frFyp8pZ_mE

[Music]
solo there and welcome to another
tutorial my name is kami Bakshi and this
time we're going to be going over how
you can use the IBM Watson visual
recognition tooling and fluff this is
the new tooling that they have just
released in fact let me tell you a
little bit about the backstory behind
all of this so it actually all began in
round November of 2016
and in November and I remember Josh
Zynga and of course thanks to him he is
a product manager at IBM Walkman and
from he actually reached out to me and
he said hey we know that you've been
using visual recognition for some time
now and we've just released the same
visual recognition tooling well not
released but they started beta testing
his new visual recognition tulane and it
wasn't you test it out you know do some
beta testing be one of the first users
and so of course I was glad to do that
because I knew at the time that Scotty
Watson visual recognition command line
utilities weren't even out and it wasn't
really that easy to get a Watson
classifier walking visual recognition
classifier train because through the
utilities and even now through utilities
it's not how easy and something like
visual recognition tooling so as you
know other Watson services do have truly
let's allow developers to easily train
them or speed their data into a train
Watson and then get output as well into
the visual recognition tooling does
practically the same thing and so of
course while I was using the visual
recognition tooling I got really
interested Josh if I could actually
create a YouTube video about this
and while unfortunately it wasn't the
correct time to do that because it
wasn't officially released now however
it has been officially released and you
can actually use this visual recognition
tooling in order to train your profit
wires in order to create visual
recognition purposes so now though I'm
going to show
you an example of how I was able to use
this tooling in order to create a water
visual recognition bonfire which can
successfully classified among different
types of flowers today we're going to
using four favorite species of flowers
and door going to be painting in quite a
few images of these four species we're
going to expedient images of rodents all
right we're going to see move images of
daffodils relatives and daffodils all
over you feeding in iris plants and of
course we're going to be feeding in the
sunflowers as well and so what we're
going to do is we're gonna get all right
50 to 100 images of all of these
different species of plant into the
watson visual recognition tooling and
basically what we're going to do here is
we are going to have that visual
recognition tooling as a web interface
that we can access and then we're going
to do is i'm going to see these into the
tooling
and then once they're fed the choice
between will use the REST API for visual
recognition and it will then go to the
regional recognition service through
that REST API and that is how this
entire systems going to work and so now
again instead of having teams see URL
commands or instead of having to use the
command line utilities or having to use
any language specific SDK all you need
to do is upload your images to this VR
tooling over here and then once it's
done training you can then continue to
use that top of my RIT with the same API
key among all of your different
applications whether that be powered by
the SDK the REST API indirectly or even
the tooling itself and so now without me
further ado let's get into how you can
actually build this system and so now
without further ado let's get into the
code all right so welcome back to the
Mac part and now I'm going to be showing
you how exactly you can use this brand
new visual recognition tooling service
provided by IBM let's take a look so now
if I go over here as you can see even
before we get into visual recognition
the service and how we can use new
tooling first I'd like to show you a
little bit about the training data that
we're actually feeding into the visual
recognition service now as you can see
in this folder here I've got four
folders in it and these four folders
actually are titled daffodils iris rose
and sunflower and in each folder
they it contains around 100 images of
that type of flowers you can see over
here for daffodils we can go over to
iris this contains iris plant images and
again these are just IRS plant free
images gathered from Google and so if we
go back to roses as you can see all
different types of Rose is there are
some you know red roses purple roses
pink roses white roses etc etc and then
if we go back over here as you can see
in these sunflowers we've got all this
in touches on
flowers even ones with sunglasses on as
you'll see later in the data set but if
we go down as you can see we've got all
these different types of sunflowers and
again these are not strictly only
sunflowers we may contain a little bit
of noise in this data say for example
this image right here because again
you've just images scraped off of Google
images for free and so what's happening
is we're basically just trying to tell
visual resolution what the basic sketch
of a sunflower looks like by giving it
most of the data that doesn't contain
noise and then some noise in between
that it's able to filter out and so as
you can see eventually these are all of
our sunflowers all right so those are
our four these are our four directories
containing those images and outside we
then have in this root directory four
more images not just not not folders we
just have image files and the images are
named daffodil iris rose and sunflower I
go through these again as you can see
these are individual images of one iris
one one daffodil one rose and one or one
sunflower field and so one more thing
I'd like to point out here though is
that these are going to be our test
images and these images are not
contained in these folders and what if I
hit done after this is actually
compressed each folder into a zip file
right after that and as you can see
these are all of our zip files and that
is what we were going to be using to
train the visual recognition service all
right so now we can go over to the
bluemix here as you can see which I went
to a bluemix catalog and I was able to
create a new visual recognition service
and once you have created your service
and once you have those credentials you
are then ready to actually find your API
key now one more thing I'd like you to
note is that eventually the tooling will
be available on this managed page in
your service near this developer
resources pane or you know somewhere on
this page here the only reason it's not
currently available is since I'm
actually recording this just before it's
officially released and that means that
it isn't technically available in
bluemix
just yet but what I'm going to show you
how the almost exact same UI exact same
sort of interface for using the tooling
it's just that I cannot access it
through Linux instead I have to access
it through URL but what I'm going to do
first is I'm actually going to find my
API key and then I'm going to copy it
and go over to the Watson visual
recognition tooling you will not need to
go to the Watson visual recognition
tooling my gimmick on that page although
I am because again this will be
available through bluemix I'm going to
click on API key enter in my API key
click enter and as you can see it is
going to bring me into the Watson visual
recognition tooling then inside of this
tooling I'm ready to create a new
classifier and what I'm going to do is
I'm actually going to call this
classifier flower YouTube classifier now
what we're going to do is we're going to
give this four different classes and
these classes are going to be coal Rose
iris I'm going to add a new class call
it daffodil and I'm going to add a new
class and call this one
sunflower now of course we don't need to
upload any negative images as these are
optional and we only want our classifier
to distinguish between different types
of flowers in this case Rose IRS
daffodil and sunflower and not tell me
if something is a flower and if it is
what what species flower what type of
flower but we do not want want it to
tell us whether it's not a flower that's
that's a feature that we do not want
with this classifier so we are not going
to be uploading a negative data set so
I'm going to go back here and I'm going
to take the Rose dot bit file and I'm
going to drag it into my Rose class in
the tooling once that is done I'm going
to go back and do the exact same thing
for the iris once that thing done I'm
going to do that for the daffodil and
once that's done I'm going to do it for
sunflower one more thing I'd like to
note here is that these you can only
upload up to 50 megabytes of images and
you zip files to watts and visual
recognition that's the most that it can
take at once so one thing I'd like you
to note here is that I have not actually
I think they optimized my JPEG
JPG files that's why these are actually
relatively large files 41 megabytes in
total for around 100 images these of
course can be shrunk even further using
some sort of JPEG öktem optimizer which
you can get by a homebrew or you can
download by the Mac App Store if you're
using Linux or Windows there are
alternatives for that as well there are
also online tools so gun if you have a
lot of different training data and you
want to see it all in to visual
recognition a few things you can try you
know reducing your file sizes key
weather that we through optimization
making your image is smaller or
something of that sort again at the end
of the video I will be getting my
contact information so if you'd like to
contact me and I can help you out with
dial your visual recognition training
all right so now in March do know is I'm
going to click on create and now the
visual recognition tooling will take all
these files and upload it to the Watson
visual recognition API now once it goes
into the API we will be ready to start
using the classifier after it trains and
so now though this is currently
uploading our images to the API now this
may take several minutes to complete due
to the fact that the image files are not
small so I will speed up to click now
and then we will be right back with the
training classifier
all right so as you can see it is done
uploading our images to the API and it
is now training our classifier now all
we need to do is wait for the visual
declination service we've done training
our classifier and once it is complete
we'll be able to use our classifier and
test other images against the model to
see how it performs all right I will be
back in just a few minutes right as my
classifier is done training and then
we'll take a look at the performance of
the model all right so as you can see
I'm back again and visual recognition
tool is done training the classifier and
so now what we can do is we can actually
go ahead and give it our test images and
we should be able to see the results
that has for each one let's actually
begin with the daffodil image so what
I'm going to do is I'm going to take the
daffodil image and drag it over to the
tooling and drag it into this box under
our classifier it's going to upload and
as you can see Watson visual recognition
responds with a confidence value in as
you can see it says that this indicates
not necessarily confidence but this is
this looks eighty-three percent like a
sunflower to Watson because what's
actually happening is this technically
isn't a confidence or an accuracy value
what instead is happening is Watson is
actually taking it's actually creating
four different models internally and
what it's doing is it's taking let's say
the first model and it's saying okay
let's just say we have daffodil is one
class and everything else is one class
and it will compare daffodils to
everything else and then we have three
other models that do the exact same
thing it'll compare relevance to
everything else is just everything else
and sunflowers to everything else
Erol then use that as basically an
ensemble of different models and it'll
use those in order to find which one
actually has the most probability of
being credit or it's this impressive
collection of probabilities from which
it then selects and in this case
daffodil turns out to be the correct
answer but if you see Rose actually is
not far behind and the reason for this
is because if you think I'll let's think
a little bit about this and if you think
about a rose and if you were to look at
this image it would see a little odd
resemblance to Rose and I mean if you
were again
look at this I have a little bit of that
rose pattern the other thing it has more
of a daffodil pattern and that's why
what we be able to tell us that this is
indeed a daffodil or at least that's
what I believe and so that's why a
watching is able to classify this as a
daffodil then we can go ahead and take
the IRS in inch huh of course hasn't
also open that up but okay sure so we
can take the iris image drag that into
the box it's going to upload the image
to Watson as you can see again these
four models have returned this
probability that iris is most probably
the correct answer because it's able to
find the iris pattern the iris color and
what's usually in the background of
these IRS images it's able to take
everything into account using its you
know internal algorithms and it's been
able to actually return to us that iris
is most probably what is inside of this
image we can then go ahead and take the
rose drag that in and as you can see in
just a moment it tells us that Rose is
the correct answer again these flowers
are not in the training set desert this
is a separate test set that we have not
trained lots of visual recognition to
understand yet it is able to tell us
that this most probably contains a rose
even despite the black background
despite the little water droplets on the
Rose despite you know all these all this
noise around the roads we're still able
to determine that this contained the
Rose because of the Rose patterns that
can then are contained inside of this
image and then of course now here comes
a slightly a little bit more difficult
one something that something that
something that really shows you the
power of the service a perfect daffodil
when the sunflower one's a little bit
challenging as well and the reason I'm
going to say this is because of color
and it's not necessarily because hey
it's going to be unable to find you know
sunflower patterns it's going to be able
to do that perfectly it'll be able to
find okay we've got so many sunflowers
here this is probably a sunflower field
and return some flowers the correct
answer but if you see there's a lot of
conflicting noise here
with the red sky and blue sky here in
the clouds and the other other plants
around here all the other vegetation
here and you can see basically this has
a lot of noise but yet technically if we
want to take this image and drag it into
the service we should be able to get an
answer and it should be sunflower so
again it's uploading the file as you can
see again Watson does tell us that
sunflower is most probably the correct
answer what's currently in this image or
what this image contains out of the four
classes that it was that it was trained
to recognize inside of images and again
note because the technique that it uses
here Watson visual recognition is not
able to say hey this contains a
sunflower and a rose or a sunflower and
iris if not able to say hey contain both
of these or three of these or nothing
what happens is lots of visual
recognition instead it can only find
individual objects inside of those
images and so for like for example if
you wanted to create a classifier that
could detect some flowers and roses
together you have to create an entirely
new class for sunflower Rose because
then you would give images that contain
both of sunflower and a rose and then it
will be able to classify it as that
class just a few notes about the facial
recognition service but that's what
makes it so interesting and so great and
fun and easy to work with again this is
the brand new tooling and as I like to
say the visual recognition service is
actually a great service to use as your
first walking service because of its
simplicity because of how great you can
just give it a little bit of training
data and be able to take all that learn
it and take new test data and give you
your answers and with good accuracy as
well and so now this makes it even more
user friendly and even better how to use
with your first lots of service and
generally to use as a professional tool
for your applications and now the visual
recognition service has just gotten so
much better and that's what I have to
cover in this tutorial thank you very
much for watching that's going
of course so if you enjoyed this
tutorial and if it was able to help you
out please do make sure to leave a like
down below and if you believe this could
help anybody else you know like your
friends or family make please make sure
to share video as well alright that
point for this tutorial but if you have
any comments feedback suggestions
suggestions or questions you can use
them down in the comments below
email them to me at 10 G man.you
gmail.com or tweet them to me at
humanity
alright so that's going to be for this
video but if you'd like to see a lot
more content like this and of course if
you want to try your support for the
channel please to make sure to subscribe
as well as it really does help out a lot
and if you'd like to be notified via
email and Google notifications whenever
I release a new video please do make
sure to turn on notifications as well by
clicking the little bell icon beside the
subscribe button
alright so thank you very much for
watching today that's going to be all
for this tutorial
goodbye