FYZld6SSCnY

[Music]
hi t uh thank you so much for your
videos about services in Blu miix I'm
interested in a visual recognition
service in Blu miix and uh it would be
great if you could explain to me how I
can use this service I'm Bernie from
Germany thank you hello there and
welcome to another tutorial my name is
Hy Baki and this time we're going to be
going over how you can use the IBM wats
and visual recognition service in order
to classify images now to begin uh as
you can see Benny has sent me a question
she is one of my subscribers and is from
Germany uh and so she sent me a question
about the IBM Watson visual recognition
Service uh and so uh as you can tell uh
I'm going to be creating a video about
the topic now I've actually been getting
quite a few requests about this service
uh and to actually create a video about
it and so I decid
why not uh all right so now another
thing is uh its possibilities in the iot
field as well since it was actually
merged with Alchemy Vision since IBM
bought that a few years ago uh its
possibilities in the iot field are
endless because of course it's much more
accurate now all right so let's begin so
now first of all as you can see I have
my uh Helper and model Bruno here uh who
is going to be sort of modeling for the
uh visual recognition classifier now if
you're wondering how a dog relates to
this uh video well essentially what's
going to happen is I'm going to train
the visual recognition classifier in
order to classify jog breeds and so it's
going to be able to differentiate
between a golden retriever uh a
Dalmatian and a husky uh now one more
thing this is extremely accurate so if
you were to put in uh another uh sort of
picture of an animal uh that's colored
very closely to a golden retriever uh it
would still be able to say you know what
this isn't a golden retriever uh because
what's going to happen is we're going to
feed in positive examples of a golden
retriever positive examples of a dalmi
and positive examples of a Husky but
also negative examples of animals that
are not uh dogs like for example cats
and lions uh and other animals like that
and humans uh and so it should be able
to differentiate between them all right
now we're going to be getting into the
Mac part where I'm going to be showing
you how to use crl commands in order to
actually train your own classifier using
the default data given by IBM Watson and
then we'll see if uh it can classify
Bruno here all right let's get to the
Mac part again thank you very much Bruno
all right let's get to it so welcome
back to the mag part and now I'm going
to be showing you how you can use the
visual recognition service in order to
classify dog breeds so let's begin now
first of all as you can see in blue miix
over here uh I have I'm about to create
a visual recognition Service uh and so
I'm going to just choose the space
application just leave it Unbound if
you'd like uh the service name the
credential name and of course the
selected
plan uh which in this case is going to
be free for me if I click click on
create
now
then um in just a
second should create the
uh the uh classifier Ser I mean the
recognition
Service uh and as you can see here it is
done so now as you can see what we need
to do is go to the service
credentials uh and as you can see the
API key is over here now this is
actually a bit different than other and
services in that it doesn't give you a
username and a password instead it gives
it just one API key uh which is I I
usually like that it's more convenient
to have just one key rather than two
username and passwords so much better
all right so now continuing so now what
we're going to do uh is first of all I'm
going to show you how you can actually
create uh a corpus of knowledge that
we're going to give to Watson to train
it uh so let's begin first of all uh as
you can see what I have is text files of
URLs of lots of images of
dalmations golden
retrievers huskys and animals that are
not dogs like cats lions and
cheetahs so now remember these are
hundreds of URLs all right uh and so
basically uh what I'm doing is I've
created this python script and what this
uh really simple script does I I whip
this up in about five minutes not even
five and so basically uh what this does
is it will uh basically take uh a text
file
it will take the name of a text file in
its current directory it'll take a
directory name and what it'll do is
it'll read that text file uh for lots of
or sorry uh it'll read that text file
for lots of links uh to images uh and so
basically what can happen uh is I
essentially take these images I download
them uh and put them into a respective
directory that I told it to uh and then
later what I can do manually is just
compress these into zip files and then
send these to Watson to
train all right so now let's begin first
of all uh in order to actually uh sort
of download uh these text F first of all
though I won't be telling you how to get
the links into these text files though
that you're going to have to find out
yourself uh but uh this python file will
be available on GitHub for you to
download so that you can actually make
use of it uh so now what you need to do
in order to use this uh is it's a very
simple script all it is just python
download images give it the URL file for
example URLs Dalmations and tell it
where you want it to save uh in this
case dations Okay uh just enter and then
it'll download all the images but I have
them pre-downloaded
already all right so now that's how
you're going to download all of your
images uh once you have downloaded all
of your images as you can see uh
hundreds of images here 280 to be exact
uh once those are all downloaded you can
of course uh send them to Watson now in
order to do that uh you need to compress
these into zip files uh and ensure that
these zip files are no larger than 50
megabytes if they are well back to the
drawing board for you uh you have to uh
somehow reduce the size of these zip
archives continuing though so now we're
going to be using a set of curl commands
uh because uh they haven't come out with
a toolkit for the visual recognition
Service uh to train with yet uh but we
will be using uh curl commands in order
to train our classifier all right so now
as you can see I have this little curl
command set up and basically what this
means is we're going to be sending a
post request uh to the Watson platform
Gateway uh and basically I'm going to
enter my API key uh as you could see
here though I'm just going to copy and
paste in my API key to this
URL and so uh quite generic actually uh
we're just creating a classifier with a
name TB dog classifier tany back she's
dog classifier is what it stands for
we're giving it some Dalmation positive
examples over here uh um in fact if I
can just yeah so as you can see we're
giving it some uh Dalmation positive
examples here uh and we're telling it
dalmati dzip uh to find them husky
positive examples husky. zip uh and
golden retriever positive examples
golden retriever do zip uh and of course
negative examples things that are not
dogs are not dogs
.zip and so as you can see over here so
now one more thing is in the actual curl
command I said Dalmation not Dalmations
and golden retriever not golden
retrievers and um
husky all right so now as you can see uh
I have this really simple curl command
ready that will start the training of
our classifier uh and I have these four
zip files ready as well so I'm going to
copy this curl command all right go to
terminal and as you can see if I list
directory here uh we have the zip files
here so now uh if I were to run this
curl command in
theory uh it says as you can see could
not open Dalmation doz let's see what
the problem is here
um oh yes I miss uh misspelled Dalmation
do zip sorry uh so it's not o n it's
just a n okay so now as you can see we
run this once more it'll be able to post
to
Watson
and in just a second now again it is
uploading uh quite a few images to
Watson so it may take a second and as
you can see it has submitted back to us
our classifier ID the classifier name
the owner which is me uh the status
which is currently uh it's currently in
training and the three classes that
it'll classify between all right now one
more thing this is actually a curl
command which you can use to check the
training status uh just like you can of
like uh let's say retrieving rank uh and
so if I go back to the result of this um
uh curl command over here as you can see
I'm going to copy the classifier
ID copy
it uh wherever it says uh classifier ID
in this document I'm going to paste that
in uh and where it says API key I'm
going to paste in my API key all
right then all I need to do is copy this
curl command move it uh to my terminal
paste it in and as you can see it is
currently training now we've sent it
around I'd say 600 images which is a lot
of images and each of them uh are not
too huge I mean in total it was around 5
megabytes of data that we're sending or
6 megabytes of data uh so it shouldn't
be too long until it trains this might
take up to an hour or two uh then again
if you're sending many more images I
currently did like around 200 uh per
class uh and they were just under 2
megabytes
uh and so I mean you could fit thousands
of images uh in 50 megabytes so yeah so
now what I'm going to do is I'm going to
pause the video uh and right as it is
done training whether it be in a few
hours or maybe even tomorrow I'm going
to continue uh recording uh and then
we're going to move uh sort of Stitch
these two videos together uh and then we
will uh see how our classifier turned
out uh and as you know my good old
friend Bruno uh that I was showing you
in the beginning of the video uh he is
going to be one of our test subjects uh
in our video today all right so uh I'm
going to be back in a few hours uh once
it is done
training so the classifier is finally
done training uh it's been around 15
minutes uh not as long as I expected
actually uh and actually took much less
than I expected uh and so yeah that's
great uh so it took 15 minutes I was
expecting 1 to two hours maybe even a
day uh to classify around 600 images but
it was able to do it within 15 minutes
I'm very excited about that so as you
can see this is actually a picture of
Bruno that we are going to be
classifying over here uh and so he is a
Golden Retriever as you should probably
know uh and so yeah we're going to see
if Watson can label him as a golden
retriever so first of all as you can see
uh I've actually set up a tiny URL link
uh that that is a direct link to Bruno's
picture this Bruno picture uh and so
basically uh Yes actually Watson's
visual recognition takes a URL as a
parameter uh for an image uh and so
basically it can follow those redirects
so it will find the final link uh and
then download that and send it um and so
yeah it's able to resolve those links
now as you can see in Adam over here I
mean Sublime
Text uh now basically I'm running the
classify um sort of endpoint here uh but
I'm giving it the API key if you know
the normal API key and I'm setting a URL
to the tiny URL that is a direct link to
Bruno's picture then I'm setting the
classifier IDS to just my uh class
classifier ID that I set up the owner is
me not IBM and the threshold is
zero and I'm setting the version of the
classifier to use to 20160520 which is
the current latest
version all right so now I just put
these enters in so it was a bit easier
to understand so if I remove all of
these real quick as you can see this is
a valid curl command which I should
technically be able to paste into my
terminal and as you can see what's going
to happen is if I actually copy this
tiny URL
link um and if I paste it
in it will download a
file um which is Bruno's
picture again sometimes the Internet
isn't always cooperating with you oh as
you can see it's done uh and is over
here this is a picture we're going to be
classifying of him all right so now if I
click enter for the curl
request in just a second you should
see as you can see it says delation with
7% confidence Golden Retriever with
89.9% confidence and husky with 4.7%
confidence so that means golden
retriever is the highest and we get to
know that Bruno is a golden retriever
according to Watson and as you can see
that was a tutorial on how to use a
visual recognition service in bluex
again thank you very much I hope you
enjoyed if you did please leave a like
down below share this video with people
who you might think can help and
subscribe to the channel if you'd like
to see more of my content or you just in
general enjoy enjoy it really does help
out a lot all right if you have any
questions suggestions feedback or ideas
uh you can leave them down in the
comments below email them to me at tajim
Manny gmail.com or tweet them to me at
tajim Manny and that's going to be it
for this tutorial thank you very much
please subscribe to the channel if it
really helps you out and that's going to
be it for this tutoral
thank you
goodbye
okay hi T me uh thank you so much for
your videos about services in Blu miix
I'm interested in a visual recognition
service in Blum miix and uh it would be
great if you could explain to me how I
can use this service I'm Bernie from
Germany thank you