►
From YouTube: Unimatrix members conference call 2020-05-20
A
A
C
A
C
C
Where
is
my
point?
Okay
here
is
a
road,
and
here
is
some
pre-application
over
here,
and
this
example
first
gets
an
image
from
the
base
formula
and
send
to
the
road
and
gets
back
the
results
from
la
road
and
it's
over
here
it's
a
metadata
generator.
We
I'm
going
to
create,
develop
and
it
sends
metal
inference
metadata
into
the
video
three
minute
sender.
C
D
C
B
B
No
I'm
talking
to
how
the
container
is
communicating
with
the
sender,
so
I
at
access,
we
have
started
to
look
at
using
an
internal
broker,
mqtt
broker
for
just
communicating
between
the
containers
in
the
system
and
then-
and
if
you
have
an
external
broker,
that
also
you
you
can
do
a
broker
to
broker
breach.
So
then
you
can
bridge
some
of
the
events
or
some
of
the
metadata
that
you
want
to
publish
externally,
and
this
is
all
standard
components
that
are
already
existing.
D
B
We
have
another
proposal
on
this
topic
actually,
which
is
to
use
protobuf,
which
is
a
it's
using
a
schema
and
also
backward
compatible.
So
you
you,
don't
have
to
use
the
same
versions
in
the
in
the
consumer
and
the
producer,
so
it's
quite
flexible
and
it's
binary,
so
it's
very
much
more
efficient
than
using
xml,
so
maybe
in
before
we
can
share.
Actually
some
basic
protobuf
schema
to
see
if
it
suits
your
needs.
It's
a.
B
F
F
B
I
still
want
the
binary
format,
so
I
mean
it.
It
makes
sense
to
have
json,
but
still
it's
not
really
efficient.
If
it's
large,
let's
say
you
have,
including
you
know,
vectors
like
result,
vectors
of
the
inference
or
images
or
stuff
like
that
in
your
in
your
metadata,
then
you
have
to
basically
encode
it
in
json,
so
I
would
like
to
have
a
protobuf
format
anyway,
but
yeah,
I
see
what
you
mean,
but
I
think
we
should
start
a
discussion
also
on
a
binary
form.
F
F
B
G
B
F
B
C
C
C
H
H
H
The
reason
for
presenting
it
is
because
I
thought
that
it
would
be
necessary
to
understand
what
kind
of
system
to
build
as
a
whole
system,
including
cloud
when
proceeding
with
each
of
each
company
and
hyper
would
like
to
continue
to
update
the
figure
of
grand
design
based
on
the
progress
of
study.
Each
time
we
found
items
and
second
item
about
the
and
time
schedule.
H
C
Let's
click:
let's
let
me
confirm
about
the
time
schedule
we
searched
the
inside
is
a
guitar.
I
didn't.
I
couldn't
find
some
kind
of
document
regarding
the
that
describes
the
time
schedule.
Timescape
zero
reason
means
about
our
tasks
for
one
or
two
years,
do
you
have
some
time
schedule
already
created.
B
I
think
we
early
on
we
kind
of
made
a
rough
schedule.
I
think
it
was
in
the
hong
kong
meeting,
maybe
yeah,
but
I
think
we
said
maybe
end
of
summer.
We
would
have
some
kind
of
pre-release,
but
it
depends
on
what
you
include
if
what
you
want
to
include,
because
at
that
time
I
think
we
were
discussing
only
around
the
inference
engine
and
the
video
api
and
the
deployment
api.
B
C
C
C
B
C
C
B
Yeah
we
we
have
done
a
driver
for
opencv
which
does
zero
copy
to
to
access
hardware.
So,
but
this
is,
this
is
specific
for
access
so,
but
I
think
we
can
open
source
it
so
that
we
can
share
ideas
with
other
companies
of
how
they
can
do
similar
things
with
their
hardware,
because
this
is,
I
guess,
very
hard
specific.
B
B
I
I
B
B
I
H
B
B
B
B
C
J
J
C
Let
me
confirm
about
the
cube
edge
and
the
battery
giveaway.
I
have
not
not
studied
the
project
secret
yet,
but
I
wanted
to
know
the
difference
in
terms
of
functionality
or
feasibility
between
in
case
when
we
select
the
qubits
and
in
case
when
we
select
the
virtual
coupes,
especially
for
the
functions
that
might
be
required
or
useful,
with
uni
metrics.
B
C
And
I
think
if
the
cubbage
has
attractive
functions,
I
don't
know,
but
it
will
be
better
if
developers
can
select
the
one
from
the
alternatives
which
they
need
coverage
or
bachelorette
or
some
others.
I
don't
know
by
comparing
their
functions,
available
and
unavailable
for
each
and
also,
of
course,
by
comparing
the
required
hardware.
Resources
is
very
important.
H
J
B
G
I
I
I
just
have
one
question
on
your
presentation:
is
that
you,
you
mentioned
a
little
bit
the
resources
needed
for
for
for
this,
and
this
is
your
some
kind
of
a
minimum
resource
or
it's
a
target,
camera
resources.
Or
what
do
you
mean
target
camera
resources
is
a
something
you
want
to
do
for
this
kind
of
capacity
as
a
minimum,
or
it's
it's
like
a
standard
for
for
some
kind
of
application.
G
E
C
G
C
G
Yes,
so
yeah,
so
that's
why
that's
why?
What
I
ask
is
that
it's
a
standard
resource
or
minimum,
because
if
we
say
minimum
resources,
that
means
maybe
just
only
one
container
installed
into
a
camera.
So
this
this
should
be
considered
as
a
minimum
resource
right,
but
if,
if
multiple
container
can
be
installed
into
one
camera,
it
also,
it
depends
on
some
kind
of
camera.
Specification
and
capacity
and
different
cameras
have
a
different
thing
to
do
so.
It's
a
yeah,
it's
a
yeah
just
have
some
idea
to
to
to.
I
B
A
B
More
detailed
breakdown
of
this
300
megabytes
what
it
is,
and
so
we
I
mean
we
know
that
container
d
takes
such
and
such
amount.
We
know
that
if
we
want
to
include
cube
edge,
it
takes
a
100
megabytes,
and
so
it
would
be
good
to
have,
and
we
know
that
opencv
is
the
size
and
so
on,
so
to
have
some
kind
of
budget,
that's
as
to
see
where,
where
we,
where
we
are
right
now.
H
C
H
A
A
G
B
I
mean
we
can
do.
We
can
do
a
very
short
presentation
of
how
it
looks
for
us
right
now
with
the
resources
that,
but
we
that
includes
docker,
as
I
said,
which
is
huge,
but
it
it
gives
you
kind
of
a
rough
estimate
of
what
it
takes
and
then
I
think,
probably
panasonic.
It
would
be
great
if
you
also
could
do
a
breakdown,
and
then
we
can
just
share
those.
E
B
B
A
I
B
B
A
G
G
Most
manufacturers
like
axis
panasonic
and
the
hydration,
or
even
we
can
contribute
a
lot
of
that.
That
part.
We
have
the
most
experience
right.
So
it's
it's
a
it's.
Is
it
possible
that
we
can
break
break
down
topic
and
this
kind
of
really
topic?
We
can
make
like
a
a
a
a
better
timeline
meeting
between
europe
and
asia
and
all
the
other
topics.
For
example,
if
we
discuss
a
little
bit
cuba
age
cloud
side,
I
think
microsoft
huawei
and
they
have
more
experience
on
that
part.
G
A
B
B
C
C
B
B
D
D
I
Yeah,
a
couple
of
quick
issues:
can
you
guys
hear
me?
Okay,
yep,
okay,
great
yeah?
We
we
would
really
like
to
figure
out
how
to
kind
of
start
a
test
and
evaluation
focus
group
for
the
project,
since
that's
really
kind
of
our
area
key
area
of
interest.
I
noticed
there
are
some
line
items
on
the
the
list
that
panasonic
presented
that
are
related
to
tests
and
evaluation.
I
So
maybe
you
know
we.
We
had
a
nice
discussion
with
the
folks
at
access
this
morning
about
tne
and
and
how
we
can
support
tne
for
access.
So
maybe
the
thing
for
us
to
do
is
follow
up
with
all
the
other
teams
about
their
specific
objectives
and
and
and
go
about
kind
of
building
out
the
t
e
plan.
That
way,
then,
maybe
we
can
circle
back
during
the
hackathon,
with
a
more
comprehensive
proposal,
for
you
know,
sort
of
end
to
end
tne
within
unimatrix
project.
I
B
It's
a
really
it's
a
really
important
topic
so,
and
I
mean
in
the
long
run,
if
you
want
this
platform
to
be
successful,
the
testing
part
is
going
to
be
very,
very
important
because
we're
going
to
have
quite
diverse
hardware
so
to
be
able
to
you,
know,
be
to
to
not
have
get
bad
will
with
the
market.
We
really
need
to
to
do
good
testing,
because
if
you
do
an
application,
you
want
to
know
if
that's
gonna
run
on
all
the
devices
out
there
or
if
there's
a
problem.
B
Yeah,
so
this
is
really
I
mean
this
is
something
I've
been
thinking
about,
but
we
haven't
really
discussed
it
that
we
need
test
suites
for
everything.
So
at
this
point
you
know
we
have
the
inference
engine
and
so
we're
gonna
have
like
a
subset
of
opencv,
of
course,
opencv.
It
had
its
own
test
suite,
so
I
mean
just
setting
up
stuff
to
run
the
the
automated
test
which
that
are
already
there
should
be
something
that
is
also
think
about
how
how
do
we
test
the
firmware?
B
Because
there
is
some
there
is
some
requirements
on
the
firmware.
So
how
is
the?
How
do
we?
How
do
we
con
confirm
that
the
firmware
is
is
conformant?
How
do
we
confirm
that
a
container
application
is
conformant
stuff
like
that,
so
I
think
a
proposal
of
how
to
test
things
the
different
things,
the
firmware,
the
application
containers,
the
base
layer
like
a
proposal
on
the
strategy
of
testing-
I
guess,
would
be
really
nice.
I
Okay,
great
yeah,
we'll
put
that
on
our
to-do
list,
not
sure
we
can
deliver
on
it
before
the
hackathon,
but
it'll
certainly
be
a
work
in
progress,
and
then
you
know
I
do
think
it
maybe
makes
sense
for
us
to
follow
up
with
each
of
the
independent
teams.
Just
to
kind
of
you
know,
we
don't
have
kind
of
connections
with
each
of
the
individual
performer
teams.