►
Description
In this episode, Christos Matskas (@ChristosMatskas) joins us to share the story of an interesting application he built using the Azure SDKs for .NET and Cognitive Services. We not only get an overview of creating custom vision models, but also a demo of the docker containers for cognitive services. Christos also shares how he was able to leverage .NET standard libraries to maximize code portability and re-use.
[03:27] - Overview of Cognitive Services
[05:05] - Discussing AI Visual Provision Sample App
[06:40] - azure sdk for .net and .net standard
[08:58] - Options for deploying a custom vision model
[11:55] - container support for cognitive services
[19:10] - High level architecture breakdown
Useful Links
Cognitive Services
https://azure.microsoft.com/en-us/services/cognitive-services/directory/?WT.mc_id=ondotnet-channel9-cephilli
Cognitive Services in containers
https://azure.microsoft.com/en-us/blog/getting-started-with-azure-cognitive-services-in-containers/?WT.mc_id=ondotnet-channel9-cephilli
Visual Azure Provisioning from a Whiteboard
https://channel9.msdn.com/Shows/XamarinShow/Visual-Azure-Provisioning-From-a-Whiteboard?WT.mc_id=ondotnet-channel9-cephilli
AI Visual Provision Sample App https://github.com/Microsoft/AIVisualProvision
.NET Libraries for Azure
https://github.com/Azure/azure-sdk-for-net
A
A
Hey
welcome
to
another
episode
of
the
undead
show
today
we're
going
to
be
talking
to
Christos
and
he's
been
building
apps
for
Azure
starting
from
scratch,
and
we
thought
you
know
you'd
be
a
good
person
because
there's
probably
a
lot
of
customers
out
there
that
are
starting
from
scratch
and
I.
Think
you
know
you
told
me:
you
started
with
just
our
documentation
and
our
samples,
so
how
the
heck
did
you
do
that?
Well,.
A
B
B
It
was
a
beautiful
idea
and
everybody
said:
can
we
be
done
to
begin
with,
so
my
first
goal
was
to
create
a
proof
of
concept,
but
at
the
point
that
I
started,
the
project
I
only
had
a
very
vague
idea
of
what
services
or
how
we
could
approach
solving
the
problem
and
to
do
image
analysis
as
I
found
early
on
the
process.
It's
very
very
hard
like
if
you
want
to
do
it
yourself.
B
I
mean
it's
doable,
but
it's
very
very
hard.
You
need
a
lot
of
resources
and
at
that
point,
I
thought
there
must
be
a
better
way.
Ya
know.
Better
people
have
tackled
this
problem
better
than
me,
I'm,
pretty
sure
and
I
came
across
a
are
cognitive
services.
Now
of
you
use
cognitive
services
before
for
smaller
things,
but
never
for
images.
So,
okay,.
A
B
Was
how
do
I
do
it
and
okay,
the
documentation
was
fantastic.
Like
you
start
from
scratch,
you
deploy
the
resource
another
and
then
the
next
thing
you
need
to
do
is
and
decide
which
service
you
want.
We
have
a
couple
of
competing
services
and
different
variations
of
these.
So
we
have
the
computer
vision,
which
can
do
analysis
on
images
to
return
tags
and
what-have-you.
B
A
B
A
B
Good
yep
I,
my
name
is
christos,
much
guess:
I
work
in
them
as
a
protect
marketing
manager
for
dev
tools.
Anything
has
to
do
with
mobile
gaming
and
other
detling.
So
anytime,
you
use
our
tools
to
do
stuff
on
Azure
we
have
Visual
Studio
or
the
command
line.
Then
that's
the
staff
highmarket
okay
and
as
a
developer,
I'm
really
passionate
about
these
things
and
I
always
like
to
try
them
out,
and
this
project
gave
me
the
opportunity
to
again
bring
some
of
these
things
together.
We
have
addressed
to
you,
okay,.
B
A
B
B
Wasn't
pump
sweating,
but
it
was
definitely
a
case
of
doing
a
lot
of
research
trying
to
find
out
what's
out
there
in
the
dotnet
space,
especially
because
I
was
doing
a
summary
nap.
We
thought
Ness
I
wanted
to
see
how
that
would
run
and
and
then
I
found
I.
You
know,
I
came
across
their
cognitive
services
and
I
thought
there
must
be
something
in
there.
They
can
help
us
out
and
we
had
two
problems
to
solve.
B
B
B
With
the
samurai
knob
that
we
created
and
then
we
use
cognitive
services
to
analyze
that
image,
it
returns
back
a
list
of
services
and
then
we
take
those
and
say:
would
you
like
to
deploy
them
to
Azure
now
and
then
you
press
the
deploy
and
then
it
happens
automatically.
So
you
go
from
whiteboard
into
Roger
and
let's
say
your
brainstorm
of
your
team
and
you
want
to
know
you
know
how
much
is
this
going
to?
How
is
it
going
to
look?
B
B
And
the
app
is
actually
open
source.
So,
if
I
switch
to
github,
we
have
everything
for
you
to
try
out.
We
even
gave
you
the
resources.
You
can
go
and
train
your
own
model,
but
we
thought
that
this
would
be
probably
too
much
effort.
So
we
have
a
public
endpoint
that
you
can
use
right
now
to
run
the
app
against
our
cognitive
services,
which
we
train
as
well.
So
this
is
all
open
source
right.
B
We
love
to
open
sourced
off
and
we
accept
contributions
and
any,
and
if
anyone
has
ideas
or
they
want
to
extend
the
solution
and
we're
more
than
happy
to
accept
them
cool
and
again,
you
said:
how
do
you
do
that?
How
do
you
Asha
right?
So
that's
the
other
thing.
If,
when
it
comes
to
deploying
things
to
Azure,
we
have
lots
of
different
avenues,
but
we
wanted
to
do
everything
dotnet,
and
for
that
we
have
an
amazing
SDK
enough.
Everyone
knows
that
right.
B
A
Know
install
this
thing
onto
my
box
and
then
somehow
visual
studio
would
know.
You
know
c
program
files
will
applause
where
this
assembly
was
gonna
come
from,
but
at
some
point
there
was
kind
of
a
transition
made
a
few
years
ago,
and
now
it
just
seems
to
be
mostly
new
get
delivery.
Absolutely,
and
I
you
know
now,
I
just
I
was
actually
ready.
A
I
was
telling
you
earlier
about
this
tool
that
I
was
writing
I
think
on
the
weekend
or
something
and
you
know
I
just
go
to
nougat
org
type,
Azure
storage
or
whatever
it
comes
up
just
actually.
Now
they
have
the
nice
done
at
CLI,
yep
thing
I,
just
literally
copy
and
paste
that
into
my
terminal
window
and
I'm
just
off
to
the
races,
correct.
B
A
A
B
So
dotnet
stander
was
something
that
we
focused
a
lot
and
the
class
library
that
we
created
to
drive
the
whole
business
logic
was
based
on
the
standard,
which
means
that
we
can
take
that
out
of
the
mobile
app
put
inside
a
faction
and
run
it
and
we'll
do
the
exact
same
thing
put
in
a
website
and
we'll
run
exactly
the
same
if
you're
running
a
snake
or
or
put
in
a
desta
plan
behind
WPF,
and
it
will
still
run
and
consume
it.
In
the
same.
A
B
Going
the
the
other
thing
I
want
to
talk
about
is
that
when
we
were
looking
at
how
we
were
going
to
create
the
backend
for
mobile
app,
we
had
a
couple
of
choices
and
for
this
design
path,
we
decided
to
make
everything
self-contained
within
the
phone,
so
the
doulton
standard
library
drives
the
business
logic.
We
call
to
the
cognitive
services
which
is
running
on
a
juror,
and
then
we
get
the
results
back
and
deploy
to
a
drove
via
the
azure
sdk
job.
A
B
Be
a
mobile,
app
or
a
web
app
again
and
I
think
down
to
the
constraints
of
the
project.
We
wanted
to
make
everything
self-contained
and
people
be
able
to
carry
their
phone
and
run
it
with
them,
and
we
could
have
taken
one
step
further.
So,
if
I
switch
to
my
to
my
custom
vision
project,
when
you
create
a
brand-new
project,
you
actually
have
the
option.
As
you
decide
what
you
want
to
do
to
or
which
services
you
want
to
target,
to
make
it
a
complex
and
that
compact
actually
makes
it
poor
exportable.
B
B
I
mean
if
you
don't
have
to
deploy
something
to
Azure
about
a
different
scenario
that
wouldn't
work
perfectly
well
and
again.
This
is
something
that
we
need
to
make
people
aware
of
that
you
can
run
it
on
the
phone
and
then
on
the
other
side
of
things
like
let's
say,
I
want
to
run
cognitive
services,
but
I
want
to
scale
out
because
I've
got
millions
of
requests
coming
in
now.
This
might
be
something
that
becomes
a
blocker
for
you.
So
what
we
have
actually
started
looking
into
is
running
your
cognitive
services
on
the
container
nice.
B
B
It
means
that
that
once
you
train
your
model,
then
that's
your
static
model.
So
if
you
want
to
add
more
things,
then
you
have
to
redeploy
your
app
or
whatever,
whatever
code.
You
have
running
that
against
that
model
against
the
next
one.
So
it's
not
as
as
easy
to
update
or
as
easy
to
to
grow,
but
once
you
deploy
that
it
becomes
quite
static
like
it's
more
compact
and
that's
the
whole
point,
you
decide
what
you
want
to
run
against.
If
your
API
or
if
your
needs
change,
you
need
to
redeploy
that
I.
B
Well,
I
mean
these
days.
It
becomes
a
lot
easier
to
have
that
kind
of
a
C
ICD
process
that
I
update
something
I,
export
it
and
sticking
into
my
CSD
pipeline
and
eventually
the
OP
Center.
You
can
push
it
to
your
users,
go
cursor,
it's
not
as
hard,
but
it
means
that
there's
a
little
more
a
manual
maintenance,
but
at
the
same
time
he
gained
up
portability.
Okay,
so
I
did
say
about
container
support
for
custom
vision
right.
So
there
are
specific
services
under
cognitive
services
umbrella.
Now
that
we're
starting
to
push
into
containers.
B
A
B
A
A
B
A
B
B
Yeah,
okay
and
then
down
here
you
have
the
local
endpoint
that
you
can
hit
and
in
fact
we
can
put
this
to
test
you.
If
I
come
up
here
and
open
a
new
browser
and
then
put
it
there,
kaboom
nice.
So
that
tells
us
that
the
container
is
running
locally.
But
if
I
wanted
to
test
that
we
have
a
swagger
right.
A
B
B
B
A
B
B
A
B
A
B
B
It
tells
you
where
you
can
find
the
results,
so
we
could
do
a
post
or
we
could
just
can
do
it
get,
and
this
is
one
get
back
like
this-
is
all
that
text
that
we
got
and
can
see,
authorize
to
actually
the
spades,
I
guess:
here's
an
additional
tips.
So
this
is
the
image
that
took
and
then
I
processed
that
and
it
got
every
single
water
sentence
out
of
that
and
there's
also
Wow.
B
Is
that's
exactly
for
that
I
mean
they're
working
right
now.
Do
you
have
handwriting
as
well?
Add
it
to
that,
and
one
of
the
proof
of
concepts
are
working
against
is
take
the
code
that
we
have
for
for
our
application
to
do
all
the
business
logic
puts
it
into
a
function
and
other
functions
based
API.
That
goes
into
all
these
things
and
then
have
a
and
I
am
a
Cuban
in
this
cluster
running
my
containers,
so
I'm
decoupling,
everything
and
then
making
it
more
scalable
just
to
see
how
he
would
perform
makes.
B
A
B
A
B
Was
mainly
about
following
around
our
documentation
right,
we
have
excellent
documentation,
Docs
or
Microsoft
or
comb
that
explain
how
to
use
it
service,
and
you
know,
make
the
right
decisions
as
well
based
on
what
you
need
to
do
and
which
tools
you
need
to
use.
So
for
me
it
was
a
no-brainer
to
use
dotnet
to
try
to
create
a
mobile
app.
It's.
B
Mean
how
can
you
go
wrong
and
summering?
Was
there
no
brainer
for
creating
the
app
to
target
both
iOS
and
Android,
so
no
code,
duplication
and
then
taking
the
existing
skills
to
use
the
Azure
SDK,
which
is
an
ed
standard
library
that
can
run
anywhere
and
then
using
a
service
that
can
analyze
our
data,
anything
something
meaningful
back,
and
that
was
the
cognitive
services
family
of
services
make.
A
A
B
Mean
again,
our
documentation
is
excellent.
We
have
some
really
good
images,
and
might
you
know
Microsoft
has
contributed
quite
a
lot
as
part
of
the.net
core
and
drive
to
make
these
as
easy
for
everyone
as
possible.
Velocity
has
the
tooling
there
to
allow
you
to
add
these
very
easily
and
then
it's
all
reverse
engineering
right.
You
take
what
you
have
there
and
then
try
to
work
out
what
what
it's
doing
at
every
step
and
then
putting
it
all
together,
working
on
individual
bits
and
then
trying
to
put
everything
together,
and
that
was
my
approach.
Okay,.
A
A
A
Spadina
core,
our
docker
containers,
cognitive
services
and
a
bunch
of
other
Azure
services.
Exactly
okay,
thanks!
Thank
you.
Well,
there's
been
another
episode
of
the
on
dotnet
show
and
we
learned
about
you
know,
starting
from
scratch,
with
an
azure
project
and
how
it
can
be
successful.
Thank
you.
Thanks.