hey my name is Felipe and welcome to my channel, in
this video I'm going to show you how to train a a pose detector using yolov8 and I'm going to show
you how to do the entire process on Google colab, so let me show you the first thing you should do
in order to work on this tutorial is to create a directory in your Google drive, this is a very
important step because this is the directory we are going to use in order to upload the data we
are going to use in order to train this model and it's also the directory where we are going to
upload all the files we need in this training process, so this is the first step in this process
please make sure you create a directory in your Google Drive, now let me show you the data we are
going to use in order to train this pose detector, let me show you the data I am going to use, but
obviously you can use whatever other dataset you want, if you want to follow along this tutorial
you can obviously use exactly the same dataset I am going to use but I'm going to show you a very
generic process which you can apply to absolutely any other dataset you want right, as long as the
dataset complies with a given structure with a given format you can definitely use it in order
to train this pose detector, now let me show you a few samples so I show you how the data looks
like, you can see that this is a license plate keypoint detection dataset and this is how the
data looks like this is only an example of how the data looks like, you can see we are going to
detect license plates and we are going to detect these four keypoints we have over here, 1 2 3 4,
which are the corners of the license plates right so this is exactly the pose detector, the keypoint
detector we are going to train today we are going to detect these four keypoints of the license
plates right for example this is another example we have two license plates and in each one of these
license plates we have these four keypoints over here right let me show you another example you
can can see this exactly same situation we are going to take these four keypoints over here
and so on so this is the data I am going to use in this process but remember in this tutorial I'm
going to show you a very generic way to train a pose detector which you can apply to whatever other
dataset you want, now let me show you something which is I have already downloaded this dataset
into my computer so let me show you how the data looks like, you can see we have three different
directories which are called test train and valid and if I open one of these directories you can see
I have two folders one of them is called images and the other one is called labels, images is the
directory which contains all the images right let me show you a few of them you can see that these
are images with cars and license plates so this is how the images look like but let me show you
something which is very important and it's the labels the annotations because remember we are
going to train a pose detector we are going to detect four keypoints and we definitely need to
provide the information of where the keypoints are located for each one of these images and this
is how the labels look like right you can see I just opened one of these files and this is how we
are providing the information to yolov8 and you can see that this looks very weird this looks very
crazy this looks like it doesn't have any sense whatsoever it doesn't make any sense whatsoever so
this is exactly the format we need for our labels right this is exactly how we need to specify the
labels the annotations and something that's very important is that this video, this tutorial, is
a very quick and a very high level description of the entire process of how to train this pose
detector but I am not going into the details of everything that's involved in this process if
you want more details of absolutely everything that's involved if you want a much more detailed
and a much more comprehensive explanation of the entire process and if you want to make sense
of these labels you see over here of all these numbers you see over here then I invite you to
take a look at other of my previous videos where I show you exactly the same process how to train
a post detector using YOLO V8 but in a much more detailed and a much more comprehensive way, in that
other video oh my God I show you so many details and for example I show you how to make sense of
this data over here I show you how to create your own annotations in case you are training this
pose detector for your own custom data I show you how to annotate the data how to create all these
labels and so on so if you want more details of this process and if you want to make sense of
these numbers of these very crazy numbers you can see over here then I invite you to take a
look at this previous video which I'm going to be posting over there, and now let's continue with
this process and the only thing we need to do now is to take this folder which is called data
and we need to compress this folder and we need to create this file which is data zip and
once we have this file the only thing we need to do is to get back to our Google Drive and we
need to upload this file into our Google Drive, in my case you can see that I have already uploaded
this file in my Google Drive so I'm not going to do it again but please make sure to upload the
data into your Google Drive in order to continue and then another file which is very important
is this config dot yaml file let me open this file to show you how it looks like and this is a file
which contains very important configuration for this training process this is where we are
specifying what's the location of the data we are going to use in order to train this model
also all the keypoints how many keypoints we are going to detect how many classes we are going
to detect this is a very important file with some very important configuration we need in order to
train this model and remember in this tutorial I am not going to show you all the details which
are involved in this configuration file if you want to know more more details if you want a much
more comprehensive explanation of all the details which are involved in this file then I you to
take a look at the previous video I mentioned a couple of minutes ago but for now let's continue
this is a very important file we're going to use in the next step in this process which will be
to execute all the cells in this notebook right remember this is the notebook we're going to use
in order to train this model and the only thing I'm going to do now, the only thing I'm going
to do is to execute all the cells in this notebook one at a time and that's going to be pretty much
all in order to train this pose detector so let me show you, the only thing you need to change
in order to make it work on your side on your computer on your Google colab is to change
this value over here which is the location of your data in your Google Drive you can see
in my case it says my drive train yolov8 pose detection Google colab data.zip which is the location
of my data in my Google Drive you can see that this is how how this directory is called train yolov8
pose detection google colab and then the data is located here and it's called data.zip so just make
sure you change this value to whatever location you have chosen for this file and that's going to
be pretty much all and obviously you also need to change this other location which is the location
of your configuration file in your Google Drive so just make sure you change this value as well
and everything else will be just the same so let me explain super quickly all the cells in this
notebook in this cell over here we are going to mount google drive into Google colab this is a very
important step because we want to access all the files in our Google Drive we want to access these
two files in our Google Drive then the second step in this process will be getting this file
getting the data.zip file from our Google drive into this Google colab environment and then we are
going to extract all the content in this Google colab, then then the next step in this process
will be installing ultralytics this is a very important step because this is the python package
we need in order to train this model so it's very important we install ultralytics and then the
next step in this process will be training the model so you can see that these are only a few
instructions we are creating a new variable and then we're calling model dot train and we are just
providing the location of the configuration file and we're also providing for how many epochs we
are going to do this training process you can see in my case it says 50 but you can just select
whatever value you want for your training process you can just train this model for as many epochs
as you want so let me show you how I execute all the cells up to here and then I'm going to
show you this other cell which is getting all the results from this training process I'm just
going to execute all the cells one at a time and I'm going to resume this video once the training
process is completed okay now the the training process is completed so we can just continue and
executing all the cells in this notebook was a very straightforward process so the only thing
you need to do is to execute all the cells one at a time, now let me show you this all the output
we created once we execute this model you can see that there is a huge output and this is going
to give you a lot of information regarding the training process let me show you if I scroll
up... over here you can see that for all of our training epochs we have a lot of information of the
training process we can see what's the loss value and we can see a lot of information which is very
important so this is all the output we got once we train this model and now let's get back here
because now the next step in this process will be to get all the results from this training process,
let me show you something if I scroll down again you can see it says results were saved here in
runs pose train7 so this is where we have saved all the results from this training process
and now the only thing we need to do is to copy these results into our Google Drive so we can
just download these results and we can analyze the results and so on right, you can see it says train7
because I have already trained this model many many times when I was preparing this tutorial
so this is the seventh time I trained this model and this is why we are saving the results here
and also pleasee remember to change this value before you execute this cell, this has to be the
location of your directory in your Google Drive now I have already executed this cell so I have
already copied the results into my Google Drive so let me show you if I go back to my Google Drive
you can see that this is the directory which is called runs, if I open this directory you can see I have
another directory which is called pose and then I have all these results from all the training
processes I have done so far this is the last one which is train7 and you can see that this
is a lot of information to just knock yourself out analyzing all these training process all the
performance of the model and so on, so I'm going to show you a couple of images I'm going to show you
this image over here which is a very quick way to analyze the performance of your model right this
is how the predictions look like on a few images and if we see a few of them for example this one
over here you can see that we are predicting the keypoints very accurately, maybe it's not
a perfect detection but I would say it's a very good detection anyway we are detecting all the
corners of this license plate the same goes here the same goes here, here... also here, I would say we
have a very good detection with this model so this is a very quick way to analyze the performance
of this model with only a few samples this is the same over here with some other images so you can
see it's not perfect we have a few mistakes, but this is... I would say it's a very good model
nevertheless so this is going to be pretty much all for this tutorial this is how you can train
a pose detector on your own custom data using yolov8 and remember this was a very quick and a very
high level description of this process, I didn't show you all the details which were involved
in this process and if you want a much more comprehensive explanation please remember to take
a look at the other video I showed you before and in that other video oh my God I show you so many
details and I show you for example how to evaluate the performance of this model but I show you in
a much more comprehensive way right I show you how to analyze this plot you have over here which
is all the results from the training process this is a much more detailed video which I definitely
recommend you to watch but 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.
