►
From YouTube: UniMatrix video conference call 2020-09-23
Description
Agenda: https://gitlab.com/unimatrix/meetings/-/blob/master/20200923_mom.md
A
A
C
B
B
A
A
B
B
A
A
B
A
A
A
A
D
A
Options
enabled
so
that
you
can
run
basically
run
containers
on
it.
So
I
haven't
I've,
looked
at
a
couple
of
this,
so
there's
a
script
for,
for
example,
for
docker
to
check
a
view.
So,
but
it's
quite
big,
it
checks
for
a
lot
of
options
which
are
not
really
required.
I
would
say
so.
We
should
make
our
own
script.
I
think
which
checks
for
the
necessaries
and
also
maybe
check
for
optional
things,
which
could
be
a
sign
that
you're
maybe
not
running
as
effective
as
you
can.
A
So
there
I
we
can
so
I
I'm
gonna
have
a
look
and
see
if
I
can
write
that
script
also
that
the
docker
script
is
not
really
so.
The
idea
here
is
that
you,
you
will
run
that
on
your
on
your
host
and
it
it
needs
to
have
the
kernel
configuration
available,
so
normally
that
is
available
through
the
proc
file
system.
So
it's
like.
So
if
you
don't
support
it,
exposing
your
kernel
configuration
over
the
proc.
You
must
provide
it
as
an
option
to
the
script.
A
E
A
E
A
E
A
A
Yeah,
maybe
yoon
can
comment
on
this.
I
think
from
the
so
right
now
at
least
the
the
the
d
bus
connection
is
needed.
I
think
it's
mainly
needed
to
negotiate
a
unix
domain
socket
so
that,
basically,
when
you
do
inference
you,
you
don't
use
d
bus
for
that.
You
use
a
basically
a
unix
domain
socket
for
it,
but
I
believe
the
initial
negotiations
to
get
that
socket.
You
need
to
get
that
over
d
bus.
F
F
A
F
A
It's
a
it's,
basically,
a
limitation,
the
dbos
protocol.
I
think,
because
there's
no
way
when
you
send
a
message
from
inside
the
container
to
when
you
change
it
it
doesn't.
You
can't
map
one
user
to
the
other
to
the
host
use
a
namespace.
Really,
I
think
canonical
has
done
a
workaround
for
it,
so
they
have
a
way
of
translating
the
user
between
the
container
and
the
host.
F
F
The
initial
the
the
the
the
socket
pair
endpoint
is
communicated
over
d
bus,
but
that
I
mean
that
that
could
be,
of
course,
be
communicated
over
a
socket
in
in
the
first
place.
So
if
we
could
have,
we
could
have
a
server
socket
open,
for
example,
and
we
could
do
a
unix
domain
server
socket
open
for
for
a
session
initialization.
F
F
F
A
A
H
G
E
F
F
F
A
E
F
F
F
F
I
don't
know
really,
I
think
we
need
to
to
look
that
over
and
try
to
estimate
that,
but
I
I
don't
we
I
think
it's
I
mean
to
just
just
implement
the
a
socket
based
session
initialization.
I
don't
think
that
is
is
particularly
advanced,
but
I
mean
we
need
to
there.
There
is
a
lot
of
things
around
the
authentication
and
so
on
that
we
need
to
think
through.
So
I
think
that
is
the
most
tricky
part.
F
A
A
F
F
A
E
Yeah
these
issues
are
are
pretty
thorny,
so
you
know
the
sort
of
the
docker
way
of
doing
things
the
docker
best
practice
on
initialization
scripts.
You
know
they,
don't
they
don't
use
system
d
for
anything
right,
so
you're,
basically
forced
into
writing
sort
of
custom,
initialization
scripts,
usually
in
bash,
it's
actually
quite
quite
painful,
but
there's
no
standard
way
of
dealing
with
this.
Everything
feels
like
a
bit
of
a
hack,
and
this
all
comes
down
to
you
know
these
these
issues
around
d
bus.
E
I
would
love
to
see
some
stop
gap
here
and
it
looks
like
red
hat
is
trying
to
provide
one,
but
it's
not
ready
to
ride
yet,
but
yeah
I
mean
if,
if
unimatrix
accepted
systemd
as
a
as
an
initialization
standard,
and
we
were
able
to
put
stop
gap
in
place
that,
I
think
would
be
great,
but
it
feels
like
the
stop
gaps
that
are
under
development.
Right
now
are
just
not
ready
for
prime.
E
B
A
A
But
for
the
future,
this
is
how
I
see
it
now,
because
that's
the
easiest
way
to
go
forward.
I
think
then
each
host
can
provide
their
something
that
works
with
opencv,
but
for
the
future,
I'd
like
to
see
a
unimatrix
driver
for
opencv,
but
it's
it's
something
for
the
future.
Then
we
will
have.
I
have
an
idea
of
using
red
is
as
a
a
cache
so
that
you
can
read
metadata
information.
Also
it
did
cut.
A
A
A
E
A
A
A
All
right,
so
my
next
task
will
be
to
kind
of
define
the
video
api
so
which,
which
subset
of
opencv
that
will
support,
which
parts
which
which
properties
that
you
can
read
for
an
image,
for
example,
so
or
which
properties
you
you
can
set.
So
we,
which
video
formats
are
supported
and
so
on,
and
also
the
inference
api
will.
A
A
F
F
Looking
at
that
as
an
example,
there
is
no
probably
no
real
need
for
the
cpu
to
touch
that
data
or
to
to
even
map
it.
So
we
are
trying
to
find
a
way
to
pass
the
data
through
this
pipeline
in
in
in
a
way
that
is
as
efficient
as
possible,
with
as
little
cpu
overhead
as
possible,
so
that
we
we
avoid.
We
try
to
avoid
the
cpu
copies
primarily,
but
also
memory,
mapping
and
cache
management,
so
that
is.
That
is
one
thing
that
we
are
going
to
start
up
now.
F
F
F
A
E
Yeah,
that's
right.
We
basically
at
the
point
here
in
our
future
releases,
where
self-service
should
be
ready
to
go
here
in
the
next
month
to
six
weeks.
We've
consistently
pushed
that
back
over
the
last
two
quarters,
because
we've
had
feature
requests
on
on
other
fronts
and
we've
just
ended
up
adding
features
to
farms
that
we're
hosting
and
managing
for
the
end
customer,
but
yeah
self-service
is
coming
back
to
the
top
here
in
q4,
so
yeah.
E
E
F
Yeah
we
we
will
try
to
to.
We
will
try
to
upload.
We
will
upload
what
we
had
today
on
our
local
master,
and
that
was
the
things
that
are
part
that
will
be
part
of
that
release
was
the
things
that
I
described
on
the
lat
on
on
the
call
during
the
summer
and
it's
the
pri.
I
think
the
primary
the
primary
features,
the
prior
primary
features
of
that
is
our
pre-processing
support.
F
F
F
F
F
A
A
F
You
have
a
larynd
with
with
the
tensorflow
lite
the
the
edge
tpu
tensorflow
lite
back
end
enabled
that
one
will
run
on
a
pc
with
a
u
an
edge
tpu
usb
stick.
We
we
have
done
a
lot
of
development
using
that
setup.
Actually,
so
we
do.
We
do
a
lot
of
our
layout
framework
development,
the
core
framework
development
on
on
on
our
host
computers,
using
such
a
setup.
F
F
A
F
Yeah,
I
could
just
comment
on
that.
We
we
need
to.
We
still
need
a
few,
I
think,
a
few
a
few
weeks
to
be
able
to
get
up
and
running
with
more
continuous
releases
to
to
get
lab,
but
we
we
are
working,
we
that
that
is
work
in
progress
here
at
axis
now.
So
I
expect
that
to
happen
within
some
weeks
from
now.
F
A
E
A
B
A
F
F
D
F
D
Meeting
if,
in
fact
in
fact
as
a
our
unimatrix
goes
a
lot
larger
and
larger,
I
think
more
and
more
new
comers
are
very
interesting
to
join.
And
I
don't
know
I
don't
know
you,
but
we
we
got
some
contact
with
the
ews,
and
maybe
we
are
very
like
to
here
to
to
to
to
say
how
they
can
do
something
with
with
this
on
ice.
So
you
mean
from
amazon.
A
D
Yeah
we've
talked
to
amazon
as
well,
so
it's
but
yeah,
but
I
think
from
different
parts.
I
we
we
connect
with
amazon
asia
and
they
have
some
specific
department
for
new
solution.
Development
like
they
say
their
goal,
is
to
seek
out
some
new
technology
new
possibilities
and
join
the
merging
them
into
into
their
current
solution
counter
architecture.
So
it's
maybe
it's
a
different
view
from
them
to
to
their
headquarters.
So
maybe
I
don't
know.