►
From YouTube: Unimatrix Technical Board call 2021-07-21
B
Today
so
our
prgcv
members
are
joined
from
eyebrow
site.
B
D
B
E
C
A
Okay,
so
the
idea
for
today
was
to
continue
the
discussion
from
last
month
about
first
the
components
and
then
around
orchestration
and
also,
if
we
have
time,
maybe
discuss
a
little
bit
about
a
development
environment
like
id
and
so
on.
B
From
our
precise,
so
I've
already
said,
email
to
fredericksen
and
so.
B
We
should
decide
the
actual
items
and
assignee
and
due
date
about
the
technology.
A
A
Okay,
so
let's
dig
into
the
agenda
so
regarding
the
first,
the
host
requirements.
So
I
don't
know,
there's
been
discussions
of
replacing
container
d
with
something
else
and
I
don't
want
to
get
into
those
kind
of
discussions,
at
least
for
the
reference
implementation.
This
the
container
day
is
like
the
de
facto
standard,
and
that
is
what
we're
going
with.
A
F
D
A
F
Because
I've
been
doing
some
research,
I
also
sent
a
link
while
answering
to
the
to
the
survey
I've
been
expecting
oc.
I
containers,
so
you
can
create
the
extremely
small
ones,
but
really
extremely
small
ones
in
terms
of
300
kilobytes.
F
So
this
is
something
that
can
be
make
a
huge
difference
for
embedding
these
things.
These
algorithms
in
devices.
A
F
A
That's
not
my
priority.
My
priority
is
not
to
make
this
fit
into
32
megabytes
ram
or
something
it's
it's.
I
don't
want
to
optimize.
I
mean
other
people
are
working
on
making
these
things
smaller.
So
when
those
things
are
ready,
we
should
adopt
them,
but
right
now
just
take
what
it
off
the
shelf
stuff
and
and
concentrate
on
on
building
other
things.
F
A
Yeah,
I
understand,
of
course,
but
it
is,
it
is
a
trend
in
the
industry
or
moving
container
down
to
the
edge.
So
it's
not
only
you
know
this
product
looking
at
those
things,
so
hopefully
we'll
get
help
from
other
places,
but
yeah
I
mean
I
think
we
should
use
the
industry
standard
like
we.
I
think
it's
even
in
the
in
our
vision
that
we
use
industry
standards
when
where
it
makes
sense
and
when
it's
available
and
right
now
container
d
is
the
standard.
A
A
E
E
Yes,
so
we
could
have
a
continuous
monitoring
of
what
is
happening
here.
So
we
understand
how
the
development
is
progressing,
but
I,
under
I
agree
with
frederick
because
container
is
used
so
widely
that
there
will
be
development
to
make
it
smaller.
B
So
the
according
to
high
pros
investigation.
B
So
if
we
use
a
cryo
and
so
the
lump
size
can
be
a
20
megabyte
smaller
than
container
d
in,
however,
so
I
don't
know
about
the
compatibility
about
a
vetoing
in
the
continent
and
the
trial
so
hansa,
could
you
share
the
knowledge
about
the
contingency
and
cryo.
D
D
G
Can
you
see
my
screen
yeah?
Okay?
So
yes,
that
I
I'm
trying
to
create
a
comparisons
when
run
the
continuity
with
the
one
c
and
the
cryo
with
the
c
run
and
here's
the
memory
consumptions
between
toothpaste,.
G
A
G
Yes,
I
also
test
this
with
the
continuity
and
yeah
q1
is
the
totally
one
with
the
continuity.
G
So
the
point
here
is
that
I
want
to
is.
I
want
to
show
you
that
the
cluster
with
using
the
cryos,
I
think
is
this
will
consume,
is
a
less
result
than
the
container
continuity.
G
So,
for
example,
you
can
see
that
on
the
cloud
on
the
crease
here
in
cluster
that
using
the
continuity
and
run
c's
it,
we
took
the
around
the
200
megabytes
grams
and
on
the
cryos
and
silent
it
only
took
about.
G
150
and
I
think
we
should
consider
the
continuity
and
crimes
to
select
to
choose
which
one
which
one
is
the
default
on
our
host
device.
D
G
So
so,
yes,
that's
also.
I
want
to
share
about
the
continuity
and
the
cryos
that
we
test
on.
A
A
A
Or
possibly
as
much
smaller
demon
running
to
do
to
pull
containers
and
so
on
on
the
camera,
not
not
something
huge
like
kubernetes,
we
don't
need
all
those
things.
Kubernetes
provides,
at
least
for
our
use
cases.
It's
more.
We
don't
really
really
need
for
our
cases,
full
orchestration
capabilities
like
load,
balancing
and
such
things
and
migration
of
workloads,
and
it's
more
from
our
point
of
view
when
it
comes
to
orchestration.
A
A
D
A
A
D
B
So
maybe
so,
first
of
all,
so
we
have
to
decide
whether
the
quantities,
for
example
k3s,
should
be
introduced.
A
kubernetes
agent
should
be
introduced
on
the
each
side
or
not
is
my
understanding,
correct.
A
Yes,
I
think
today
it's
like
that,
because
I
think
you
can
still
use
docker,
maybe
a
few
other
run
times
as
well,
but
the
default
I
would
say
now
is
to.
D
H
Yeah
I've
been
looking
at
these
ecosystems
quite
a
bit
lately
for
our
own
use.
I
mean
one
of
the
challenges
right
is
that
not
everything
is
compatible
with
cube
adm
based
distributions
of
kubernetes?
H
So
if
you,
you
know
you
bite
on
something
like
micro,
k8s
or
k3s,
or
one
of
the
related
slim
down
kubernetes
distributions
you,
basically
you
know:
you're
you're
not
compatible
with
with
cube
adm
central
services
running
on
on
the
cloud,
and
so
if
cloud
to
edge
is
a
key
piece
of
the
puzzle
for
unimatrix,
then
you're
kind
of
hosed
unless
you're
willing
to
run
a
cube,
adm
based
distro
at
the
edge
as
well-
and
there
may
be-
you
know
there
may
be
some
fixes
to
that
in
the
pipe
for
at
least
for
k3s,
and
it
is
possible
to
conceivably
fork
k3s
and
implement
support
for
kube
adm
upstream,
but
it
would
be,
it
would
appear
to
be
a
fair
amount
of
work.
H
So
that's
that's
something
to
think
about.
I
hate
to
add
one
more.
H
You
know
one
more
concern
to
the
kubernetes
selection
piece
of
the
puzzle,
but
it
seems
like
kube
adm
on
the
cloud
is
sort
of
the
standard.
H
H
I
guess
the
other
major
concern
and
a
reason
that
we've
been
looking
at
podman.
Quite
a
bit
is
standardization
around
startup
scripts
and
things
of
that
nature.
This
sounds
sort
of
trivial
right.
I
guess
from
an
embedded
perspective.
H
You
know
people
aren't
thinking
worrying
too
much
about
hey.
What's
your,
you
know,
what
are
your
scripting
entry
points
to
your
containers
but
from
a
developer's
perspective,
especially
if
we
try
to
attract
more
naive
developers
who
maybe
are
just
getting
their
feet
wet
with
this
technology,
then
you
know
having
things
that
work
with
systemd
out
of
the
box.
Is
a
very
nice
convenience?
H
A
H
H
H
That's
a
major
advantage
of
podman.
In
my
opinion,
it's
obviously
not
optimized
for
edge
at
all.
A
H
Yep
yeah
so
podman
has
you
know
its
own
compose
alternative.
That's
supposed
to
be
backward
compatible
with
compose
yeah,
I
mean.
Sadly,
I
don't
think
there
is
a
good
answer
right
now.
Frankly,
the
this
just
you
know
this
ecosystem
is
too
fragmented
for
there
to
be
a
good
choice
right
at
the
moment.
H
You
know
cloud
cloud
kubernetes
distributions?
Are
we
optimizing
for
small
footprint
size
on
embedded
devices
or
are
we
optimizing
for
you
know
the
naive
developers
that
we
might
hope
to
attract
to
the
project
somewhere
down
the
line.
Those
are
really
the
three
choices
and
each
one
has
a
distinct
answer.
A
More
of
footprint
than
I
I
don't
think
I
mean
the
kind
of
developer
we
are
focusing
on
is,
I
guess
the
cloud
native
developer
already
bit
used
to
containers.
So
it's
not
the
naive
developer.
I
would
say
fair
enough.
That's
fair!.
H
Yeah,
in
that
case,
if
you're
really
hoping
to
to
attract
a
the
cloud
developer,
then
I
think
the
right
answer
is
to
do
things
that
are
consistent
with
kubereum-based
kubernetes
distributions.
If
anything,
kubernetes
related,
because
they're
going
to
expect
a
degree
of
portability
between
the
cloud
and
the
edge
that
I
don't
see
yet
in
k3s
or
micro,
kds.
H
A
We
have
been
looking
at
making
a
very
small
docker
compose
compatible
launcher
that
is
without
even
using
dockers,
where
you're
just
using
container
d.
So
it's
it's
compatible
with
compose
and
you
can
so.
You
can
kind
of
structure
application
in
a
nice
way,
but
still
the
footprint
is,
is
quite
small
compared
to
the
full-blown
walk-around
composition.
A
A
A
Says
so,
let's
move
on
to
d
bus?
That's
I
guess.
A
An
interesting
topic
as
well
so
currently
lara
is,
has
that
as
a
requirement,
and
but
I
I
was
thinking
we
could
probably
provide.
A
I
mean
there's
two
choices
here.
One
is
to
change
the
large
code
to
another
pro.
A
A
So
that's
that's
one
choice,
at
least
in
the
meantime
until
we
have
if
we
want
to
change
to
another
implementation,
so
I
was
going
to
have
a
look
to
try
to
do
that.
Using
the
red
hats,
bus,
one.
A
E
I
asked
google,
it
said
about
half
a
meg,
so
478
kilobytes,
according
to
google
yeah,
if
you're
just.
E
A
E
Do
we
have
a
list
of
tasks,
so
we
could
see
who
is
working
on
what.
E
Well,
maybe
it's
a
good
idea
to,
since
we
are
talking
more
specifically
now
about
tasks,
we
could
list
them
and
then
each
company
could.
If,
if
I
choose
to
work
on
a
task,
they
could
put
the
name
and.
A
Yeah
I
can
update
that
at
least
anyway.
So
so
what
I
will
do
is
go
back
to
kind
of
where
we
started
and
do
the
reference
implementation
using
york,
though,
and
add
those
require
host
requirements
and
then
make
a
container
layered
with
opencv
and
so
on
and
plus
an
example.
So
we
have
something
that
you
can
actually
show.
B
No,
I
will
have
it
on
on
hyper
camera.
E
B
So
so
far,
according
to
our
discussion,
our
internal
discussion,
so
the
architecture
change
is
not
a
welcome
for
eyebrow
and
therefore
so
we
are
looking
for
the
way
of
working
and
route
without
debuff.
B
Therefore,
so,
if
we
need
to
add
the
debuss
broker
on
ipro
camera,
so
I
can't
answer
it's
okay
in
this
timing
and
therefore
so
I
have
to
discuss
that
scene
with
hyper
internal
members.
E
Hey,
please
do,
but
I
must
I
must
have
to
make
sure
that
you
understand
that
this
is
not
adding
it
to
your
camera
system.
It's
adding
it
to
the
container.
So
it
doesn't
affect
your
camera
system
you're,
just
adding
the
debuss.
The
broker
would
only
be
available
inside
the
container
for
running
and
separate
container
as
well.
Maybe
that's
better
so
it
doesn't
affect
or
cure
your
product.
In
other
terms,
you
don't
have
debuffs
in
your
system,
just
in
the
container.
A
C
Okay,
oh.
I
B
So,
in
addition,
so
if
we
we
introduce
devas
blogger
into
the
container
in
that
case,
so
unimatericus
project
provides
the
developers
the
sdk.
I
Yeah,
I
think
they
need
time
to
see
if
the
debuff
broker
can
imply
can
be
implemented.
It's
very
small
right.
I
A
A
There
are
other
things
for
doing
starting
other
demons
based
on
d
bus
signals,
and
so
on
that
you
don't
need
all
those
things
so
yeah.
I
think
that's
that
single
binary
is
is
only
for
500k.
I
B
So
I
understand
so
the
insulation
size
is
very
small.
So,
however,
so
my
question
is
so
the
yeah:
how
will
we
provide
to
the
developer.
B
The
way
of
creating
austria
image,
so
we
will
share
the
builder
tool
to
the
developers
or
the
reference
on
the
github,
which
one
is
better.
A
So
I
think
we
will
have
an
sdk
container
right,
so
you
would,
and
that
container
has
all
the
dependencies
and
then.
B
I
also
think
so.
Unimatrix
project
should
provide
the
sdk
to
the
developers,
because
so
they
don't
understand
how
to
create
the
oci
image
with
the
debug
broker.
Individual.
A
E
Yeah,
so
it's
transparent,
that's
what
you're
saying
frederick
they
don't
have
to
understand.
If
your
system
has
d
bus
or
not
the.
A
A
A
G
I
got
a
bunch
of
questions
so
with
the
bus
brokers
that
were
installed
by
the
container.
So
do
we
have
to
care
about
the
system
d.
G
A
Lara
doesn't
have
doesn't
have
a
dependency
on
system
ds,
so.
G
Yes,
because
the
the
current
version
of
narrows
that
I
think
it
still
needs
systemd
to
control
the
device.
A
A
D
A
That's
also
something
I
will.
I
will
look
at
just
to
create
some
a
default
configuration
for
opencv.
That's.
It
will
include.
A
Video
for
linux,
backend,
plus
ffm
and
then
some
of
the
default,
the
the
most
common
features
like,
of
course,
the
base
and
then
there's
a
high
level
api.
And
but
when,
when
we.
A
And
then,
depending
on
how
you
make
your
application
yeah,
depending
on
which
libraries
you
actually
use
some
of
those,
I
guess
will
be
stripped
away.
So
it's
not
it's
not
guaranteed
that
everything
will
be
included.
I
guess
but
yeah.
That's
that's!
Roughly
the
size
of
opencv
15
yeah
as
a
small,
quite
minimal.
A
A
A
B
So
I
don't
understand
so
your
intention,
so
so,
according
to
your
explanation,
so
the
video
apis
as
the
the
video
api
standard,
so
opencv
is
suggested.
B
A
We
we
had
to
write
our
own
back
and
for
opencv,
because
we
don't
support
any
of
the
existing
back-ends.
I
mean
there
is
back-ends
for
android.
There
is
back-ends
for
video,
for
linux,
for
ffmpeg
and
other,
but
axis.
Doesn't
we
don't
support
video
for
linux,
so
we
had
to
write
our
own?
It's
not
it's
not
so
difficult,
but
at
least
it
says
so
we
want
to
include
that
in
the
opencv
as
well.
It's
it's
a
small
piece
of
code.
I
guess
so
my
question
is:
do
you
is?
A
I
I
Okay
and
also
for
open
api,
I
think
maybe
we
should
specify
which
apis
need
to
be
mandatory
and
which
apis
are
optional,
and
this
kind
of
specification.
I
A
I
A
Yeah
I
mean
we
could
all
I
mean
we
could
all
certainly
have
a
look
and
then
we'll
just
make
a
compromise.
What
what
people
think
are
important?
I
mean
there's
a
huge
set
of
variables
to
you
can
configure
because
it's
such
a
huge
framework,
but
so
the
I
think,
but
you
have
to
compromise
and
make
kind
of
because
otherwise
it
can
become
quite
huge.
A
So,
for
example,
opencv
has
feature,
detection
has,
or
has
I
mean,
there's
so
many
algorithms,
which
ones
are
needed.
I
there's
like
deep
learning
and
there's
like
a
lot
of.
A
A
And
then,
if
you
want
to
publish
something
to
an
external
broker,
you
can
set
up
rules
in
the
broker
which
messages
to
relay.
So
I
think
also.
We
should
provide
a
container
mqtt
broker.
A
A
All
right,
I
guess
there
was
no
input
on
that.
Maybe
we
should
discuss
this
a
little
bit
about
security
and
orchestration,
so
I
noticed
that
both
ipro
and
hikvision
is
not
happy
about
having
ssh
enabled,
and
I
mean
I
would
be
surprised,
because
this
is
how
you
normally,
if
you
I
mean
if
you
used
any
cloud
platform,
it's
always
ssh.
That
is
the
control
channel
some
and
but
it
all
depends
on
how
you
secure
ssh.
Of
course,
you
don't
want
to
log
in
there
with
root
and
have
full
access
to
everything
and
so
on.
A
So
I'm
a
little
I'm
a
bit
surprised
that
it's
it's
like
saying
that
no,
we,
we
can't
use
tcp
because
it
can
be
attacked
by
or
it's
I
see
it
just
as
a
fundamental
infrastructure.
B
By
doing
so
so
the
security
is
established
and
therefore
so
if
we
need
to
be
accessed
to
the
camera
via
ssh
by
the
end
users.
So
in
that
case,
so
we
would
like
to
know
how
how
to
keep
the
security
from
the
attack.
A
Yeah,
okay,
I
understand,
then
then
we
can
explain
how
we
are
using,
how
we
are
doing
that
with
what
how
we
are
doing
that
with
ansible,
but
the
person
from
axis
who
will
do
he's
not
here
right
now
he's
on
vacation,
so
hopefully,
next
time
we
can
do
a
small
presentation
around
that.
It's
a
token
based
authentication
that
gives
certain.
B
Sir,
so
could
you
please
summarize
the
actual
items,
for
example
in
the
excel
sheet
and
so
share
the
attendees
all
right
and
in
addition,
so
it
seems
we
have
a
lot
of
things
to
discuss
about,
discuss
the
uni
matrix
and
therefore
so
so
it's
better
for
us
to
have
the
tv
call
more
short
spa.
A
B
A
All
right,
okay,
I'll
summarize,
I'm
not
gonna
summarize
in
excel.
I
put
it
in
the
agenda,
as
I
mean
it's
in
this
in
this
agenda
that
I
sent
out
it's
also:
okay,
okay,.