Real Time Object Segmentation and Tracking using YOLOv8 on Custom Dataset: Complete Tutorial

hi guys in this video we will see how we can use yellow V8 for segmentation and tracking on any custom data set we will go through all the implementation step by step and I will help you to implement your lobby Aid with segmentation and tracking on any custom data set and for this uh tutorial we will be using this repository so you want to be a segmentation deep sort of jet packing with this repository uh will be used in this implementation so let's move towards the implementation please do watch the complete video and don't skip any part of the videos so that you can get full understanding of all the core concept so first of all before running the step we need to make sure that we have select the runtime as GPU uh to be sure just go to it here and stack I have select the runtime as GPU so I am quite fine over here so now we need to import all the required libraries from IPython dot display import image basically this library is used when we need to display any output image into our collab notebook let's say I have been using this Library over here like to display confusion Matrix I am using this Library uh to display a prediction on the validation batch I am also using this Library so good let's move ahead so first of all we need to run this plan because this Library we use from ipython.display import image as this Library we use to load any image output image into our Google call app notebook then we need to turn this GitHub repo so just click over here and just click over this and just paste it over okay so just now just run this cell so it might take one to two minutes okay that's done so just check where what is our present working practice so currently this is our present working directory so okay so now you can see that we have the cloud folder over here so just we need to move uh as we need to implement segmentation we do slot tracking so we need to move towards this folder so we are just uh going through here and just copy path and just uh paste it over here and just run this cell so if you can just run this cell from here as well or you can just click on shift and enter to run the cell now we need to install all the dependencies or you can say we need to install all the required libraries to implement this uh project because if we don't install all the dependencies uh some required liabilities which are not pre-installed might create issue like Hydra library or pandas Library numpadability so the libraries which are not installed will create an issue so it's better to install all the dependencies at the start might take few seconds so let's wait um it might take 100 seconds so Plus Moving ahead as well so as in this tutorial we need to implement segmentation so we need to move us towards the segmentation folder which is over here and just go here copy path and just paste it over here so now we are to into the segmentation folder so to implement this project we will be using this data set let me show you so for the implementation I am using this data set of drone traffic data set which is of multi-class uh this data set is available publicly on roboflow so I am using this data set for the implementation of this project you can see over here so this data set is already available publicly on roboflow so it is multi-class data set which contains around four different classes so we will be using this data set in this project and first as we are implementing segmentation so this is this linear segmentation instant segmentation data set of drone traffic which include different vehicles and cars bicycles Lorry and others okay so here we are just importing this data set uh from the from basically roboflow into our notebook so as I have already explained you how to import basically data set you just need to import this data set you just need to go to the data set and just click on download and just show download code and just copy this and just first before this you need to be signed in into your account so I have been signed in so just copy this and just paste it away you can just paste it over here okay so just now click on just run this cell so the data set will be uh downloaded from Rover flow into your Google collab notebook so it might take one to two minutes uh so that's the latency so the data set is being downloaded over here you can see that this is our data set over here so now as we are going to implement object tracking using uh deep sort so we need to download the Deep sort files into our collab notebook so I'm just downloading a deep sort files from my derive into the Google app notebook so just downloading a zip file which I will unzip in the next step okay now I will train the custom models for this just go here data set location so I have already trained the model for 30 bucks you can train it for high number of because just for this tutorial I have trained the model for 30 number of epochs so here are the results of the modern uh the model gives very fine results we have a good mean every precision as well so and here is the confusion Matrix uh so let me explain you the quad confusion Matrix status basically confusion Matrix is the chart that shows us how our model handle different classes for example in this case our model successfully detect 99 or 99 of the time that this is a bicycle while for only one percent load a Time the model is unable to detect anything okay for bike for bus about a hundred percent uh model successfully detect that this is a bus for 100 of the times and while for car that the model also detects uh 99 of the times that's the car so while four uh this is the point where we are getting a weight issue like or there are the directions are weak let me see so in all cases we are getting around very good results like for example we have four different classes bicycle Buzz car Lorry and for all classes we are getting around 99 to 100 of detection successfully order is directing a 99 200 percent of time uh successful detections are being done okay so let's move ahead and see so these are the model predictions on the validation badge so these are basically image these images are not basically used for the training so it is better to look take a look and see how our model is basically behaving so basically I have saved my model weights into the Google Drive so let me download the bits from the Google drive after uh running a training over here here I just saved the model bits into the Google Drive so I'm just downloading the model bits from the Google Drive and just check it so downloading the grids from the Google Drive and then we will validate the customer model as well so the size of the weights around 368 MB so let's download all the views and just check it out as well so now I have downloaded the views on the Google Drive you can see this is my Dev passport video file over here and here I have validated my custom model you can see over here so here of validation of custom model is being done so now uh we will download a sample video you can just skip this tab we don't need this I'm just basically previously I'm just uh importing exporting my builds from the core lab notebook into the drive because when I just run the script so just skip this step you don't require it so now I will download a demo video from my drive so test my model so just downloading a demo video from the drive okay download it now just run this cell and see what results do we get it might take one to two minutes so let's wait and see the results so the script is running so let me pause the video until it runs successfully and then get back to you and this process completes foreign and see what presents we get let's see over here so it might take one to two minutes again so I have been testing the model on two demo videos so I will show you the second demo video as well so let's first see with the first demo video so here are those words let's download it and see point again I'm just checking your Tower let's see that so you can see that we have the unique ID with each of the detective object plus we have the trails as well so we are able to get very fine results with our model like Communications are very good plus it is uh now let's test on the demo video 2 as well so just download it over here and just keep it might take one to two minutes again so let's so the currently script is going to run so it is just finishing here it's a short video so it will finish a little bit on the previous one so okay there's around it's done now now let's see what our results will get over here just checking the output video we got I will be sharing this polar file with you as well uh this will be this polar point will be uploaded into this Google Drive so you can just press the file from this Google sorry in this case will be uploaded so you can access this folder file from this GitHub reference thank you so here are our output videos and just download it so you can see about broader view of it so just downloading it and just download it so here are our results and you can see that detections are very good in the very good results so these are the results and so in this video we have implemented our object segmentation and tracking probability [Music] upload this files so you can download it from here as well so do subscribe the channel and in the next video will be coming into some new topic thank you for watching have a great day

Channel: Muhammad Moin

Views: 2,778

Keywords: object detection, objects segmentation, object tracking, yolo, machine learning, deep learning, yolov8, artificial intelligence, computer vision, research and development

Id: e-uzr2Sm0DA

Length: 15min 51sec (951 seconds)

Published: Wed Jan 25 2023