►
From YouTube: GitHub Satellite India 2021 - Lightning talk: AI for web: An introduction to TensorFlow.js
Description
Presented by Shivay Lamba, Production Engineer @ Layer5
Conventional methods of uploading machine learning models for web applications can be a daunting task for web developers who specialize in Javascript. This is where Tensorflow.JS (TFJS) comes into the picture.
In this talk, you'll get an introduction to TensorFlow.js—and learn how this powerful tool helps create dynamic web apps using machine learning.
As always, feel free to leave us a comment below and don't forget to subscribe: http://bit.ly/subgithub
Thanks!
Connect with us.
Facebook: http://fb.com/github
Twitter: http://twitter.com/github
LinkedIn: http://linkedin.com/company/github
About GitHub
GitHub is the best place to share code with friends, co-workers, classmates, and complete strangers. Millions of people use GitHub to build amazing things together. For more info, go to http://github.com
A
Everyone
who
are
watching
this
lighting
talk
I'm
so
I
am
currently
a
tensorflow
js,
sig
member
and
working
group
member
as
well,
and
I
was
also
previously
a
google
tensorflow
coding
mentor
and
I'm
currently
also
a
production
engineer
at
layer
5..
So
the
talk
of
my
lightning
presentation
is
machine
learning
for
the
web
and
introduction
to
tensorflow.js.
A
A
Javascript
has
support
for
all
of
these
different
spectrum
of
devices,
and
you
can
introduce
machine
learning
with
javascript.
You
know
one
of
the
biggest
reasons
why
we
want
to
do
that
is
for
creative
professionals,
who
are
only
learning
javascript.
It
can
be
very
difficult
for
them
to
probably
learn
another
type
of
language
like
python
and
introduce
machine
learning
models
in
a
cross-platform
application,
or
that
uses
both
python
and
also
uses
javascript
so
because
of
the
versatility
of
javascript.
A
A
So
that
includes
things
like
augmented
reality
or,
let's
say,
gesture,
based
interactions
or
sound
recognition,
and
all
of
these
different
types
of
applications
that
we
build
in
the
web
platforms
can
be
coupled
with
tensorflow.js
to
create
machine
learning
experiences
that
are
going
to
blow
your
mind
with
some
of
the
examples
that
we'll
be
showing
you
so
the
first
category
of
the
such
kind
of
models
are
the
pre-trained
models.
Now
these
are
completely
easy
to
use.
A
You
can
simply
go
to
tensorflow.org
js,
slash
models
and
you'll
find
the
github
links
for
all
of
these
different
models.
At
the
same
time
now,
a
few
examples
of
these
include,
for
example,
object
recognition
that
is
one
of
the
most
fundamental
deep
learning
or
machine
learning
based
projects
that
can
be
taken
up
and
focused,
which
is
a
really
popular
object,
recognition
model
also
available
in
tensorflow.js,
and
it
can
be
trained
up
to
90
different
classes.
A
So,
as
you
can
see
in
this
picture
that
we
have
a
picture
of
two
dogs
and
it
can
actually
create
bounding
boxes
on
these
dogs
images
of
these
dogs
and
give
a
really
great
confidence
at
the
same
time
to
the
pictures.
So
it's
really
great
in
doing
it
quick
and
also
is
highly
accurate.
At
the
same
time
now
there's
another
example
of
a
face.
A
Mesh
now
face
mesh
is
a
model
that
is
just
three
megabytes
in
size
and
it
actually
recognizes
over
468
different
landmarks
on
your
face
and
again,
as
you
can
see
that
it's
really
popular
and
being
used
right
now
in
production
in
in
places
like
l'oreal,
which
you
can
see
on
the
right-hand
side
being
used
in
production
environments
and
being
actually
used
in
products
that
are
ready
to
go
and
ready
to
ship
as
and
when
required.
Then
there's
another
model
that
we
have.
A
That,
basically,
is
the
body
segmentation
that
can
basically
distinguish
24
different
body
areas
in
human
bodies,
and
it
can
also
do
that
in
real
time
and
also
across
multiple
different
bodies.
So
there's
this
need.
It
doesn't
need
to
be
just
one
single
body.
As
you
can
see
in
this
video,
there
are
three
different
individuals
whose
bodies
are
being
segmented
as
compared
to
the
background
in
real
time
and
again
without
any
performance
overheads.
A
At
all.
Now,
with
a
bit
of
creativity,
we
can
emulate
a
lot
of
different
superpowers
using
tensorflow.js
at
the
same
time
and
just
to
sort
of
show
you
examples
of
cross
development
applications
over
here.
This
a
person
who's
from
a
community,
a
tensorflow
js
community
in
the
united
states-
has
created
this
amazing
laser
application,
using
webgl,
shaders
and
with
tensorflow.js,
and
these
are
just
some
of
the
examples
that
you
know
we
can
think
of
to
create
some
kind
of
unique
creative
applications.
A
Now
there
is
another
one
that
was
created
by
jason
was
a
real-time
clothing
size
estimation
that
uses
the
body
segmentation
model
and
can
provide
you
the
details
of
your
body
size
by
just
scanning
your
entire
body
and
within
the
next
10-15
minutes.
It
can
provide
you
a
very
quick
body,
size
estimation
for
your,
so
that
you,
this
could
be
used,
probably
in
sites
like
amazon
or
other
kind
of
real
world
applications
or
when
it
comes
to
any
kind
of
estimation
that
needs
so
not
just
related
to
the
closing
size.
A
At
the
same
time
now,
the
idea
is
that
we
can
combine
different
types
of
tech
with
the
tensorflow.js
and
again.
This
shows
a
really
great
example
where
basically,
we
are
using
web
xr.
That
is
the
mixed
reality,
and
we
are
also
using
webgl
to
create
this
3d,
rendering
of
an
augmented
reality
application
where,
for
example,
which
sort
of
scans
an
image
and
creates
a
real-world
model
of
an
augmented
reality
model
of
this
statue
right.
A
So
the
idea
is
that
we
can
combine
different
types
of
technologies
that
are
supported
and
web
very
seamlessly
and
again,
the
integration
is
usually
very
simple.
At
the
same
time
now,
the
second
part
is
using
transfer
learning
so
basically
retraining,
some
of
the
existing
models
that
we
have
to
work
with
your
own
data,
to
sort
of
customize
those
models
to
your
own
data,
to
provide
you
better
accuracy
and
better
functionality.
A
At
the
same
time,
so
transfer
learning
is
essentially
taking
an
existing
model
and
use
it
to
basically
apply
to
similar
kind
of
problems
or
in
some
kind
of
domain
such
as
probably
like,
you
know
like
say,
recognizing
a
cat
instead
of
a
dog
right.
So
again,
there
are
two
different
ways
in
which
we
can
use
the
transfer
learning.
One
of
the
most
easiest
ones
is
teachable
machines.
A
A
And
if
your
data
is
in
gigabytes
or
if
you
want
to
create
a
production
ready
model,
we
can
also
use
the
cloud
automl
that
basically
lets
you
train
custom
engine
models,
and
if
you
have
data
in
more
than
gigabytes
in
size,
it
can
simply
just
use
this
and
we
can
deploy
that
directly
to
tensorflow.js
once
we
deploy
it.
It
creates
basically
the
small.json
file
that
essentially
has
the
entire
model
of
how
it
works,
and
we
can
use
that
simply
with
our
application.
A
At
the
same
time-
and
this
is
a
very
small
example
of
a
code
so
in
the
first
two
lines
you
see
is
what
we
have
imported
the
tensorflow.js
library
and
also
the
auto
ml
library.
In
this
third
line,
you
see,
we
have
simply
just
taken
an
image
of
a
daisy,
that
is
a
flower
and
we
have
just
given
it
the
id
of
daisy
for
the
image
and
then
the
script
that
you
see
is
not
more
than
five
lines
in
length
and
what
we
do
in
this.
A
What
we
do
in
this
is
that
we
have
created
an
async
function,
because
the
machine
learning
models
take
some
time
to
load,
so
we
would
want
to
make
the
predictions
only
after
the
model
has
been
loaded.
So
what
we
do
is
that
we
use
an
async
function
or
an
asynchronous
function
in
javascript
to
first
load
our
image
classification
model
from
our
model.json
file.
That
has
the
characteristics
or
basically
the
dependencies
of
our
machine
learning
model.
And
then
we
just
use
the
const
image
to
get
the
document
by
id
for
our
flower.
A
So
we
can
get
that
particular
image
id
and
then
what
we
do
is
that
we
simply
run
the
predictions
using
this
and
we
use
the
classify
to
basically
classify
okay.
What
type
of
flower
is
it
so
it
will
basically
tell
you
what
is
the
classification
or
the
label
of
the
image
right
and
the
probably
the
last
one
and
the
most
powerful
one?
Is
writing
your
own
code
intensive
projects,
so
tensorflow
basically
provides
you
two
different
types
of
apis.
A
One
is
the
layers
api
that
is
very
similar
to
how
keras
works
on
top
of
tensorflow
and
that
allows
you
to
use
some
high
level
functions,
but
we
also
have
the
ops
api,
or
that
is
the
lower
level
api.
Where
you
can
also,
you
know,
change
the
mathematical
functions
like
the
linear
algebra
functions
in
your
code.
That
provides
you
a
lot
more
flexibility
and
functionality
when
it
comes
to
changing
or
writing
your
own
code.
Now
this
sort
of
gives
you
the
architecture
of
what
we
have
discussed
so
far.
A
So
the
top
level
is
the
pre-built
models.
The
next
one
is
the
layers
api,
that's
similar
to
how
keras
has
been
built,
and
finally,
we
have
at
the
most
low
level,
the
ops
api,
and
this
basically
allows
it
to
run
either
on
the
client
side
or
on
the
server
side
or
the
client
side
with.
Basically
is
the
interaction
with
the
browser.
A
Basically,
this
is
a
mobile
net
model
that
we
have
and,
as
you
can
see,
the
difference
between
the
performance
of
python
based
model
and
like
a
javascript
based
model
is
very
minuscule,
and
the
difference
is
not
that
much
right
and
again.
This
shows
another
example
of
an
of
like
an
np
based
model,
that
is
the
distilbert
model
and,
as
you
can
see
that
this
is
actually
better
as
compared
to
the
one
that
we
see
in
python
now,
there
are
also
five
different.
A
A
We
get
a
lower
cost
and
again
interactivity
at
the
same
time,
and
also
we,
if
you're
using
it
on
the
server
side,
then
you
basically
get
the
power
of
javascript
just
in
time
compiler
or
we
can
also
just
write
in
just
one
particular
language
and
of
course
we
can
run
models
that
are
much
more
powerful
as
compared
to
the
client
side
because
of
the
dedicated
cpu
and
gpu
power
that
comes
with
node.js.
So
these
are
some
of
the
server-side
benefits
at
the
same
time
now
just
to
share
some
of
the
resources.
A
Basically,
you
can
visit
all
of
these
different
links,
the
website,
tensorflow.org
js,
and
the
models
on
testfloat.org
models.
And
again
you
can
visit
all
these
different
links
to
get
started
with
tensorflow.js
at
the
same
time
and
you
can
join
the
community
by
using
a
hashtag
made
with
this,
and
you
can
follow
this
on
twitter
and
on
links
like
linkedin
and
again.
Thank
you
so
much
for
staying
in
touch,
and
you
can
follow
me
on
these
links
to
get
started
again.