Episode 17 | Train YOLOv8 Pose Model on Tiger-Pose Dataset Using Ultralytics HUB

hey guys welcome to new video in this video here we're going to take a look at Alo litic Hub we're going to take a look at how we can train our own custom models we can connect the data set we can upload and data set to the altic hub train our model with just a few clicks either on own computer Alo litics cloud-based systems as well but we can also train them in Google collab directly so we're going to use a tiger data set for post estimation we're going to cover the whole process I'm going to walk you through every single step so let's not just jump straight into Al lytics documentation up here at the top you can go inside the Hub we just quickly go and scroll through it and then we're going to go into the fun part seeing how we can actually like use it but we have this quick start we have the data sets projects and also models and then we're going to see how we can use them so they need a specific data set format where you can actually just uploaded in this video here I'm going to show you that for the tiger post data set so if you go inside data sets hop tab you can see the whole procedure data set format and all the steps that you need to run through I'm going to show you that in a live demo here so you don't have to go through the documentation we also have different projects and also models that you can set up again is not too important but we have different data set that we can use directly if we just go inside our data set tab up here at the top if you just want to try out um specific data set they also have this tiger post which is the one that we're going to use in this video so right now let's just go inside this post estimation data set so you can see the data set JL file you just need to specify the training path the validation path and so on um the key Point shapes flip indexes the different classes that you want so right now we only have a tiger that we want to post this maion on and if you want to download the data set you can just directly download the Sip file here so I've already done that and then we're going to see these results here and how we can train our own custom model with the ultra liix Hub so first of all here let's just go in and open up the data set I've have downloaded the data set we have this ZIP file I'm just going to open it up we have tiger post validation set training set we have our yl file here that I just showed you we have license and also read me file so we have everything here we just need the zip file and then we can go inside our data set upload data set choose post here let's just do tiger post YouTube we can add our description or else we can just drag and drop our SI data set directly you can choose any data set you just need to have it in a specific format so right now we have already uploaded it you can do it both for detection classification segmentation and post we're going to create our data set it will just take a couple of minutes and then everything should be in here they already have some predefined data sets if you want to do like for example like kugo data set B Drone ESU here as well if you want to do some um retail store analytics so they have some cool data set you can use directly out of the box or upload your own data sets so here we can see that it act like only takes probably around 10 seconds to before we have uploaded our data once we have uploaded our data we can go inside our projects we can see all the different projects the updation we can even create our own project Yol V 8 and Yol V 5 if you just go inside here we can directly create our own project we won't do that now we're just going to use some default ones here we can see YOLO V8 all the YOLO models that we have we can see the mean error position and also the size of our files now let's go inside the models we can see all these different models that you have trained on the ultr litic Hub or also the pre-trained ones the cool thing here is that once we have our data set we can directly go in and train our models they also have these Integrations if you want to integrate it with roof flow you have your account and so on but right now let's just go inside our data set wait until has it has uploaded okay so now our data set has been uploaded we can directly just go in and click on it we have 263 images a sing class we want to do Post estimation and we can also see this size of our data set so here we can just see a preview of all the different images that we have in our training set validation set and also our test set so just just take a look at the validation set again we can see all the key points for the specific joints and the key points that we want to detect with this data set so now we have everything here we can go up hit train model directly upload your data set click train model go in train the model use the model in your own project and application it can't be easier than this so after we created our project we can just go in and specify it here I'll just have this test project we can specify model name so tiger post we you can also choose the post architecture that we want to use so if you want to use Nano small medium large model ex large model and so on we can even see the accuracy and the speed so depending on like what you act like need let's go just go with the small one pretty good trade-off between accuracy and speed we can see we have some advant options if you want to use a pre- model or not if you want to have 200 images we'll have to use a pre-t model but again sometimes you want to train your models from scratch so yeah we're going to use a pre-ra model let's hit continue now we actually just get the different steps that we can use so first of all here let's see the first example where we can train it in Google collab notebook but we can also train it with our own agent which is our own computer so the first step here is to log into the altic top you get your API key then we can create an instance of our YOLO model then we just need to specify the model from this URL on the altic Hop it will do everything for you both the model and also the data set connect those together and then we can call model. train directly step two here is to open up Google cab notebook and then just follow the steps throw in the couple of lines of code here and then we can train our own models without doing anything you don't need to do any code modifications or anything we just need to click click click and we have trained the model so here we have different Advanced options so we can have specify number of epoch image size patient and also the cast strategy we can also specify this um the device here and also bat size so if you want to use like a custom bat size let's say we want to use aen bat size again we will get a warning here if you change any of the parameters so we should only change these parameters if you're an expert and that is the case here so let's go in and change the bat size so yeah we can specify the number of epoch let's just go with like 10 epochs in this example here or like 50 epochs so now we have all our high parameters we can then open up a Google collab notebook so we're going to do that it will connect automatically to the hop going to run the notebook we're going to log in with our API key so now we have opened up our Google cab notebook we're connected to runtime we just copy paste this code go inside here delete this override it and now we can just directly train it in here let's go back here we see that now we're now connected into altic Hub and we don't have to do anything we can just hit done we can even see the training progress in here so we can lock the different metrics we can see how far is the training we see that we're using this Google col notebook how many Epoch is remaining and so on and then you can see all the statistics in here so this is pretty nice for just tracking your model so now we can see that the first Epoch has started let's go back into Al litics hop and see if we actually get some metrics we should get that after the first Epoch is done now we start to see some metrics so right now we just have a single data point because we have only been running it for one Epoch now we can see that we're starting with the second Epoch so while I'm model training we can just see all these metrics right now they're basically just increasing per number of epochs that we're running so again our model is training fine and we should just wait until it converges so thank you guys for watching this video here I hope you have learned a ton and that you actually saw how easy it is to train custom uptic detection models you don't have to do anything just connect your data set set up a project use the model here directly from ALR litics but this is pretty awesome definitely test it out try it out yourself and letting me know in the comments how it works out and how cool you think this is so again thank you guys for watching this video here I hope to see you in one of the upcoming ones stay tuned bye for now

Channel: Ultralytics

Views: 712

Keywords: Object Detection, YOLOv5, YOLOv8, Tiger-Pose, Computer Vision, Pose Estimation, Ultralytics, Ultralytics Hub, PyTorch, Artificial Intelligence, Deep Learning

Id: Gc6K5eKrTNQ

Length: 7min 13sec (433 seconds)

Published: Wed Nov 29 2023