►
From YouTube: ONNX Edge WG meeting 20190417
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
C
A
A
A
A
A
A
A
A
A
Basically,
you
know
I'm,
assuming
or
like
everybody
to
acknowledge
and
give
consent
for
recording
into
this
meeting,
because
you
know
that's
five
girls,
so
I
think
in
the
future.
Indian
rights
I
think
you'll
be
may
be
returned
like
automatically.
If
you're
done
into
the
meeting,
you
give
you
a
consent
to
be
recorded,
and
so
on
so
well
on
is.
A
A
A
A
A
Discussion,
okay,
so
agenda,
just
quick,
a
quick
updates
main
agenda
is
really
to
go
to
the
state
homework
and
really
reduced
it.
No
work,
so
we
can
really
get
the
potentially
final
or
close
to
final
draft
that
we
can
then
share
with
the
steering
committee
and
get
really.
You
know
approved
that
this
is
our
statement
of
work,
and
this
is
what
we're
gonna
deliver
as
a
working
group
and
be
the
clear
exit
criteria
and
so
on.
So
that's
a
focus
of
this
meeting.
A
A
So
base
misper,
you
know
agenda
and
some
action
items
from
last
time
I
did.
I
did
update
that
the
readme
and
the
working
groups
edge
directory
with
some
more
details
about
you
know,
as
you
discussed
last
time,
and
it's
some.
You
know
people
that
are
participating
in
this
working
group
as
per
attendance.
So
far
we
can
update
this
as
it
goes
and
there's
some
you
know
basic
meeting,
logistics
and
so
on.
A
A
A
A
Yeah,
so
if,
if
nobody
has
any
comments,
then
I
guess
the
approach
would
be
that
we
go
through
it
and
me
enough
for
a
mere
lead,
and
you
cannot
explain
rationale
for
the
statement
of
work
and
what
we
are
going
to
sign
up
to
do
and
then
hopefully
we
come
with
a
bid,
a
conclusion
at
the
end
of
the
meeting
that
this
is
it
over.
We
need
to
extend
more
so
if
no
questions
or
comments
or
directly
that
anybody
else
could
raise
right
now,
then
I
would
proceed
with
just
going
through
a
document.
I
think.
B
That
maybe
what
we
should
do,
because
statement
of
folk
had
to
be
delivered
to
the
SIG's
and
the
steering
gray
is
a
first
step.
We
need
to
try
forgive
a
a
date
to
ourselves.
Let's
say
we
talked
about
this
meeting.
We
give
one
more
meeting
and
if
there
aren't
any
other
comments,
it
means
that
on
the
next
meeting,
we're
going
to
approve
it
and
give
it
to
the
steering
committee
to
fork
over.
A
We
need
to
it's
our
first
deliverable,
really.
We
need
to
make
sure
we
have
a
clear
definition.
What
we
were
working
on
and
I
give
it
over.
We
need
to
have
some
deadlines
and
when
by
when,
we
need
to
finish
this
so
hopefully
by
by
you
know
we
finalize
not
later
than
next
meeting
as
make
sense.
Okay,.
A
A
B
B
B
Cool
so
basically
running
onyx
moderns
on
wider
set
of
platforms
is
required,
and
there
is
a
class
of
platforms
that,
with
a
famous
edge-
and
you
know,
part
of
the
work-
would
actually
be
to
classify
what
phones
into
age
in
the
previous
face-to-face.
We
already
saw
that
I
somewhat.
Our
definition
and
the
definition,
for
example,
of
Huawei
are
different
and
both
Megan
and
I
believe
that
we
can
encompass
all
of
that
under
age.
We
just
need
to
partition
it
to
different
types
of
devices.
B
Of
course,
you
cannot
treat
a
IOT
device,
the
same
way
that
you're
13
gay
influence
age
several,
so
we
need
to
define
the
range
of
age
and
we
need
to
inside
edge
somehow
politician
it
to
whatever
favors
affects.
So
the
purpose
is
to
explore
what
are
the
implications
and
constraints
on
these
platforms
and
projected
into
some
of
onyx
aspects,
for
example,
operations
as
well
going
to
talk
and
provide
these
accommodations
to
the
sick.
That's
basically
in
it's
a
very
high-level
description
of
what
we're
going
to
do.
B
B
B
We
would
like
to
identify
scenarios
and
use
cases
that
are
applicable
to
edge
devices.
This
is
where
we
have
a
document
for
whywe,
which
is
a
very
good
starting
point
to
map
down
these
use
cases
and
I.
Believe.
If
we
look
at
these
scenarios
and
use
cases,
we
can
map
them
into
Poe
fangs.
These
are
fins
that
you
expect
the
IOT
device
profile
to
do,
and
these
are
the
things
that
you
would
expect
a
a
mobile
phone,
for
example.
B
B
These
combines
for
edge
devices,
because
this
is
feedback
that
we're
getting
for
customers
and
I
believe
that
others
are
getting
is
well
that
it's
important
to
to
define
some
frame
some
structure
that
would
be
able
to
say
this
is
an
onyx
modem
and
I
expect
a
compliant
device
to
meet
this
behavior
in
this
behavior,
in
terms
of
mostly
accuracy,
we're
going
to
get
to
the
next
point
so
on.
But
in
my
view
at
least,
compliance
is
more
on
in
terms
of
a
functionality
and
oqz
twisting
in
the
scope
of
the
onyx,
a
foundation.
D
B
E
A
E
A
Yeah,
so
that
pool
request
I
actually
I
was
looking
at,
provide
some
comments
actually
as
to
where
to
pull
it.
It
seems
to
me
that
that
document
is
kind
of
capturing
the
discussion,
and
that
was
that
happened
in
the
previous
workshop.
So
I
was
thinking.
That
document
really
should
be
description.
Okay,
here's
the
proposal
that
was
discussed
enough,
say
December
workshop
and
make
it
into
the
workshop
directories
right.
So
we
have
kind
of
this
is
maybe.
E
E
A
The
one
that
they
came
from
your
original
document
and
a
Google
Doc
right
and
yes,
that
one
seemed
like
something
that
was
really
covering
material
or
discussion
that
happened
in
the
last
march,
worship
of
the
Divine
December
yeah
right.
Okay,
that's
my
understanding.
Okay,
that's
good!
So,
okay,
I
guess
both
of
them
I
think,
would
need
to
be
then
placed
in
some.
What
offer
is
suggesting
seems
fine
to
me
like
discussion,
materials
or
something
and
then
and
then
we'll
do
that
right
right.
B
The
one
we're
going
to
probably
pull
from
the
founder
of
discussion
materials
into
a
folder
of
documents
that,
but
when
we
create,
for
example,
in
description
of
scenarios,
description
of
poor
files,
so
we
kind
of
separate
between
things
that
people
put
there
versus
things
that
we
created
as
efficient
delivery.
In
a
sense,
this
is,
in
my
view
why
we
need
to
create
this
separated
found.
Oh
yes,
a
taqueria.
A
So
yeah
and
actually
to
that
time
you
know
going
back
now
to
the
actual
document
I
mean
we
are
having
the
section
that
is
about
their
liberal
bones
right,
so
I'm
guessing
there
will
be
potentially
area
or
subdirectory
they're
called
renewables
or,
or
you
know,
documents
or
whatever
we
decided
to
call
it
right.
Mm-Hmm,
okay,
yeah
offer
it
then
okay
sounds
good
yeah
thanks
for
the
slight
digression
sure.
B
B
Going
back
to
compliance,
so
compliance
not
of
the
problem
that
I've
been
having
for
a
long
period
of
time
with
onyx
is
that
there
is
no
defined
compliance.
Criteria
like
people
can
be
compliant
with
onyx,
without
even
supporting
even
a
single
operation
and
use
the
logo
and
I
believe
that
the
moderns
will
seek,
and
others
are
already
working
on.
Creating
some
compliance
criteria
and.
B
We
should
look
at,
we
should
talk
with
these
groups
and
we
should
look
at
compliance
for
edge
devices,
mainly
in
terms
of
subset
of
operations.
The
way
that
I
see
it
to
define
a
subset
of
operations,
which
is
a
clickable
toe
edge
device,
is
some
of
the
operations
in
onyx.
Oh,
what's
the
number
now
one
hundred
and
lt
I
forgot:
what's
the
updated
number
you
know
is
another
set
of
these
operations
main
to
be
applicable
to
the
scenarios.
B
Okay,
we're
going
to
the
middle.
The
first
iteration
may
be
a
year
from
now,
there's
going
to
be
another
short
update,
iteration
right
devices,
new
use
cases,
everything
updates
and
we
will
have
to
revisit
everything
that
we
did
and
say.
Ok,
we
need
to
add
a
few
use
cases.
We
need
to
add,
maybe
a
few
operations.
F
B
We
we
have
an
idea
how
to
define
a
process
which
doesn't
rely
on
having
any
kind
of
shared
on
time
or
anything.
We
can
share
this
process.
The
process
is
creates
a
test
package
that
includes
golden
references,
and
then
you
will
be
able
to
compare
outputs
of
a
device
that
wants
to
be
combined
with
the
reference
based
on
the
criteria
and
if
it
passes
your
compliant,
and
this
entire
thing
has
to
be
again
collaborated
with
the
other
things.
But
I
see
this.
Establishing
this
mechanism
is
one
of
the
important
deliveries
of
this
washing
soap.
D
B
Okay,
so
this
is
part
of
the
process
that
we
need
to
discuss.
We
need
to
describe
the
ideal
is
in
an
arching.
Let's
say
that
you
have
a
modern,
you
have
inputs
and
you
have
outputs
that
you
need
to
compare
to
and
you
have
a
pass
criteria
and
it's
the
responsibility
of
the
person
or
the
company
that
wants
to
be
compliant
to
give
out
a
report.
B
So
he
has
a
list
of
tests.
It's
his
responsibility
to
run
the
test
and
to
deliver
a
report,
and
it's
ours
or
Onix,
seek
responsibility
to
look
at
the
report
came
to
validate
it
say.
Ok,
the
report
is
finally,
there
needs
to
be
a
toast
mechanism
here.
Ok,
it
also.
This
is
the
mechanism
that
exists
in
other
organizations.
There's
kind
of
two
techniques
in
the
world.
One
technique
is
Microsoft
technique
of
having
certification
labs.
It's
a
very
costly
technique.
The
other
technique,
more
common
in
the
open
source
world
is
it
last
mechanism
we
define
this.
B
We
define
how
a
output
to
go,
go
need
to
look
like
you
deliver
the
goal.
We
validate
a
look,
and
you
say
it's
ok.
This
is
it's
compliant,
so
it's
the
responsibility
of
the
person
who
wants
to
be
compliant.
You
know
there
was
a
trust
mechanism
here
to
give
us
an
actual
log
of
running
the
actual
tests
on
the
e
system.
D
B
But
by
the
way,
keep
in
mind
that
the
working
group,
the
purpose
of
the
working
group,
is
exactly
to
do
these
discussions
and
come
up
with
other
commendations,
the
purpose
of
the
cigs.
In
that
case
it
will
be
a
more
than
zu
sig
and
the
operation
sig,
and
what
are
the
Razr
comprehend
see,
go
not
to
approve
whatever
we
suggest.
So
it's
not
that
what
we
suggest
becomes
a
described
process
of
onyx.
It
is
a
recommendation
of
how
to
do
something
and
we
will
have
to
come
out
to
these
working
groups.
B
B
Okay,
so
the
last
section
of
the
scope
talks
about
collaboration
with
a
member
of
organization,
so
onyx
is
at
least
do
the
way
that
I
in
others
perceive
it
onyx.
Repo
and
organization
is
not
about
performance.
It's
about
a
way
to
represent
moderns
and
more
functionality
and
accuracy.
Okay,
so
anything
related
to
perform
benchmarking,
probably
his
telescope
of
fornix.
But
this
is
well.
The
M&P
elf
organization
goes
into
comes
into
the
picture
that
NFL
Forgan
ization
is
was
created
exactly
two
benchmarks
and
to
show
to
have
a
unified
benchmarking
framework.
B
Is
one
of
them
needing
gray
inputs
to
do
testing
package?
Maybe
we
won't
come
to
a
point
where
the
onyx
will
be
the
only
input
to
an
internal.
Maybe
even
everybody
agrees,
but
we
would
like
to
collaborate
with
the
aged
informants
working
group,
which
is
part
of
the
payment.
Elf
organization
to
a
nine
on
terminology
and
operation,
sets
everything
that
data
types
so
that
it
would
suit
their
performance
and
a
benchmarking
purposes.
So.
A
B
D
A
A
D
A
D
D
A
B
E
E
D
B
D
A
E
B
B
In
terms
of
collaboration,
we
want
to
emphasize
that
we
should
avoid
overlapping.
You
know,
though,
to
avoid
it.
We
have
to
sync
and
collaborate
with
a
few
potential
onyx
for
groups
to
realize
what
they
are
doing
to
properly
communicate.
What
we
are
doing
so
there's
the
quantization
wall
can
go
up
to
the
modern,
zoo,
sig
conferencing.
B
A
D
B
Sorry,
so
from
the
amount
of
organization
point
of
view,
it's
going
to
simplify
things
by
saying
if
these
are
the
models
that
onyx
guarantees
to
all
college
devices,
so
these
models
can
be
a
great
good
starting
point
for
the
benchmarking,
because
we
know
these.
These
models
are
guaranteed
to
work
on
any
device
which
classifies
itself
is
age,
but
there
is
definitely
room
for
kind
of
a
feedback,
one
Sweden
either.
Let's
say
we
do
some
work.
B
B
B
So
my
point
is:
if
you
want
to
in
influence
whatever
the
onyx
edge
walking
group
is
doing,
you
should
be
part
of
onyx.
You
should
be
in
the
onyx
edge
working
group
meetings
and
so
on.
So
if
there
is
some
feedback
coming
from
the
M&P
of
attendees,
my
recommendation
to
them
would
be
go
for
the
onyx
Channing's
to
give
that
feedback.
Yeah.
E
B
B
There's
no
reason
why
someone
forum
who
is
attending
the
event
of
meeting
he
sees
something
that
we
present.
He
wants
to
give
feedback.
There's
no
reason
for
him.
Why
not
to
give
that
feedback
for
the
onyx
Channing's
by
Chanin's
I
mean
not
just
the
ghetto
challenge,
but
you
know
issue
well,
whatever
mechanism
you
would
like
to
attend
a
meeting.
E
A
This
is
not,
then,
if
that
org
turns
out
to
be
not
interested
to
really
support
onyx
as
input
formats
or
whatever,
then,
which
I
would
doubt
but
anyways
if
they
do
Dan
the
ice.
It's
you
know,
then
it's
not
working
well
for
onyx
community,
and
so
we
need
to
find
potentially
different
partner,
all
right,
yeah,
but.
B
Your
point:
if
they
just
want
to
give
some
feedback
and
they
want,
they-
don't
want
to
give
it
for
onyx
channels.
Probably
we
can
we
can
someone
from
the
Gulf
can
pick
it
up
and
put
it
into
a
next
like
the
recommendation.
Pat
would
be
do
it
for
the
right,
onyx
ripple
or
get
all
channels
or
whatever.
If
you
don't
want
to
do
it,
and
you
want
to
use
another
active
contributor
to
do
it.
Okay,
that's
fine!
B
The
goal
is
to
gather
feedback.
The
goal
is
not
to
to
be
open
to
a
reason.
I
would
say.
Okay,
so
talking
about
organization
so
edge.
Is
the
working
group
we're
going
to
gather
requirements,
the
connotations
whatever
we
also
talked
about
code,
not
metals,
and
for
me
to
documents
and
we're
going
to
put
everything
we
have
a
directory
under
the
walk
and
go
deposit
away.
So
the
goal
is
that
everything
that
we
do
everything
that
people
want
to
either
contribute
discuss
whatever
it
has
to
go
into
that
repo.
B
Right
now
we
have
some
Google,
Documents
I,
think
it's
also
a
possible
mechanism.
If
people
prefer
to
put
documents,
it's
also
happening
in
other
workouts,
so
discussion
materials
can
be
a
document
which
is
inside
Google
Docs
and
you
just
provide
a
link.
For
example,
oh
you
can
convert
it
to
an
MD
file.
B
We
have
a
dedicated
little
home,
which
is
actually
it's.
It's
really
nice
people,
it's
becoming
very
active.
It's
very
good
I.
Think,
as
in
the
training
detail,
are
the
most
active
one
right
now
very
nice,
we're
going
to
have
periodic
meetings
like
the
one
that
were
opening
we're
having
today
it
is
open
to
any
active
contributor.
Actually,
it's
even
open
to
anyone.
You
can
all
we
publish
them
in
in
zome.
Anyone
that
wants
to
join
can
join.
B
We
promise
that
we're
going
to
give
notification
and
agenda
a
few
days
before
so
far
I
will
manage
to
do
it,
mehgan's
a
keeping
an
eye
on
that
and
nothing
just
as
we
opened
that
section.
The
combination
are
going
to
be
delivered
to
the
relevant
SIG's
that
stay
part
of
the
open
government's
definition
of
what
this
walking
dope
needs
to
do.
B
B
Maybe
just
to
emphasize
one
thing:
this
walking
group
is
in
a
sense:
it's
only
pushes
things
into
their
walking
go
repo.
Everything
has
to
be
inside
the
repo.
So
when
we
finish
and
we
have
deliveries
even
if
this
deliver
is
a
code,
the
code
that
we
do
is
into
that
inside
the
type
of
the
working
groups,
it's
the
purpose
of
the
SIG's
to
pull
this
code
and
put
it
into
the
major
Onix,
a
repo.
Oh,
we
can
initiate
a
proven
quest,
but
it's
not
our
own
to
push
into
the
may.
B
B
B
If
you're
familiar
the
presentation
that
we
kind
of
gave,
we
show
how
you
can
talk
of
a
few
different
vectors
of
defining
the
profile
in
terms
of
compute
in
terms
of
connectivity
in
terms
of
security,
there's
many
vectors
and
we'll
trying
to
capture
and
define
the
essence
of
what
makes
an
IOT
device
different
than
a
mobile
device
different.
Then
a
aged
infrastructure
cell.
Well,
of
course,
we
understand
that
is
different,
that
we
want
to
map
it
into
vectors,
which
are
either
a
I.
B
Another
thing
that
was
not
mentioned
so
far
in
the
SOC
is
the
work
with
the
quantization
walking.
Cuates
ation
is
very
important
to
edge
devices
which
are
more
no
power,
no
compute,
mobile
and
IOT.
Although
the
authorization
is
a
major
use
case,
and
we
have
to
make
sure
that
whatever
onyx
stand
out,
don't
say
container
the
Onix
operations,
whatever
datatypes,
whatever
they
define
on
quantization,
is
going
to
fit.
B
So
that's
the
collaboration
with
the
quantization
working
group.
I,
don't
know
what
form
it's
going
to
take.
It
may
be
a
document,
maybe
a
set
of
issues,
pull
requests
or
something,
but
we're
going
to
start
kicking
off
this
poses
probably
here
I
mean
he
may
be
in
the
next
few
weeks
kick
off
this
activity.
B
The
last
item
on
the
list
is
to
collaborate
with
moderns,
oh
and
operators,
to
notice
Asian
SIG's
to
define
to
defer
the
compliance
process,
the
workflow
and
what
is
a
test
package,
the
compliant
test
package,
for
example,
let's
take
inception
v3
or
mobile
net.
We
okay,
we
want
to
say
here,
is
a
compliance
package
for
mobile
net
v2,
and
then
we
say
a
something
like
edge
device
which
has
a
profile
of
mobile
phone,
has
to
meet
these
test
packages,
and
one
of
them
would
be
the
motherland
v2
s
package.
B
So
how
do
we
play
this
test
package?
What
does
it
contain?
What
is
the
pass
criteria?
All
of
this
has
to
be
done
in
collaboration
with
the
other
six
and
the
zone.
I
hope
will
be.
Actually,
you
know
how
test
packages,
and
here
is
a
document
that
connects
test
packages
to
profiles
in
terms
of
compliance.
B
B
Don't
know
what
type
of
code
we'll
have
to
provide,
but
as
an
example,
let's
say
we
define
it
in
is
part
of
the
compliance
in
terms
of
accuracy.
We
define
symmetric,
you
know,
does
well-known
metrics,
like
top
k1
or
top
one
or
top
five,
but
maybe
we
will
decide
that
we
want
to
define
a
slightly
different
metric
or
we
want
to
standardize
how
top
one
is
measured
between
a
golden
output
and
the
output
that
wants
to
be
compliant
that
may
translate
to
code,
and
that
code
can
be
something
that
we
agree.
B
A
A
Guess
not
sure
if
you
want
to
you
know
close
the
meaning
right
now
or
give
a
guidance
for
the
next
one,
because
we
are
running
out
of
time.
Basically,
we
want
to
put
some
goals
and
milestones.
That
is
the
next
chapter.
What
we
can
achieve
in
the
reasonable
time
sorry
offer
to
throw,
but
still
the
meeting.
So
we.
B
B
A
Yes,
that's
what
we
captured
in
the
meeting
minutes
and
then
so
that
you
know
not
later
that
next
meeting.
We
would
like
to
define
this
and
agree
on
statement
work.
Okay,
so
you
know.
Basically
there
are
a
couple.
You
know
these
last
two
sections
goals
and
milestones,
and
maybe
we
can
do
it
next
time.
What
is
a
practical
and
achievable,
and
what
else
we
can
change
here
and
exit
criteria
is
really
another
spin
on
deliverable.