vier O
r
la
n
d
o
P
in
z
ó
n
A
r
en
a
s
)
1019
(
a)
(
b
)
Fig
u
r
e
1
.
E
x
a
m
p
les o
f
th
e
d
ata
s
et,
w
h
er
e
(
a)
a
r
ec
en
tl
y
p
r
ep
ar
ed
d
is
h
an
d
(
b
)
alr
ea
d
y
s
tar
te
d
to
b
e
ea
ten
,
w
it
h
t
h
eir
r
esp
ec
tiv
e
lab
el
s
T
h
e
f
ir
s
t
ar
ch
itect
u
r
e
is
a
Seg
Net
o
f
d
ep
th
3
(
o
r
Seg
Net
D3
)
,
i.e
.
it
is
f
o
r
m
ed
b
y
a
n
E
n
co
d
er
n
et
w
o
r
k
th
at
co
n
s
is
ts
o
f
3
s
et
s
o
f
la
y
er
s
.
E
ac
h
s
et
co
n
tai
n
s
2
g
r
o
u
p
s
o
f
co
n
v
o
lu
tio
n
,
i
n
o
th
er
w
o
r
d
s
,
a
g
r
o
u
p
is
co
m
p
o
s
ed
o
f
a
co
n
v
o
lu
tio
n
l
a
y
er
,
a
b
atch
n
o
r
m
aliza
t
io
n
l
a
y
er
an
d
a
R
e
L
U
la
y
er
.
At
t
h
e
en
d
o
f
ea
ch
s
et,
a
d
o
w
n
s
a
m
p
li
n
g
o
r
m
a
x
p
o
o
lin
g
la
y
er
i
s
ad
d
ed
.
T
h
e
E
n
co
d
er
is
f
o
llo
w
ed
b
y
a
Dec
o
d
er
n
et
w
o
r
k
t
h
at
is
b
asicall
y
a
m
ir
r
o
r
o
f
th
e
en
co
d
er
,
w
it
h
t
h
e
d
if
f
er
en
ce
th
a
t
in
s
tead
o
f
h
a
v
i
n
g
d
o
w
n
s
a
m
p
l
i
n
g
,
t
h
e
d
ec
o
d
er
h
as
u
p
s
a
m
p
l
in
g
la
y
er
s
ca
l
led
u
n
p
o
o
lin
g
la
y
er
.
I
n
o
r
d
er
to
i
m
p
r
o
v
e
th
e
d
elin
ea
tio
n
o
f
th
e
s
e
g
m
en
tatio
n
b
y
m
ea
n
s
o
f
t
h
e
r
ete
n
tio
n
o
f
t
h
e
d
et
ails
o
f
t
h
e
e
n
co
d
er
’
s
ea
r
l
y
la
y
er
s
,
t
h
e
i
n
d
ices
o
f
ea
c
h
p
o
o
lin
g
la
y
er
ar
e
in
ter
co
n
n
ec
ted
w
it
h
t
h
e
i
n
d
ic
es
o
f
t
h
eir
r
esp
ec
ti
v
e
m
ir
r
o
r
u
n
p
o
o
lin
g
la
y
er
,
as
d
e
f
i
n
ed
in
[1
6
]
.
T
h
e
s
ec
o
n
d
ar
ch
itect
u
r
e,
ca
lled
Seg
Net
D
4
,
h
as
a
s
i
m
ilar
co
n
f
i
g
u
r
atio
n
as
th
e
p
r
ev
io
u
s
o
n
e,
w
it
h
a
s
l
i
g
h
t
v
ar
iatio
n
i
n
its
d
ep
th
,
b
ein
g
o
f
4
,
th
at
is
,
w
it
h
an
ad
d
itio
n
al
s
et
o
f
la
y
er
s
in
b
o
th
th
e
e
n
co
d
er
an
d
d
ec
o
d
er
.
Fo
r
ar
ch
itectu
r
e
3
,
w
h
ic
h
is
a
m
o
d
i
f
icat
io
n
o
f
t
h
e
Seg
Net
D3
,
t
h
e
f
i
lter
s
’
s
ize
o
f
th
e
f
ir
s
t
s
et
o
f
t
h
e
en
co
d
e
r
an
d
th
e
last
s
et
o
f
th
e
d
ec
o
d
er
w
er
e
v
ar
ied
,
estab
lis
h
i
n
g
th
e
m
in
5
x
5
p
ix
el
s
q
u
ar
es,
s
o
th
at
th
e
n
e
t
w
o
r
k
co
u
ld
lear
n
in
a
b
etter
w
a
y
tex
tu
r
es a
n
d
in
ter
n
al
s
h
ap
es o
f
th
e
f
o
o
d
.
L
ik
e
w
is
e,
t
h
e
n
u
m
b
er
o
f
f
ilter
s
w
er
e
i
n
cr
ea
s
e
d
in
s
o
m
e
la
y
er
s
.
I
n
o
r
d
er
to
ex
p
lo
r
e
ar
ch
itectu
r
es
d
if
f
er
e
n
t
f
r
o
m
co
n
v
e
n
tio
n
al
o
n
es,
it
w
a
s
p
r
o
p
o
s
ed
to
u
s
e
r
esid
u
al
la
y
er
s
w
it
h
in
ea
ch
la
y
er
s
et.
A
s
r
ep
o
r
ted
in
[
2
4
]
,
th
e
ad
d
iti
o
n
o
f
th
i
s
t
y
p
e
o
f
la
y
er
i
m
p
r
o
v
es
th
e
ac
c
u
r
ac
y
o
f
th
e
n
et
w
o
r
k
i
n
co
n
j
u
n
ctio
n
with
th
e
q
u
alit
y
o
f
th
e
s
e
g
m
en
tatio
n
w
i
th
r
esp
ec
t
to
th
e
co
n
to
u
r
o
f
th
e
o
b
j
ec
t
th
an
k
s
to
th
e
tr
an
s
f
er
o
f
th
e
f
ea
tu
r
e
s
o
f
ea
r
l
y
la
y
er
s
to
d
ee
p
er
lay
er
s
.
Fo
r
th
i
s
r
ea
s
o
n
,
th
e
f
o
llo
w
i
n
g
ar
ch
itect
u
r
es,
ca
lled
E
-
R
e
s
id
u
al
v
.
1
a
n
d
v
.
2
,
co
n
s
is
t
o
f
a
Se
g
Net
o
f
d
ep
th
th
r
ee
w
i
th
s
et
s
o
f
la
y
er
s
co
m
p
o
s
ed
o
f
3
g
r
o
u
p
s
o
f
co
n
v
o
lu
tio
n
,
ad
d
in
g
r
es
id
u
al
la
y
er
s
in
ea
ch
s
et
o
f
t
h
e
e
n
co
d
er
.
Ho
w
e
v
er
,
d
u
e
to
th
e
in
ter
co
n
n
ec
ti
v
it
y
o
f
th
e
e
n
co
d
er
w
it
h
th
e
d
ec
o
d
er
,
th
e
r
e
s
id
u
al
la
y
er
i
n
p
u
t
is
ta
k
e
n
f
r
o
m
th
e
f
ir
s
t
g
r
o
u
p
o
f
co
n
v
o
lu
tio
n
o
f
th
e
s
e
t,
an
d
its
o
u
tp
u
t
is
co
n
n
ec
ted
to
th
e
p
o
o
lin
g
lay
er
i
n
p
u
t
.
T
h
e
c
o
n
n
ec
tio
n
s
ca
n
b
e
o
b
s
er
v
ed
m
o
r
e
g
r
ap
h
icall
y
i
n
Fig
u
r
e
3
,
w
h
er
e
t
h
e
o
u
tp
u
t
o
f
t
h
e
co
n
n
ec
tio
n
is
th
e
R
e
L
U
f
u
n
ctio
n
.
I
n
t
h
is
f
i
g
u
r
e,
“
co
n
v
”
r
ef
er
s
to
a
c
o
n
v
o
l
u
tio
n
la
y
er
an
d
“
B
-
n
o
r
m
”
to
a
b
atch
n
o
r
m
aliza
t
io
n
la
y
er
.
T
h
e
m
ai
n
d
if
f
er
e
n
ce
b
et
w
ee
n
t
h
ese
t
w
o
n
et
w
o
r
k
s
is
t
h
e
n
u
m
b
er
o
f
f
ilt
er
s
u
s
ed
p
er
la
y
er
.
T
o
f
u
r
th
er
s
tr
e
n
g
th
e
n
t
h
e
p
r
ev
io
u
s
ar
ch
i
tectu
r
es,
r
esid
u
al
co
n
n
e
ctio
n
s
ar
e
ad
d
ed
in
th
e
d
ec
o
d
er
.
T
h
is
v
ar
iatio
n
is
n
a
m
ed
R
e
s
id
u
al
Se
g
m
e
n
tatio
n
C
o
n
v
o
l
u
tio
n
al
Neu
r
al
Net
w
o
r
k
o
r
R
esS
e
g
,
w
h
ich
h
as
a
d
ep
th
o
f
3
,
o
r
R
esS
eg
D3
.
Fi
n
all
y
,
t
h
e
last
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
is
a
R
es
Seg
w
i
th
a
d
ep
th
o
f
5
s
ets
(
R
e
s
Se
g
D5
)
.
I
ts
m
ai
n
c
h
ar
ac
ter
is
tic
th
at,
i
n
th
e
f
ir
s
t
s
et
o
f
th
e
e
n
co
d
er
an
d
last
o
f
t
h
e
d
ec
o
d
er
,
r
esid
u
al
b
r
an
ch
es
ar
e
n
o
t
u
s
ed
,
an
d
f
ilter
s
o
f
s
ize
3
ar
e
u
s
ed
in
t
h
e
f
ir
s
t
co
n
v
o
l
u
tio
n
l
a
y
er
,
ex
ce
p
t
in
s
tag
e
4
,
b
ec
au
s
e
it
is
n
ec
ess
ar
y
to
ad
j
u
s
t th
e
o
u
tp
u
t v
o
l
u
m
e
to
a
v
o
i
d
lo
s
s
o
f
in
f
o
r
m
atio
n
,
d
u
e
to
th
e
s
ize
o
f
t
h
e
i
m
ag
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
0
:
1
0
1
7
-
1
0
2
6
1020
I
t
s
h
o
u
ld
b
e
n
o
ted
th
a
t
f
o
r
th
e
co
n
v
o
l
u
tio
n
la
y
er
s
w
it
h
3
x
3
f
ilter
s
ize
in
ev
er
y
ar
ch
itect
u
r
e
,
p
ad
d
in
g
o
f
1
is
u
s
ed
,
in
s
u
ch
a
w
a
y
t
h
at
t
h
e
s
ize
o
f
t
h
e
i
n
p
u
t
v
o
l
u
m
e
o
f
t
h
e
la
y
er
is
m
a
in
ta
in
ed
.
T
h
e
n
et
w
o
r
k
w
ith
w
h
ic
h
it
w
ill
b
e
co
m
p
ar
ed
to
th
e
p
er
f
o
r
m
a
n
ce
o
f
all
p
r
o
p
o
s
ed
ar
ch
itectu
r
es
is
a
S
eg
Net
v
er
s
io
n
o
f
th
e
VGG
-
16
[2
5
]
,
s
in
ce
it
s
d
ep
th
,
an
d
i
n
g
en
er
al,
it
s
ar
ch
i
tectu
r
e
is
s
i
m
ilar
to
t
h
e
p
r
o
p
o
s
ed
o
n
es,
b
u
t
w
i
th
a
g
r
ea
ter
n
u
m
b
er
o
f
co
n
v
o
lu
ti
o
n
la
y
er
s
a
n
d
lear
n
i
n
g
f
ilter
s
.
Fig
u
r
e
2
.
P
r
o
p
o
s
ed
ar
ch
itectu
r
es
In
p
u
t
3x3
,
64
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
64
3x3
,
64
Softm
ax
3x3
,
64
3x
3,
64
Poo
l
/
2
3x3
,
64
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x
3,
64
3x3
,
64
Unp
o
o
l
x2
3x
3,
64
3x3
,
6
In
p
u
t
3x3
,
64
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
64
3x3
,
64
Softm
ax
3x3
,
64
3x
3,
64
Poo
l
/
2
3x3
,
64
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
64
3x3
,
64
Unp
o
o
l
x2
3x3
,
64
3x3
,
6
3x3
,
64
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
64
3x3
,
64
In
p
u
t
5x5
,
64
5x5
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
256
3x3
,
128
Softm
ax
3x3
,
64
3x
3,
12
8
Poo
l
/
2
3x3
,
128
3x3
,
256
Poo
l
/
2
Unp
o
o
l
x2
3x
3,
12
8
3x3
,
64
Unp
o
o
l
x2
5x
5,
64
5x5
,
6
In
p
u
t
3x3
,
128
1x1
,
64
Poo
l
/
2
Unp
o
o
l
x2
1x1
,
512
3x3
,
512
Softm
ax
3x3
,
128
1x1
,
128
3x3
,
256
Poo
l
/
2
1x1
,
128
1x1
,
128
3x3
,
512
Poo
l
/
2
1x1
,
512
1x1
,
128
Unp
o
o
l
x2
1x1
,
128
3x3
,
256
1x1
,
128
Unp
o
o
l
x2
3x3
,
128
1x1
,
128
3x3
,
6
1x1
,
128
1x1
,
512
In
p
u
t
3x3
,
64
1x1
,
32
Poo
l
/
2
Unp
o
o
l
x2
1x1
,
256
3x3
,
256
Softm
ax
3x3
,
64
1x1
,
64
3x3
,
128
Poo
l
/
2
1x1
,
64
1x1
,
64
3x3
,
256
Poo
l
/
2
1x1
,
256
1x1
,
64
Unp
o
o
l
x2
1x1
,
64
3x3
,
128
1x1
,
64
Unp
o
o
l
x2
3x3
,
64
1x1
,
64
3x3
,
6
1x1
,
64
1x1
,
256
Se
gN
e
t
D3
Se
gN
e
t
D4
Se
gN
e
t
D3
M
o
d
if
ied
Se
gN
e
t
D
3
E
-
R
esid
u
al
v
.1
Se
gN
e
t
D3
E
-
R
esid
u
al
v
.2
In
p
u
t
3x3
,
64
1x1
,
32
Poo
l
/
2
Unp
o
o
l
x2
1x1
,
256
3x3
,
256
Softm
ax
3x3
,
64
1x1
,
64
3x3
,
128
Poo
l
/
2
1x1
,
64
1x1
,
64
3x3
,
256
Poo
l
/
2
1x1
,
256
1x1
,
64
Unp
o
o
l
x2
1x1
,
64
3x3
,
128
1x1
,
64
Unp
o
o
l
x2
3x3
,
64
1x1
,
64
3x3
,
6
1x
1,
64
1x1
,
256
1x1
,
64
1x1
,
256
In
p
u
t
3x3
,
32
3x3
,
64
Poo
l
/
2
U
n
p
oo
l
x2
3x3
,
128
3x3
,
128
Softm
ax
3x3
,
64
3x3
,
64
3x3
,
64
Poo
l
/
2
1x1
,
128
3x3
,
128
3x3
,
128
Poo
l
/
2
1x1
,
128
1x1
,
128
1x1
,
128
4x3
,
128
3x3
,
128
Poo
l
/
2
1x1
,
128
3x3
,
128
3x3
,
128
Poo
l
/
2
1x1
,
128
Unp
o
o
l
x2
3x3
,
128
3x3
,
128
2x3
,
128
Unp
o
o
l
x2
3x3
,
128
3x3
,
128
1x1
,
128
Unp
o
o
l
x2
3x3
,
64
3x3
,
64
1x1
,
64
Unp
o
o
l
x2
3x3
,
6
3x3
,
64
3x3
,
64
1x1
,
128
1x1
,
128
1x1
,
64
In
p
u
t
3x3
,
32
3x3
,
64
Poo
l
/
2
Unp
o
o
l
x2
3x3
,
512
3x3
,
512
Softm
ax
3x
3,
12
8
3x3
,
128
Poo
l
/
2
3x3
,
256
3x
3,
25
6
Poo
l
/
2
3x3
,
256
3x3
,
512
3x3
,
512
3x3
,
512
Poo
l
/
2
3x3
,
512
3x3
,
512
3x3
,
512
Poo
l
/
2
3x3
,
512
Unp
o
o
l
x2
3x3
,
256
3x3
,
512
3x3
,
512
Unp
o
o
l
x2
3x3
,
128
3x3
,
256
3x3
,
256
Unp
o
o
l
x2
3x3
,
64
3x3
,
128
U
n
p
oo
l
x2
3x3
,
6
3x3
,
64
3x3
,
256
3x3
,
256
3x3
,
512
3x3
,
512
3x3
,
512
3x3
,
512
Se
gN
e
t
V
GG16
R
esSe
g
D5
R
esSe
g
D3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
R
esS
eg
:
R
esid
u
a
l e
n
co
d
er
-
d
ec
o
d
er c
o
n
vo
l
u
tio
n
a
l
n
eu
r
a
l n
et
w
o
r
k
…
(
Ja
vier O
r
la
n
d
o
P
in
z
ó
n
A
r
en
a
s
)
1021
Fig
u
r
e
3
.
R
esid
u
al
co
n
n
ec
t
io
n
s
tr
u
ct
u
r
e
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
NS
3
.
1
.
P
er
f
o
r
m
a
nce
w
it
ho
ut
la
beled ba
c
k
g
ro
un
d
Fo
r
th
e
p
u
r
p
o
s
e
o
f
co
m
p
ar
in
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
ar
ch
ite
ctu
r
es,
id
en
tical
tr
ai
n
i
n
g
p
ar
a
m
eter
s
ar
e
s
et,
s
o
th
at
t
h
e
y
ca
n
lear
n
u
n
d
er
th
e
s
a
m
e
co
n
d
itio
n
s
.
T
ak
in
g
in
to
ac
co
u
n
t t
h
e
ab
o
v
e,
th
e
i
n
itial lea
r
n
i
n
g
r
ate
i
s
s
et
to
1
0
-
4
w
ith
a
d
r
o
p
f
ac
to
r
o
f
0
.
5
p
er
1
0
0
e
p
o
ch
s
,
u
s
i
n
g
a
b
atch
s
ize
o
f
2
an
d
a
to
t
al
o
f
4
0
0
ep
o
ch
s
o
f
tr
ain
i
n
g
.
T
h
o
s
e
p
ar
a
m
eter
s
wer
e
s
et
b
y
d
o
in
g
iter
ati
v
e
test
s
,
w
i
th
w
h
ic
h
b
etter
r
esu
lt
s
wer
e
o
b
tain
ed
d
u
r
in
g
th
e
tr
ai
n
in
g
.
W
it
h
th
is
,
t
h
e
b
eh
av
io
r
s
s
h
o
w
n
i
n
Fi
g
u
r
e
4
ar
e
o
b
tain
ed
,
w
h
er
e
n
o
n
et
w
o
r
k
s
u
r
p
as
s
es
an
ac
cu
r
ac
y
o
f
5
0
%,
ev
en
h
a
v
i
n
g
t
h
e
VGG1
6
as t
h
e
w
o
r
s
t p
er
f
o
r
m
er
,
w
ith
3
0
% a
cc
u
r
ac
y
.
Fig
u
r
e
4
.
T
r
ain
in
g
b
eh
a
v
io
r
o
f
ea
ch
ar
ch
itect
u
r
e,
u
s
i
n
g
a
d
ata
b
ase
w
it
h
o
u
t la
b
eled
b
ac
k
g
r
o
u
n
d
Du
e
to
th
e
lo
w
p
er
f
o
r
m
an
ce
o
f
t
h
e
n
et
w
o
r
k
s
d
u
r
in
g
t
h
eir
t
r
ain
in
g
,
it
i
s
n
ec
es
s
ar
y
to
o
b
s
er
v
e
w
h
at
th
e
y
lear
n
ed
to
u
n
d
er
s
tan
d
t
h
eir
b
eh
av
io
r
s
.
Fo
r
th
i
s
,
test
s
ar
e
m
ad
e
w
ith
te
s
t
i
m
a
g
es,
o
b
tain
in
g
r
es
u
lt
s
li
k
e
th
e
o
n
e
in
Fi
g
u
r
e
5
,
w
h
er
e,
d
esp
ite
b
ein
g
ab
le
to
s
e
g
m
en
t
ea
ch
t
y
p
e
o
f
f
o
o
d
,
th
e
y
ar
e
n
o
t
ab
le
to
eli
m
i
n
ate
th
e
b
ac
k
g
r
o
u
n
d
o
r
p
ar
ts
o
f
th
e
d
is
h
es,
ca
u
s
in
g
lar
g
e
a
m
o
u
n
ts
o
f
f
alse
p
o
s
iti
v
es
to
b
e
g
en
er
ated
f
r
o
m
all
f
o
o
d
s
,
w
h
ic
h
m
a
k
es t
h
e
n
et
w
o
r
k
i
n
ef
f
icie
n
t.
Fig
u
r
e
5
.
C
o
m
p
ar
is
o
n
b
et
w
ee
n
g
r
o
u
n
d
tr
u
t
h
an
d
s
e
g
m
e
n
tati
o
n
o
b
tain
ed
f
r
o
m
R
e
s
Se
g
D3
Co
n
v
R
e
LU
Co
n
v
B
-
N
o
r
m
B
-
N
o
r
m
Re
LU
Co
n
v
B
-
N
o
r
m
R
e
LU
Co
n
v
B
-
N
o
r
m
+
CON
V
CON
V
CON
V
R
e
s CON
V
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
0
:
1
0
1
7
-
1
0
2
6
102
2
3
.
2
.
P
er
f
o
r
m
a
nce
w
it
h la
beled b
a
ck
g
ro
un
d
T
o
s
o
lv
e
th
e
ab
o
v
e
p
r
o
b
le
m
,
it
is
p
r
o
p
o
s
ed
to
ad
d
an
ad
d
itio
n
al
ca
te
g
o
r
y
ca
lled
"
B
ac
k
g
r
o
u
n
d
"
,
w
h
er
e
ea
ch
u
n
lab
eled
p
ar
t
o
f
th
e
i
m
ag
e
b
ec
o
m
e
s
p
ar
t
o
f
th
is
ca
teg
o
r
y
,
g
en
er
ati
n
g
i
m
ag
e
s
lik
e
t
h
e
ex
a
m
p
l
e
s
h
o
w
n
i
n
Fi
g
u
r
e
6
.
W
ith
t
h
i
s
m
o
d
if
ica
tio
n
,
it
is
p
r
o
ce
ed
ed
to
p
er
f
o
r
m
t
h
e
tr
ai
n
in
g
o
f
e
ac
h
n
et
w
o
r
k
a
g
ai
n
.
H
o
w
e
v
er
,
a
s
l
ig
h
t
m
o
d
if
icati
o
n
is
m
ad
e
i
n
th
e
lear
n
i
n
g
r
a
te
p
ar
a
m
et
er
,
u
s
i
n
g
an
in
i
tial
v
al
u
e
o
f
10
-
3
,
w
it
h
a
d
r
o
p
r
atio
o
f
0
.
5
p
er
1
5
0
ep
o
ch
s
,
i
n
s
u
c
h
a
w
a
y
t
h
at
th
e
n
et
w
o
r
k
s
s
tar
t
w
it
h
a
r
ap
id
le
ar
n
in
g
p
r
o
ce
s
s
,
a
n
d
th
en
f
i
n
e
-
tu
n
i
n
g
th
e
p
ar
a
m
eter
s
lear
n
ed
ev
er
y
ce
r
tai
n
n
u
m
b
e
r
o
f
ep
o
ch
s
.
T
h
is
m
o
d
i
f
icatio
n
w
as
d
ec
id
ed
af
ter
lo
o
k
in
g
at
t
h
e
i
n
itial
b
e
h
av
io
r
o
f
th
e
n
et
w
o
r
k
s
co
m
p
ar
ed
to
th
e
f
ir
s
t
lear
n
in
g
r
ate
u
s
e
d
,
s
ee
in
g
th
at
w
it
h
a
s
m
al
ler
lear
n
i
n
g
r
ate,
t
h
e
n
et
w
o
r
k
s
te
n
d
ed
to
o
b
tain
ac
cu
r
a
cies lo
w
er
th
a
n
7
0
% a
f
ter
2
0
0
ep
o
ch
s
.
W
ith
t
h
ese
p
ar
am
eter
s
,
a
tr
ai
n
in
g
b
e
h
av
i
o
r
lik
e
t
h
e
o
n
e
s
h
o
w
n
in
Fi
g
u
r
e
7
is
o
b
tain
ed
,
h
av
i
n
g
as
th
e
t
w
o
b
est
n
et
w
o
r
k
s
th
e
R
esS
e
g
o
f
d
ep
th
5
an
d
th
e
Se
g
Net
w
it
h
VGG
-
1
6
ar
ch
itect
u
r
e,
w
ith
9
4
.
1
%
an
d
9
5
.
5
%
ac
cu
r
ac
y
,
r
esp
ec
tiv
el
y
,
s
u
r
p
ass
i
n
g
b
y
m
o
r
e
th
an
3
%
t
h
e
m
o
d
i
f
ied
Se
g
Net
D3
,
w
h
ich
w
as
th
e
o
n
l
y
o
n
e
to
ac
h
iev
e
m
o
r
e
th
an
9
0
% a
m
o
n
g
th
e
r
e
m
ai
n
in
g
n
et
w
o
r
k
s
.
Fig
u
r
e
6
.
Gr
o
u
n
d
tr
u
t
h
w
it
h
th
e
a
d
d
itio
n
al
ca
teg
o
r
y
“
B
ac
k
g
r
o
u
n
d
”
Fig
u
r
e
7
.
T
r
ain
in
g
b
eh
a
v
io
r
o
f
ea
ch
ar
ch
itect
u
r
e,
u
s
i
n
g
a
d
atab
ase
w
it
h
lab
eled
b
ac
k
g
r
o
u
n
d
3
.
3
.
T
est
perf
o
r
m
a
nce
o
f
f
ina
l t
ra
ined a
rc
hite
ct
ures
Sin
ce
th
e
ar
ch
itect
u
r
es
h
ad
b
etter
b
eh
av
io
r
d
u
r
i
n
g
t
h
eir
tr
ain
i
n
g
,
it
i
s
p
r
o
ce
ed
ed
to
v
er
if
y
th
e
ir
p
er
f
o
r
m
a
n
ce
w
it
h
t
h
e
tes
t
i
m
a
g
es.
T
ab
le
1
s
h
o
w
s
th
e
r
es
u
lts
o
b
tai
n
ed
w
i
th
ea
ch
ar
ch
itect
u
r
e.
T
h
e
ar
ch
itectu
r
es
w
i
th
t
h
e
lo
west p
er
f
o
r
m
a
n
ce
w
er
e
th
e
Se
g
Net
s
to
w
h
ic
h
t
h
e
r
esid
u
al
la
y
er
s
w
er
e
ad
d
ed
o
n
l
y
in
th
e
e
n
co
d
er
s
tag
e.
On
th
e
o
th
er
h
an
d
,
th
er
e
ar
e
n
et
w
o
r
k
s
w
i
th
p
er
ce
n
ta
g
es
o
f
ac
cu
r
ac
y
g
r
ea
ter
th
a
n
9
0
%,
w
it
h
Se
g
Net
D4
,
m
o
d
if
ied
D3
,
an
d
R
e
s
Se
g
D5
,
w
i
th
s
l
ig
h
t
d
if
f
er
e
n
ce
s
.
H
o
w
e
v
er
,
a
n
i
m
p
o
r
tan
t
f
ac
to
r
f
o
r
it
s
ev
alu
a
tio
n
is
p
r
ese
n
ted
i
n
t
h
e
in
ter
ce
p
t
o
v
er
u
n
io
n
(
I
o
U)
,
w
h
er
e
t
h
e
R
esS
e
g
o
b
tai
n
ed
a
h
i
g
h
er
r
ela
tio
n
,
i.e
.
h
o
w
w
ell
t
h
e
n
et
w
o
r
k
c
lass
if
ies
t
h
e
p
ix
el
s
o
f
th
e
c
lass
e
s
,
g
i
v
en
b
y
(
1
)
,
w
h
er
e
T
P
ar
e
th
e
tr
u
e
p
o
s
iti
v
es,
FP
th
e
f
alse
p
o
s
it
iv
e
s
an
d
t
h
e
FN
t
h
e
f
alse
n
e
g
ati
v
es,
b
ein
g
t
h
e
v
a
lu
e
s
h
o
w
n
i
n
t
h
e
tab
l
e
th
e
a
v
er
ag
e
o
f
al
l
th
e
clas
s
es o
f
t
h
e
ev
a
lu
ated
d
a
taset.
T
ab
le
1
.
A
r
ch
itect
u
r
es p
er
f
o
r
m
an
ce
w
i
th
t
h
e
te
s
t d
atab
ase
N
e
t
w
o
r
k
S
e
g
N
e
t
D3
S
e
g
N
e
t
D4
S
e
g
N
e
t
D
3
M
S
e
g
N
e
t
D
3
E
-
R
v
.
1
S
e
g
N
e
t
D
3
E
-
R
v
.
2
R
e
sS
e
g
D3
R
e
sS
e
g
D5
S
e
g
N
e
t
V
G
G
1
6
Me
a
n
Ac
c
u
r
a
c
y
(
%
)
8
9
.
6
1
9
0
.
0
6
9
0
.
3
5
8
7
.
8
5
8
6
.
7
1
8
8
.
1
7
9
0
.
4
3
9
4
.
8
6
Me
a
n
I
o
U
0
.
6
3
5
0
.
6
6
3
0
.
6
6
1
0
.
6
3
8
0
.
6
3
1
0
.
6
4
2
0
.
7
5
6
0
.
8
2
1
Me
a
n
BF
S
c
o
r
e
0
.
5
4
7
0
.
5
6
8
0
.
5
9
0
0
.
5
6
7
0
.
5
5
4
0
.
5
6
7
0
.
7
0
7
0
.
7
6
8
Me
a
n
Pr
o
c
e
ss
i
n
g
T
i
m
e
(
m
s)
1
6
3
.
9
1
8
3
.
3
2
6
0
.
3
3
1
7
.
6
2
0
3
.
3
2
1
0
.
0
2
6
4
.
1
3
5
4
.
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
R
esS
eg
:
R
esid
u
a
l e
n
co
d
er
-
d
ec
o
d
er c
o
n
vo
l
u
tio
n
a
l
n
eu
r
a
l n
et
w
o
r
k
…
(
Ja
vier O
r
la
n
d
o
P
in
z
ó
n
A
r
en
a
s
)
1023
(
1
)
Ma
k
i
n
g
a
co
m
p
ar
is
o
n
b
et
w
e
en
th
e
t
w
o
p
r
o
p
o
s
ed
R
esS
eg
ar
ch
itectu
r
e
s
,
it
ca
n
b
e
s
ee
n
th
at,
w
it
h
th
e
in
cr
ea
s
e
in
t
h
e
d
ep
th
o
f
th
e
n
et
w
o
r
k
,
th
e
a
v
er
ag
e
ac
c
u
r
ac
y
i
n
cr
ea
s
ed
b
y
2
.
2
6
%.
L
ik
e
w
i
s
e,
th
e
m
I
o
U
o
b
tain
ed
a
s
i
g
n
if
ica
n
t
i
m
p
r
o
v
e
m
e
n
t,
in
cr
ea
s
in
g
b
y
0
.
1
1
4
,
lik
e
th
e
p
ar
a
m
eter
B
F
(
b
o
u
n
d
ar
y
F1
)
w
h
ic
h
i
s
d
ef
in
ed
a
s
t
h
e
p
r
ec
is
io
n
i
n
t
h
e
alig
n
m
en
t
o
f
th
e
p
r
ed
icted
b
o
u
n
d
ar
y
w
it
h
r
esp
ec
t
to
t
h
e
g
r
o
u
n
d
tr
u
t
h
,
g
r
o
w
i
n
g
b
y
0
.
1
4
,
o
b
s
er
v
in
g
an
in
cr
ea
s
e
in
t
h
e
p
r
o
ce
s
s
i
n
g
t
i
m
e
o
f
o
n
l
y
5
0
m
s
.
W
it
h
t
h
is
,
it
c
a
n
b
e
in
f
er
r
ed
th
at
b
y
in
cr
ea
s
i
n
g
t
h
e
d
ep
th
to
a
ce
r
tai
n
d
eg
r
ee
,
it c
a
n
b
e
ea
s
il
y
r
ea
c
h
ed
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
Se
g
Net
VGG1
6
,
ev
e
n
b
ein
g
ab
le
to
h
a
v
e
a
b
etter
p
r
o
ce
s
s
i
n
g
ti
m
e.
A
co
m
p
ar
i
s
o
n
o
f
th
e
ar
c
h
itec
tu
r
es
i
n
d
if
f
er
e
n
t
d
is
h
es
a
n
d
en
v
ir
o
n
m
e
n
t
s
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
8
,
w
h
er
e
th
er
e
ar
e
f
r
esh
l
y
s
er
v
e
d
d
is
h
es,
s
tar
ted
to
b
e
ea
ten
a
n
d
f
in
is
h
ed
.
Si
m
ilar
l
y
,
tab
les
w
it
h
f
lat
co
lo
r
an
d
co
m
p
le
x
tex
tu
r
es
ar
e
u
s
ed
in
s
u
ch
a
w
a
y
t
h
at
it
i
s
m
o
r
e
d
if
f
i
cu
lt
to
d
if
f
er
e
n
tiate
it
f
r
o
m
f
o
o
d
.
W
h
en
th
e
d
is
h
es
ar
e
f
r
esh
l
y
s
er
v
ed
,
s
u
c
h
as
n
u
m
b
er
4
,
t
h
e
n
et
w
o
r
k
s
ten
d
to
g
en
er
ate
a
g
o
o
d
class
i
f
icat
io
n
an
d
s
e
g
m
e
n
tatio
n
o
f
th
e
f
o
o
d
,
alth
o
u
g
h
t
h
e
y
h
av
e
p
ar
ts
o
f
th
e
tab
le
th
at
ar
e
r
ec
o
g
n
ized
as
s
o
m
e
t
y
p
e
o
f
f
o
o
d
,
ex
ce
p
t
R
esS
e
g
D5
an
d
VGG1
6
.
I
n
th
e
s
a
m
e
w
a
y
,
w
h
e
n
th
e
p
late
is
al
m
o
s
t
e
m
p
t
y
(
co
lu
m
n
5
)
,
th
e
y
g
e
n
er
ate
a
g
o
o
d
class
i
f
icatio
n
,
ev
en
o
f
p
ar
ts
n
o
t
lab
eled
in
th
e
g
r
o
u
n
d
tr
u
th
,
alth
o
u
g
h
th
e
y
also
m
a
n
a
g
e
to
lab
el
lef
to
v
er
s
o
f
s
au
ce
lo
ca
ted
in
t
h
e
u
p
p
er
p
ar
t
o
f
th
e
p
late.
I
n
th
e
f
ir
s
t
co
lu
m
n
,
t
h
e
n
et
w
o
r
k
s
te
n
d
to
b
e
w
r
o
n
g
m
a
in
l
y
i
n
th
e
s
eg
m
e
n
tatio
n
o
f
t
h
e
j
u
ice
,
m
ai
n
l
y
b
y
it
s
co
lo
r
,
w
it
h
t
h
e
e
x
ce
p
tio
n
o
f
th
e
las
t
t
w
o
ar
ch
itect
u
r
es,
e
v
en
th
o
u
g
h
p
ar
t
o
f
th
e
m
ea
t
s
au
ce
w
a
s
lab
eled
as
m
ea
t
i
n
th
e
g
r
o
u
n
d
tr
u
th
,
t
h
ese
t
w
o
w
er
e
ab
le
to
id
en
tify
w
h
ic
h
w
a
s
th
e
p
r
o
tein
a
n
d
s
ep
ar
ate
it f
r
o
m
th
e
s
au
ce
.
Fig
u
r
e
8
.
Dif
f
er
en
t te
s
ts
o
f
t
h
e
ar
ch
itectu
r
e
s
w
it
h
th
e
te
s
t d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
0
:
1
0
1
7
-
1
0
2
6
1024
T
h
e
m
o
s
t
co
m
p
lex
i
m
ag
e
is
lo
ca
ted
in
co
l
u
m
n
2
,
w
h
er
e
th
er
e
is
a
p
late
t
h
at
h
as
alr
ea
d
y
b
e
en
s
tar
ted
an
d
a
b
ac
k
g
r
o
u
n
d
w
i
th
a
co
m
p
lex
tex
tu
r
e.
I
n
th
i
s
,
t
h
e
f
ir
s
t
6
ar
ch
itectu
r
es
g
e
n
er
ate
a
lar
g
e
n
u
m
b
er
o
f
f
alse
p
o
s
itiv
es,
esp
ec
iall
y
p
r
o
tein
d
u
e
to
th
e
d
ar
k
co
lo
r
o
f
th
e
tex
tu
r
e.
T
h
e
R
esS
eg
D5
n
et
w
o
r
k
,
alt
h
o
u
g
h
it
s
a
m
o
u
n
t
o
f
f
al
s
e
p
o
s
iti
v
es
i
s
la
r
g
el
y
le
s
s
t
h
a
n
o
th
er
n
et
w
o
r
k
s
,
ten
d
s
to
b
e
w
r
o
n
g
w
h
er
e
t
h
e
ca
n
d
y
i
s
(
w
h
ic
h
i
s
n
o
t p
ar
t o
f
a
n
y
ca
te
g
o
r
y
)
.
On
t
h
e
o
th
er
h
a
n
d
,
VGG1
6
w
as a
b
le
to
d
is
cr
i
m
in
ate
th
is
a
n
d
av
o
id
in
lar
g
e
q
u
a
n
tit
y
f
alse
p
o
s
iti
v
es
ca
u
s
ed
b
y
t
h
e
e
n
v
ir
o
n
m
e
n
t.
I
n
g
en
er
al
ter
m
s
,
th
e
t
w
o
b
e
s
t
ar
ch
itect
u
r
es
ar
e
t
h
e
R
e
s
Se
g
D5
an
d
Seg
Net
V
GG1
6
,
w
h
ic
h
ar
e
ab
le
to
eli
m
i
n
ate
t
h
e
n
o
i
s
e
th
a
t
d
o
es
n
o
t
b
elo
n
g
to
t
h
e
d
is
h
to
a
lar
g
e
ex
te
n
t
a
n
d
to
s
eg
m
e
n
t
w
it
h
g
o
o
d
p
r
ec
is
io
n
t
h
e
t
y
p
es o
f
f
o
o
d
.
On
t
h
e
o
th
er
h
a
n
d
,
alth
o
u
g
h
t
h
e
VGG1
6
h
a
s
b
etter
o
v
er
all
p
er
f
o
r
m
an
ce
,
w
h
e
n
m
a
k
i
n
g
th
e
co
m
p
ar
i
s
o
n
b
et
w
ee
n
th
is
n
et
w
o
r
k
a
n
d
t
h
e
R
es
Se
g
D5
,
e
s
p
ec
iall
y
w
it
h
f
r
es
h
l
y
s
er
v
ed
d
is
h
es,
th
e
R
e
s
Se
g
h
a
s
a
b
etter
b
eh
av
io
r
r
eg
ar
d
in
g
th
e
d
eli
n
ea
ti
o
n
o
f
th
e
co
n
to
u
r
s
o
f
t
h
e
s
eg
m
e
n
ted
s
ec
t
io
n
s
,
as
s
h
o
w
n
in
Fi
g
u
r
e
9
.
T
h
e
R
es
Seg
i
s
ab
le
to
s
eg
m
e
n
t
e
m
p
t
y
s
p
ac
es
b
et
w
ee
n
p
ar
ts
o
f
t
h
e
s
a
m
e
f
o
o
d
,
as
ca
n
b
e
s
ee
n
in
t
h
e
le
f
t
p
ar
t
o
f
th
e
d
is
h
,
w
h
er
e
a
n
o
o
d
le
lea
v
es
an
e
m
p
ty
s
p
ac
e,
w
h
ic
h
is
f
i
lled
b
y
t
h
e
VG
G1
6
.
I
t
ev
e
n
h
as
a
b
etter
s
e
g
m
en
ta
tio
n
o
f
t
h
e
v
e
g
etab
le
ca
te
g
o
r
y
.
T
h
is
al
lo
w
s
y
o
u
to
h
a
v
e
a
b
etter
id
ea
o
f
th
e
a
m
o
u
n
t
o
f
f
o
o
d
th
at
is
ac
t
u
all
y
o
n
th
e
p
la
te.
Fig
u
r
e
9
.
C
o
m
p
ar
is
o
n
o
f
b
o
u
n
d
ar
y
s
e
g
m
en
tatio
n
p
er
f
o
r
m
a
n
c
e
b
et
w
ee
n
R
es
Seg
D5
a
n
d
VGG1
6
3
.
4
.
Ca
s
es o
f
w
ro
ng
s
eg
m
ent
a
t
io
n f
o
r
ResSeg
D5
T
h
e
R
esS
e
g
D5
,
alth
o
u
g
h
it
h
as
a
g
o
o
d
b
eh
av
io
r
esp
ec
iall
y
w
h
e
n
t
h
er
e
i
s
f
o
o
d
s
till
o
n
th
e
p
late,
h
as
a
d
if
f
ic
u
lt
y
w
h
e
n
t
h
er
e
is
n
o
f
o
o
d
b
u
t
th
er
e
ar
e
r
esid
u
es
o
f
s
au
ce
s
th
at
ca
n
s
i
m
u
late
t
h
e
tex
tu
r
e
o
f
t
h
e
r
ice
b
y
th
e
d
i
v
is
io
n
an
d
co
lo
r
o
f
t
h
e
d
is
h
,
ca
u
s
in
g
it
to
g
e
n
er
ate
a
co
n
f
u
s
io
n
o
f
q
u
an
ti
t
y
o
f
f
o
o
d
.
I
t
ca
n
b
e
s
ee
n
i
n
Fig
u
r
e
1
0
a,
w
h
er
e
ac
co
r
d
in
g
to
th
eir
ac
ti
v
atio
n
s
,
t
h
e
n
e
t
w
o
r
k
ac
tiv
ate
s
ar
ea
s
m
ai
n
l
y
w
h
er
e
t
h
e
s
a
u
ce
is
lo
ca
ted
.
T
h
is
ca
n
b
e
ca
u
s
ed
b
y
co
n
v
o
l
u
tio
n
s
f
r
o
m
p
i
x
el
to
p
ix
el
(
f
ilter
s
o
f
s
ize
1
)
,
s
in
ce
th
e
y
ca
n
d
iv
er
t
th
eir
lear
n
in
g
to
w
ar
d
s
ac
ti
v
e
b
u
t
lit
tle
r
elev
a
n
t
p
ar
ts
,
tak
in
g
i
n
to
ac
co
u
n
t
th
at
p
r
io
r
to
th
e
s
e
t
h
e
r
e
ar
e
o
n
l
y
2
la
y
er
s
o
f
co
n
v
o
l
u
tio
n
an
d
d
o
n
o
t
u
s
e
a
lar
g
e
n
u
m
b
er
o
f
f
ilter
s
i
n
th
e
la
y
er
s
,
co
n
s
id
er
in
g
th
e
n
u
m
b
er
o
f
p
o
s
s
ib
le
d
etails th
at
m
a
y
b
e
in
th
e
e
m
p
t
y
p
lates.
An
o
th
er
s
i
m
ilar
s
it
u
atio
n
h
ap
p
en
s
b
u
t
th
i
s
ti
m
e
w
h
e
n
co
m
p
lex
b
ac
k
g
r
o
u
n
d
s
ar
e
p
r
ese
n
t
ed
,
s
u
ch
a
s
th
e
o
n
e
p
r
esen
ted
i
n
Fi
g
u
r
e
1
0
b
.
Her
e,
th
e
n
et
w
o
r
k
,
d
esp
it
e
h
a
v
i
n
g
b
ee
n
ab
le
to
co
r
r
ec
tly
id
e
n
ti
f
y
t
h
e
m
ai
n
f
o
o
d
d
is
h
w
it
h
s
o
m
e
m
is
ta
k
e
s
i
n
t
h
e
f
r
u
it
p
late,
p
r
esen
t
s
s
eg
m
e
n
tat
io
n
o
f
t
h
e
p
r
o
tein
ca
teg
o
r
y
in
th
e
tab
le,
d
u
e
to
t
h
e
te
x
tu
r
e
it
h
as,
m
a
k
i
n
g
it
lo
o
k
li
k
e
g
r
o
u
n
d
m
ea
t,
a
l
s
o
b
ein
g
ac
ti
v
ated
w
h
er
e
t
h
e
s
w
ee
t
is
,
a
n
d
i
n
s
id
e
th
e
g
las
s
as
i
f
it
w
er
e
r
ice.
A
l
th
o
u
g
h
th
is
d
o
es
n
o
t
h
ap
p
en
t
o
a
lar
g
e
ex
te
n
t,
w
h
en
l
ig
h
ti
n
g
is
lo
w
,
t
h
i
s
er
r
o
r
ten
d
s
to
ap
p
ea
r
.
(
a)
(
b
)
Fig
u
r
e
1
0
.
C
ases
o
f
w
r
o
n
g
s
e
g
m
en
tatio
n
O
rig
in
a
l
G
ro
u
n
d
T
ru
th
Seg
N
e
t
V
G
G
1
6
Re
s
Seg
D5
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
R
esS
eg
:
R
esid
u
a
l e
n
co
d
er
-
d
ec
o
d
er c
o
n
vo
l
u
tio
n
a
l
n
eu
r
a
l n
et
w
o
r
k
…
(
Ja
vier O
r
la
n
d
o
P
in
z
ó
n
A
r
en
a
s
)
1025
4.
CO
NCLU
SI
O
NS
I
n
th
is
p
ap
er
,
it
w
a
s
ex
p
lo
r
ed
th
e
u
s
e
o
f
r
esid
u
al
la
y
er
s
f
o
r
th
e
task
o
f
f
o
o
d
s
eg
m
en
tatio
n
i
n
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
e
t
w
o
r
k
s
,
w
it
h
in
w
h
ic
h
an
ar
ch
itect
u
r
e
w
it
h
p
er
f
o
r
m
an
ce
clo
s
e
to
o
n
e
o
f
th
e
m
o
s
t
u
s
ed
n
et
w
o
r
k
s
i
n
s
eg
m
e
n
tatio
n
w
a
s
i
m
p
le
m
e
n
ted
,
w
h
ich
w
a
s
n
a
m
ed
R
esS
e
g
.
Dif
f
er
e
n
t
ar
c
h
itect
u
r
es
w
i
th
d
if
f
er
e
n
t
co
n
f
i
g
u
r
atio
n
s
w
er
e
s
tr
u
c
tu
r
ed
in
th
e
lo
ca
tio
n
o
f
th
e
r
esid
u
al
la
y
er
s
,
m
ak
in
g
co
m
p
ar
is
o
n
s
w
ith
b
asic
s
eg
m
e
n
tatio
n
n
et
w
o
r
k
s
,
s
h
o
w
i
n
g
th
at
h
a
v
i
n
g
a
s
h
allo
w
d
ep
th
,
th
e
r
esid
u
al
la
y
er
s
m
ak
e
th
e
ar
ch
itect
u
r
e
pe
r
f
o
r
m
les
s
t
h
a
n
t
h
o
s
e
ar
ch
it
ec
tu
r
es
th
at
d
o
n
o
t
u
s
e
t
h
e
m
,
ev
en
w
it
h
th
e
s
a
m
e
d
ep
th
,
as
s
h
o
w
n
i
n
T
ab
le
1
.
On
t
h
e
o
th
er
h
a
n
d
,
w
h
e
n
i
n
cr
ea
s
in
g
t
h
e
d
ep
th
o
f
t
h
e
n
et
w
o
r
k
,
in
t
h
e
ca
s
e
o
f
co
m
p
ar
is
o
n
b
et
w
ee
n
R
esS
e
g
D3
an
d
D5
,
th
e
p
er
f
o
r
m
a
n
ce
i
m
p
r
o
v
es
g
r
ea
tl
y
,
b
y
m
o
r
e
th
a
n
2
%
in
t
h
e
a
v
er
ag
e
ac
cu
r
a
c
y
,
an
d
u
p
to
1
1
%
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
m
I
o
U.
T
h
e
co
r
r
ec
t
lab
elin
g
o
f
th
e
d
atab
ase
p
la
y
s
a
f
u
n
d
a
m
e
n
tal
r
o
le
in
t
h
e
tr
ai
n
i
n
g
o
f
ar
ch
itect
u
r
es,
s
i
n
ce
,
as
ev
id
e
n
ce
d
in
s
ec
tio
n
3
.
1
,
n
o
t
h
av
i
n
g
a
lab
el
co
r
r
esp
o
n
d
in
g
to
th
e
b
a
c
k
g
r
o
u
n
d
,
m
a
k
es
th
e
p
er
f
o
r
m
a
n
ce
s
o
f
th
ese
b
e
cr
iticall
y
i
m
p
air
ed
,
tak
i
n
g
i
n
to
ac
co
u
n
t
ev
er
y
t
h
i
n
g
n
o
t
lab
eled
as
p
ar
ts
o
f
th
e
o
th
er
ca
teg
o
r
ies.
Fo
r
th
is
r
ea
s
o
n
,
it
i
s
i
m
p
o
r
tan
t
to
in
cl
u
d
e
an
ad
d
itio
n
a
l
lab
el
th
at
in
c
lu
d
es
t
h
e
p
ar
ts
n
o
t
r
elate
d
to
th
e
ca
te
g
o
r
ies
in
t
h
e
i
m
a
g
e.
A
lt
h
o
u
g
h
t
h
e
R
e
s
Se
g
D5
h
a
s
a
h
ig
h
p
er
f
o
r
m
a
n
ce
,
e
x
ce
ed
in
g
9
0
%
ac
c
u
r
ac
y
,
it
s
till
r
e
m
ai
n
s
b
elo
w
t
h
e
Se
g
Net
VGG
-
1
6
,
h
o
w
e
v
er
,
w
h
en
p
r
o
ce
s
s
in
g
i
m
a
g
es
w
it
h
f
r
esh
l
y
s
er
v
ed
f
o
o
d
,
th
e
co
n
to
u
r
o
f
t
h
e
s
e
g
m
en
ted
s
ec
tio
n
s
i
m
p
r
o
v
e
s
,
av
o
id
in
g
s
eg
m
e
n
ti
n
g
e
m
p
t
y
s
p
ac
es
b
et
wee
n
th
e
s
a
m
e
t
y
p
e
s
o
f
f
o
o
d
an
d
b
etter
d
elin
ea
tin
g
th
e
co
n
to
u
r
s
b
et
w
ee
n
ea
c
h
ca
teg
o
r
y
,
as
s
h
o
w
n
i
n
Fi
g
u
r
e
9
.
I
n
f
u
tu
r
e
w
o
r
k
,
it
is
p
r
o
p
o
s
ed
to
in
cr
ea
s
e
t
h
e
ca
p
ac
it
y
o
f
t
h
e
n
et
w
o
r
k
b
y
ad
d
in
g
a
g
r
ea
ter
d
ep
th
in
ea
ch
p
h
ase
o
f
th
e
e
n
co
d
er
/
d
ec
o
d
er
w
it
h
3
x
3
f
ilter
s
.
T
h
i
s
ca
n
i
m
p
r
o
v
e
its
ac
cu
r
ac
y
,
s
in
ce
it
w
o
u
ld
allo
w
it
to
lear
n
i
n
a
b
etter
w
a
y
th
e
c
h
ar
ac
ter
is
tic
s
o
f
th
e
co
m
p
le
x
tex
t
u
r
es
o
f
t
h
e
en
v
ir
o
n
m
e
n
t
to
av
o
id
th
e
co
n
f
u
s
io
n
t
h
at
m
a
y
b
e
g
en
er
ated
b
y
t
h
e
la
y
er
s
w
it
h
1
x
1
f
ilter
s
,
i
m
p
le
m
e
n
ti
n
g
n
e
w
c
o
n
f
i
g
u
r
atio
n
s
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
e
au
th
o
r
s
ar
e
g
r
atef
u
l
to
th
e
Nu
e
v
a
Gr
an
ad
a
Militar
y
Un
iv
er
s
it
y
,
w
h
ich
,
t
h
r
o
u
g
h
its
Vice
-
ch
an
ce
l
lo
r
f
o
r
r
esear
ch
,
f
i
n
a
n
ce
s
th
e
p
r
esen
t
p
r
o
j
ec
t
w
ith
c
o
d
e
I
MP
-
I
NG
-
2
9
3
5
(
b
ein
g
in
f
o
r
ce
2
0
1
9
-
2
0
2
0
)
an
d
titl
ed
“Flex
ib
l
e
r
o
b
o
tic
p
r
o
to
ty
p
e
f
o
r
f
ee
d
in
g
ass
is
ta
n
ce
”,
f
r
o
m
w
h
ich
t
h
e
p
r
esen
t
w
o
r
k
is
d
er
iv
ed
.
RE
F
E
R
E
NC
E
S
[1
]
P.
P
o
u
lad
z
a
d
e
h
,
G
.
V
il
lalo
b
o
s,
R.
A
l
m
a
g
h
ra
b
i
a
n
d
S
.
S
h
irm
o
h
a
m
m
a
d
i,
"
A
n
o
v
e
l
S
V
M
b
a
se
d
fo
o
d
re
c
o
g
n
it
io
n
m
e
th
o
d
f
o
r
c
a
lo
rie
m
e
a
su
re
m
e
n
t
a
p
p
li
c
a
ti
o
n
s,"
i
n
2
0
1
2
IEE
E
In
t
e
rn
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
M
u
lt
i
me
d
ia
a
n
d
Exp
o
W
o
rk
sh
o
p
s
,
IE
EE
,
p
p
.
4
9
5
-
4
9
8
,
2
0
1
2
.
[2
]
V
.
Br
u
n
o
a
n
d
C
.
J.
S
il
v
a
Re
se
n
d
e
,
"
A
su
rv
e
y
o
n
a
u
to
m
a
ted
f
o
o
d
m
o
n
it
o
r
in
g
a
n
d
d
ieta
ry
m
a
n
a
g
e
m
e
n
t
s
y
ste
m
s,
"
J
o
u
rn
a
l
o
f
h
e
a
lt
h
&
me
d
ica
l
in
f
o
rm
a
ti
c
s
,
v
o
l.
8
(
3
)
,
2
0
1
7
.
[3
]
M
.
Bo
sc
h
,
F
.
Z
h
u
,
N.
Kh
a
n
n
a
,
C.
J.
Bo
u
sh
e
y
a
n
d
E.
J.
De
lp
,
"
Co
m
b
in
in
g
g
lo
b
a
l
a
n
d
lo
c
a
l
f
e
a
tu
re
s
f
o
r
f
o
o
d
id
e
n
ti
f
ica
ti
o
n
in
d
ieta
ry
a
ss
e
ss
m
e
n
t,
"
in
Im
a
g
e
Pro
c
e
ss
in
g
(
ICIP)
,
2
0
1
1
1
8
th
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
,
IEE
E
,
p
p
.
1
7
8
9
-
1
7
9
2
,
2
0
1
1
.
[4
]
Y.
M
a
tsu
d
a
,
H.
Ho
a
sh
i
a
n
d
K.
Ya
n
a
i,
"
Re
c
o
g
n
it
io
n
o
f
m
u
lt
ip
le
-
f
o
o
d
im
a
g
e
s
b
y
d
e
te
c
ti
n
g
c
a
n
d
id
a
te
re
g
io
n
s,
"
in
M
u
lt
ime
d
ia
a
n
d
Exp
o
(
ICM
E),
2
0
1
2
IE
EE
I
n
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
,
IEE
E
,
p
p
.
2
5
-
30,
2
0
1
2
.
[5
]
F
.
Ko
n
g
a
n
d
J.
T
a
n
,
"
Die
tCa
m
:
A
u
to
m
a
ti
c
d
ieta
r
y
a
ss
e
ss
m
e
n
t
w
it
h
m
o
b
il
e
c
a
m
e
ra
p
h
o
n
e
s,
"
Per
v
a
siv
e
a
n
d
M
o
b
il
e
Co
mp
u
t
in
g
,
v
o
l
.
8
(
1
)
,
p
p
.
1
4
7
-
1
6
3
,
2
0
1
2
.
[6
]
L
.
Bo
ss
a
rd
,
M
.
G
u
il
lau
m
in
a
n
d
L.
V
a
n
G
o
o
l,
"
F
o
o
d
-
1
0
1
–
m
in
in
g
d
isc
rim
in
a
ti
v
e
c
o
m
p
o
n
e
n
ts
w
it
h
ra
n
d
o
m
f
o
re
sts,
"
in
E
u
ro
p
e
a
n
C
o
n
fer
e
n
c
e
o
n
C
o
mp
u
ter
Vi
si
o
n
,
S
p
rin
g
e
r
,
Ch
a
m
,
p
p
.
4
4
6
-
4
6
1
,
2
0
1
4
.
[7
]
W
.
Zh
a
n
g
,
Q.
Yu
,
B.
S
id
d
iq
u
ie,
A
.
Div
a
k
a
r
a
n
a
n
d
H.
S
a
w
h
n
e
y
,
"”
S
n
a
p
-
n
-
Eat”
f
o
o
d
re
c
o
g
n
it
io
n
a
n
d
n
u
tri
t
io
n
e
sti
m
a
ti
o
n
o
n
a
sm
a
rtp
h
o
n
e
,
"
J
o
u
rn
a
l
o
f
d
ia
b
e
tes
sc
ien
c
e
a
n
d
tec
h
n
o
lo
g
y
,
v
o
l.
9
(
3
)
,
p
p
.
5
2
5
-
5
3
3
,
2
0
1
5
.
[8
]
M
.
Y.
Ch
e
n
,
e
t
a
l
.,
"
A
u
to
m
a
ti
c
Ch
in
e
se
f
o
o
d
id
e
n
ti
f
ica
ti
o
n
a
n
d
q
u
a
n
t
it
y
e
sti
m
a
ti
o
n
,
"
in
S
IGG
RA
PH
Asia
2
0
1
2
T
e
c
h
n
ica
l
Briefs
,
A
CM
,
p
p
.
2
9
,
2
0
1
2
.
[9
]
S
.
Ya
n
g
,
M
.
Ch
e
n
,
D.
P
o
m
e
rlea
u
a
n
d
S
u
k
t
h
a
n
k
a
r
R.
,
"
F
o
o
d
re
c
o
g
n
it
io
n
u
si
n
g
sta
ti
stics
o
f
p
a
irw
is
e
lo
c
a
l
f
e
a
tu
re
s,
"
in
Co
m
p
u
ter
Vi
si
o
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
(
CVP
R),
2
0
1
0
IE
EE
Co
n
fer
e
n
c
e
o
n
,
IEE
E
,
p
p
.
2
2
4
9
-
2
2
5
6
,
2
0
1
0
.
[1
0
]
F
.
Zh
u
,
M
.
B
o
sc
h
,
N.
Kh
a
n
n
a
,
C.
J.
Bo
u
sh
e
y
a
n
d
E.
J.
De
lp
,
"
M
u
lt
il
e
v
e
l
se
g
m
e
n
tatio
n
f
o
r
f
o
o
d
c
las
sif
ic
a
ti
o
n
i
n
d
ieta
ry
a
ss
e
ss
m
e
n
t,
"
in
Ima
g
e
a
n
d
S
ig
n
a
l
Pro
c
e
ss
in
g
a
n
d
An
a
lys
i
s
(
IS
PA
),
2
0
1
1
7
t
h
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
si
u
m
o
n
,
IEE
E
,
p
p
.
3
3
7
-
3
4
2
,
2
0
1
1
.
[1
1
]
Y.
L
e
Cu
n
,
Y.
Be
n
g
io
a
n
d
G
.
Hin
t
o
n
,
"
De
e
p
lea
rn
i
n
g
,"
Na
tu
re
,
v
o
l
.
5
2
1
(
7
5
5
3
)
,
4
3
6
,
2
0
1
5
.
[1
2
]
M
.
D.
Zeiler
a
n
d
R.
F
e
rg
u
s,
"
V
isu
a
li
z
in
g
a
n
d
u
n
d
e
rs
tan
d
i
n
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
tw
o
rk
s,"
in
Eu
ro
p
e
a
n
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
,
S
p
ri
n
g
e
r
,
C
h
a
m
,
p
p
.
8
1
8
-
8
3
3
,
2
0
1
4
.
[1
3
]
S
.
M
a
ll
a
t,
"
Un
d
e
rsta
n
d
in
g
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
tw
o
rk
s,
"
Ph
il
o
so
p
h
ic
a
l
T
ra
n
s
a
c
ti
o
n
s
o
f
t
h
e
R
o
y
a
l
S
o
c
iety
A
:
M
a
th
e
ma
ti
c
a
l
,
P
h
y
sic
a
l
a
n
d
En
g
i
n
e
e
rin
g
S
c
ien
c
e
s
,
v
o
l.
3
7
4
(
2
0
6
5
)
,
p
p
.
2
0
1
5
0
2
0
3
,
2
0
1
6
.
[1
4
]
W
.
S
h
im
o
d
a
a
n
d
K.
Ya
n
a
i,
"
CNN
-
b
a
se
d
f
o
o
d
im
a
g
e
se
g
m
e
n
tatio
n
w
it
h
o
u
t
p
ix
e
l
-
w
is
e
a
n
n
o
tati
o
n
,
"
in
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ima
g
e
An
a
lys
is
a
n
d
Pro
c
e
ss
in
g
,
S
p
rin
g
e
r
,
Ch
a
m
,
p
p
.
4
4
9
-
4
5
7
,
2
0
1
5
.
[1
5
]
A
.
M
e
y
e
rs,
e
t
a
l
.,
"
I
m
2
Ca
lo
ries
:
T
o
w
a
rd
s
a
n
a
u
to
m
a
t
e
d
m
o
b
il
e
v
isio
n
f
o
o
d
d
iary
,
"
in
Pro
c
e
e
d
in
g
s
o
f
th
e
IEE
E
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
C
o
mp
u
ter
Vi
si
o
n
,
p
p
.
1
2
3
3
-
1
2
4
1
,
2
0
1
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
0
:
1
0
1
7
-
1
0
2
6
1026
[1
6
]
V
.
Ba
d
rin
a
ra
y
a
n
a
n
,
A
.
Ke
n
d
a
ll
,
R.
Cip
o
ll
a
,
"
S
e
g
n
e
t:
A
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
a
rc
h
it
e
c
tu
re
f
o
r
i
m
a
g
e
se
g
m
e
n
tatio
n
,
"
IEE
E
tra
n
sa
c
ti
o
n
s
o
n
p
a
tt
e
rn
a
n
a
lys
is
a
n
d
m
a
c
h
in
e
in
telli
g
e
n
c
e
,
v
o
l.
3
9
(
12
)
,
p
p
.
2
4
8
1
-
2
4
9
5
,
2
0
1
7
.
[1
7
]
M
.
Ka
m
p
ffm
e
y
e
r,
A
.
B.
S
a
lb
e
r
g
a
n
d
R.
Je
n
ss
e
n
,
"
S
e
m
a
n
ti
c
se
g
m
e
n
tatio
n
o
f
sm
a
ll
o
b
jec
ts
a
n
d
m
o
d
e
li
n
g
o
f
u
n
c
e
rtain
ty
in
u
r
b
a
n
re
m
o
te
se
n
sin
g
im
a
g
e
s
u
sin
g
d
e
e
p
c
o
n
v
o
l
u
t
io
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s,
"
in
Pro
c
e
e
d
in
g
s
o
f
th
e
IE
EE
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
it
io
n
wo
rk
sh
o
p
s
,
p
p
.
1
-
9
,
2
0
1
6
.
[1
8
]
J.
Ch
e
n
g
,
Y.
H.
Tsa
i,
W
.
C.
Hu
n
g
,
S
.
W
a
n
g
a
n
d
M
.
H.
Ya
n
g
,
"
F
a
st
a
n
d
a
c
c
u
ra
te
o
n
li
n
e
v
id
e
o
o
b
je
c
t
se
g
m
e
n
tatio
n
v
ia t
ra
c
k
in
g
p
a
rts
,"
i
n
Pro
c
e
e
d
in
g
s o
f
th
e
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
m
p
u
ter
Vi
sio
n
a
n
d
P
a
tt
e
rn
Rec
o
g
n
i
ti
o
n
,
p
p
.
7
4
1
5
-
7
4
2
4
,
2
0
1
8
.
[1
9
]
A
.
Ke
n
d
a
ll
,
V
.
Ba
d
r
in
a
ra
y
a
n
a
n
a
n
d
R.
Cip
o
ll
a
,
"
Ba
y
e
sia
n
se
g
n
e
t:
M
o
d
e
l
u
n
c
e
rtain
ty
in
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
a
rc
h
it
e
c
tu
re
s f
o
r
sc
e
n
e
u
n
d
e
rsta
n
d
in
g
,"
a
rXiv
p
re
p
rin
t
a
rX
iv:1
5
1
1
.
0
2
6
8
0
,
2
0
1
5
.
[2
0
]
J.
T
a
n
g
,
J.
L
i
a
n
d
X
.
Xu
,
"
S
e
g
n
e
t
-
b
a
se
d
g
lan
d
se
g
m
e
n
tatio
n
f
ro
m
c
o
lo
n
c
a
n
c
e
r
h
isto
l
o
g
y
i
m
a
g
e
s
,
"
in
2
0
1
8
3
3
r
d
Y
o
u
t
h
Ac
a
d
e
mic
An
n
u
a
l
C
o
n
fer
e
n
c
e
o
f
Ch
i
n
e
se
Asso
c
ia
ti
o
n
o
f
A
u
t
o
ma
ti
o
n
(
Y
AC)
,
IEE
E
,
p
p
.
1
0
7
8
-
1
0
8
2
,
2
0
1
8
.
[2
1
]
P
.
Ku
m
a
r,
P
.
Na
g
a
r,
C.
A
ro
ra
a
n
d
A
.
G
u
p
ta,
"
U
-
S
e
g
n
e
t:
F
u
ll
y
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
b
a
se
d
a
u
to
m
a
ted
b
ra
in
ti
ss
u
e
se
g
m
e
n
tatio
n
to
o
l
,
"
in
2
0
1
8
2
5
t
h
IEE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ima
g
e
Pro
c
e
ss
in
g
(
ICIP)
,
IEE
E
,
p
p
.
3
5
0
3
-
3
5
0
7
,
2
0
1
8
.
[2
2
]
N.
In
g
,
e
t
a
l
.,
"
S
e
m
a
n
ti
c
s
e
g
m
e
n
tatio
n
f
o
r
p
ro
sta
te
c
a
n
c
e
r
g
r
a
d
in
g
b
y
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s,
"
in
M
e
d
ica
l
Ima
g
i
n
g
2
0
1
8
:
Dig
it
a
l
P
a
t
h
o
l
o
g
y
,
In
tern
a
ti
o
n
a
l
S
o
c
iety
f
o
r
Op
ti
c
s a
n
d
P
h
o
to
n
ics
,
p
p
.
1
0
5
8
1
1
B
,
2
0
1
8
.
[2
3
]
S.
A
lq
a
z
z
a
z
,
X
.
S
u
n
,
X.
Ya
n
g
a
n
d
L
.
N
o
k
e
s,
"
A
u
to
m
a
ted
b
ra
in
tu
m
o
r
se
g
m
e
n
tatio
n
o
n
m
u
lti
-
m
o
d
a
l
M
R
im
a
g
e
u
sin
g
S
e
g
Ne
t,
"
Co
mp
u
t
a
ti
o
n
a
l
Vi
su
a
l
M
e
d
i
a
,
v
o
l.
5
(
2
)
,
p
p
.
2
0
9
-
2
1
9
,
2
0
1
9
.
[2
4
]
T
.
P
o
h
len
,
A
.
He
r
m
a
n
s,
M
.
M
a
th
ias
a
n
d
B.
L
e
ib
e
,
"
F
u
ll
-
re
so
lu
t
i
o
n
re
sid
u
a
l
n
e
tw
o
rk
s
f
o
r
se
m
a
n
ti
c
se
g
m
e
n
tatio
n
in
stre
e
t
sc
e
n
e
s
,"
in
Co
mp
u
ter
Vi
sio
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
(
CVP
R),
2
0
1
7
IEE
E
Co
n
fer
e
n
c
e
o
n
,
IEE
E
,
p
p
.
3
3
0
9
-
3
3
1
8
,
2
0
1
7
.
[2
5
]
K.
S
im
o
n
y
a
n
a
n
d
A
.
Zi
ss
e
r
m
a
n
,
"
V
e
r
y
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
tw
o
rk
s
f
o
r
lar
g
e
-
sc
a
le
i
m
a
g
e
re
c
o
g
n
it
io
n
,
"
a
rXiv
p
re
p
rin
t
a
rXiv:1
4
0
9
.
1
5
5
6
,
2
0
1
4
.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
J
a
v
ier
O
r
la
n
d
o
Pi
n
z
ó
n
A
r
e
n
a
s
w
a
s
b
o
rn
in
S
o
c
o
rr
o
-
S
a
n
tan
d
e
r,
Co
lo
m
b
ia,
in
1
9
9
0
.
He
re
c
e
iv
e
d
h
is
d
e
g
re
e
in
M
e
c
h
a
tro
n
ics
En
g
in
e
e
rin
g
(Cu
m
L
a
u
d
e
)
in
2
0
1
3
,
S
p
e
c
ializa
ti
o
n
i
n
En
g
in
e
e
rin
g
P
ro
jec
t
M
a
n
a
g
e
m
e
n
t
in
2
0
1
6
,
a
n
d
M
.
S
c
.
in
M
e
c
h
a
tr
o
n
ics
E
n
g
in
e
e
rin
g
in
2
0
1
9
,
a
t
th
e
Nu
e
v
a
G
ra
n
a
d
a
M
il
it
a
r
y
Un
iv
e
rsit
y
-
UMN
G
.
He
h
a
s
e
x
p
e
rie
n
c
e
in
th
e
a
re
a
s
o
f
a
u
to
m
a
ti
o
n
,
e
lec
tro
n
ic
c
o
n
tro
l
,
a
n
d
m
a
c
h
in
e
lea
rn
in
g
.
Cu
rre
n
tl
y
,
h
e
is
stu
d
y
in
g
a
P
h
.
D.
in
A
p
p
li
e
d
S
c
ien
c
e
s
a
n
d
w
o
rk
in
g
a
s
a
G
r
a
d
u
a
te
A
ss
i
s
tan
t
a
t
th
e
UMNG
w
it
h
e
m
p
h
a
sis
o
n
R
o
b
o
ti
c
s
a
n
d
M
a
c
h
in
e
L
e
a
rn
in
g
.
E
-
m
a
il
:
u
3
9
0
0
2
3
1
@
u
n
im
il
it
a
r.
e
d
u
.
c
o
Ro
b
i
n
s
o
n
J
i
m
é
n
e
z
M
o
r
e
n
o
w
a
s
b
o
r
n
i
n
B
o
g
o
tá,
C
o
lo
m
b
ia,
i
n
1
9
7
8
.
He
re
c
e
iv
e
d
th
e
En
g
i
n
e
e
r
d
e
g
re
e
in
El
e
c
tro
n
ics
a
t
t
h
e
F
ra
n
c
isc
o
Jo
sé
d
e
Ca
ld
a
s District
Un
iv
e
rsit
y
-
UD
-
in
2
0
0
2
.
M
.
S
c
.
i
n
In
d
u
strial
A
u
to
m
a
ti
o
n
f
ro
m
th
e
Un
iv
e
rsid
a
d
Na
c
io
n
a
l
d
e
Co
lo
m
b
ia
-
2
0
1
2
a
n
d
P
h
.
D
.
in
En
g
in
e
e
rin
g
a
t
th
e
F
ra
n
c
isc
o
Jo
s
é
d
e
Ca
ld
a
s
District
Un
iv
e
rsity
-
2
0
1
8
.
He
is
c
u
rre
n
tl
y
w
o
rk
in
g
a
s
a
P
ro
f
e
ss
o
r
in
th
e
M
e
c
h
a
tro
n
ics
En
g
i
n
e
e
rin
g
P
r
o
g
ra
m
a
t
th
e
Nu
e
v
a
G
ra
n
a
d
a
M
il
it
a
r
y
Un
iv
e
rsit
y
-
UMN
G
.
He
h
a
s
e
x
p
e
rien
c
e
in
th
e
a
re
a
s
o
f
In
stru
m
e
n
tatio
n
a
n
d
El
e
c
tro
n
ic
Co
n
tro
l
,
a
c
ti
n
g
m
a
in
ly
in
Ro
b
o
t
ics
,
c
o
n
t
ro
l
,
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
a
n
d
im
a
g
e
p
ro
c
e
ss
in
g
.
E
-
m
a
il
:
ro
b
in
so
n
.
ji
m
e
n
e
z
@u
n
im
il
it
a
r.
e
d
u
.
c
o
Cé
sa
r
G
io
v
a
n
y
Pa
c
h
ó
n
S
u
e
s
c
ú
n
w
a
s
b
o
rn
in
B
o
g
o
tá,
Co
l
o
m
b
ia,
in
1
9
9
6
.
He
re
c
e
iv
e
d
h
is
d
e
g
re
e
in
M
e
c
h
a
tro
n
ics
En
g
in
e
e
r
in
g
f
ro
m
th
e
P
il
o
t
Un
iv
e
rsity
o
f
Co
lo
m
b
ia
in
2
0
1
8
.
Cu
rre
n
tl
y
,
h
e
is
stu
d
y
in
g
h
is
M
a
ste
r’s
d
e
g
re
e
in
M
e
c
h
a
tro
n
ics
En
g
in
e
e
ri
n
g
a
n
d
w
o
rk
in
g
a
s
Re
se
a
rc
h
A
s
sista
n
t
a
t
th
e
Nu
e
v
a
G
ra
n
a
d
a
M
il
it
a
ry
Un
iv
e
rsit
y
w
it
h
a
n
e
m
p
h
a
sis
o
n
Ro
b
o
ti
c
s
a
n
d
M
a
c
h
in
e
L
e
a
rn
in
g
.
E
-
m
a
il
:
u
3
9
0
0
2
5
9
@
u
n
im
il
it
a
r.
e
d
u
.
c
o
Evaluation Warning : The document was created with Spire.PDF for Python.