Hey my name is Felipe and welcome to my channel.
In this video I'm going to show you how to train a semantic segmentation model using YOLO V8 and
Google Colab, and this is exactly the notebook we are going to use in this process you can see we have
only one two three four five steps so in only five steps you will have completed this process and you
will have trained your own semantic segmentation model using YOLO V8. Now this is very important,
this tutorial is only a very quick and a very high level description of the entire process
if you want more details or if you want a much more comprehensive tutorial then I invite you to
take a look at other of my previous videos where I show you exactly how you need to prepare the
data how you need to annotate the data how to train the model and how to evaluate the model you
trained this other video this other tutorial is a much more comprehensive description of the entire
process so if you want more details I definitely invite you to take a look at this other video
which I'm going to be posting over there, now let's continue now let me show you something else which
is the Google Drive directory we are going to use in this project you can see we have three
files one of them is config dot yaml then data.zip and then train. ipynb this file over here is the
notebook in which we are going to be working in today right train. ipynb is the notebook we are
going to use in order to train this model and then we also have these two other files, now it's very
important you create a directory in your Google Drive in order to work on this project you can
see the directory I have created is located in my drive computer vision engineer semantic
segmentation YOLO V8 Google collab that's the directory I have created in order to work
on this project but please remember to create a directory in your Google Drive which is where
you're going to upload all the data and all the files and so on so please remember to create a
directory in your Google Drive and then I'm going to show you all the files you need to upload into
this directory now let me show you this file over here which is data.zip let's go to my file system
you can see that this is the file I show you in my Google Drive data.zip and now let me show you
how this file looks like you can see we have two folders one of them is called images and the other
one is called labels within the images directory we have three additional directories one of them
is called test the other one is called train and the other one is called validation and within
each one of these directories is where we have all the data we are going to use in order to train
the model now let me show you how this data looks like in my case I am going to train a semantic
segmentation model in order to detect ducks and alpacas so you can see that all the images I have
over here they are about ducks and alpacas right because these are the two categories I am going
to detect with this model now let's go back here to this other directory which is called labels
and this is where we're going to locate all the labels all the annotations we are going to use in
order to train this model now if you want a much more comprehensive description of how you need
to prepare all these files which contain all the annotations all the labels for this process then
I invite you to take a look at this tutorial over here at these instructions which are in Yolov8
official website and these are the yolov8 official instructions on how you need to prepare all the
annotations... in order to train this model because if I show you how these files look
like you can see that they look very weird right we have many many numbers and this is a very long
file so in order to properly understand how to create each one of the files you need to locate
here and within each one of these directories then I invite you to take a very close look to these
instructions which are provided by YOLO V8, now a very quick note in my case I have downloaded
all this data from the Google open images dataset version 7 from this website over here which is
the Google open images dataset version 7 this is an amazing dataset I have used it I don't know how
many times many many many times an amazing dataset but downloading this data was a very time
consuming process so what I decided to do in order to help you to create your dataset to train
your semantic segmentation model with yolo V8 is to create a python script which takes care of
the entire process of downloading this data in a super super super easy way the only thing you need
to do in order to use this script is to edit these values over here which are all the class names
all the categories you want to download from the Google open images dataset version 7 in my case
I have selected these two class names which are alpaca and duck so this is why I have downloaded
all the images and all the annotations for alpacas and ducks but just replacing these values for
whatever other categories whatever other class names you want you will be able to use this script
in order to download a semantic segmentation dataset from the Google open images dataset version 7 this is
an amazing resource I created in order to help you to train this model and this is a resource
which is available in my patreon so it's going to be available to all my patreon supporters now
let's continue now let me show you something which is very important you can see that although we
have three directory over here which are test train and validation I am going to use only these
two which are train and validation I am not going to use the data from this directory which is the
test directory this is a very very quick note and now let's continue so I'm only going to use these
two train and validation once you have created a dataset once you have created a directory with
all the data which is formatted exactly like I show you over here then the only thing you need to
do is to zip this directory and to create this data.zip file and once you have this file the
only thing you need to do is going back to your Google Drive and to upload this file into your
directory into your Google Drive and that's it and now let's get back to the Jupiter notebook
we are going to use in order to train this model because now it's time we start with this process
now I'm going to execute absolutely every single cell in this notebook and this is how we're going
to train our semantic segmentation model so I'm going to start executing this cell over here
which is going to mount my Google Drive into this Google colab instance right I am going
to connect Google drive with Google collab and this is very important because we need to
access all the data and all the files we have in this directory we have created in Google Drive
so the only thing I'm going to do is to select my account and to click allow okay now our Google
Drive is mounted into our Google collab environment and now we're ready to continue with the next
step in this process which is preparing the data or actually this is going to be getting the
data getting this file this data.zip file from our Google drive into this instance and then this
going to take care of unzipping this file and taking all the files which are within right so the
only thing I'm going to do is to press enter and this is going to take care of getting the file
getting the data.zip file and then unzipping the file and unzipping all the content we have within
this file okay so we have extracted all the data we have within this file we have extracted all
the data in this environment in this Google Colab environment now let's continue now the next step
in this process is installing ultralytics, ultralytics is the python library we are going to use
in order to train this model with yolov8 so this is a very important library and the only thing
I'm going to do is to execute this cell, now let's continue and you can see we have already
executed three steps in this process we have only two steps left in order to train this model, this is
amazing, and now the next step will be training the model and in order to train the model now I
am going to show you this file which is config dot yaml which is the other file we have in the in
the directory in our Google Drive so let me show you this file this is a file with very important
configuration in order to train this model and if you have done everything as I showed you in
this tutorial so far then you can just leave these three values as they are currently right now
and everything is going to work just fine this is the location the absolute location the absolute
path to your data and this is the relative path the relative location of your training images and
this is the relative location of your validation images and when I say relative I mean relative
to this location over here right so long story short if you have done everything as I
showed you in this tutorial then you can just leave everything as it is right now and everything
is going to work just fine let me show you these two other keys we have over here which are nc
and names, nc stands for number of classes and in my case I have two classes so in my case this
is two and then this is a list of all the classes we are going to detect with this model remember in
my case I'm going to detect alpacas and ducks so this is exactly how this looks like in my case,
for me, but please remember to update these two values with all the class names and all the number
of classes you are going to detect with your model, right, so this is a very quick description of this
file which is config dot yaml and this is a very important file in order to continue, something that's very
important is that please remember to update this path over here with the path of the directory you
have created in your Google Drive remember in my my case this is my drive computer vision engineer
semantic segmentation YOLO V8 Google collab which is exactly what I have over here but in your case
please remember to update this directory because otherwise this is not going to work right and the
same goes for this other value over here please remember to update this directory otherwise
nothing is going to work this should be the directory in your Google Drive the directory you
have created in your Google Drive now let's continue now the only thing I'm going to do is to execute
this cell and this is going to take care of the entire process of the entire training process so
I'm just going to press enter and then we need to wait for a few minutes or actually for a few
hours until this process is completed okay now the training process is completed and this is all
the output which was generated during the training process you can see this is a lot of information
and from all this information over here we are going to mind this value which is where all
the data was saved all the data we need in order to evaluate this model and this is also where
the weights we generated with this model where the weights were saved and everything is located
here in run segment train so what we are going to do now is we are going to get all the data which
was saved here in this directory and we're going to copy this data into our Google drive this is
very important because this is going to make the process of downloading this data much easier so
what I'm going to do now is executing this cell and this is going to take care of copying all
this directory into my Google Drive okay now going back to Google Drive this is the directory
we have just copied and what I'm going to do now is to download this directory into my computer
into my file system and now let me open this file so I can show you how it looks like and basically
this is where we are going to have all the images and all the data we need in order to evaluate
this model and also this is where we're going to have all the weights of the model we created
now let me show you a couple of these images so I can show you how they look like I'm going to
open this image over here which is val batch 0 pred and you can see that these are a few examples
of how the model performs with a few images so this is an image we can use in order to evaluate
this model now let me open this other image which is val batch 1 pred and you can see that this is
exactly the same this is how the model performs with a few images in this case we have images of
ducks and we also have images of alpacas so this a very good way to evaluate the model performance
these are images from the validation set from the validation directory now let me open another...
another image which is val batch 2 pred and this is exactly the same this is how the model
performs with a few images now in this video... this was remember a very quick and a very high
level description of the entire process I am not going into the details of the model evaluation if
you want more details on how to make sense of all the data we have over here within this file then I
invite you to take a look at this other video this other tutorial which is a much more comprehensive
description of the entire process and in this other video I show you how to evaluate the model
and how to analyze all the information we have over here so I definitely invite you to take a
look at this other video and this is going to be all for today my name is Felipe I'm a computer
vision engineer and see you on my next video.
