Yolov8 FULL TUTORIAL | Detection | Classification | Segmentation | Pose | Computer vision

[@ComputerVisionEngineer]

Did you enjoy this video? Try my premium courses! 😃🙌😊 ● Hands-On Computer Vision in the Cloud: Building an AWS-based Real Time Number Plate Recognition System bit.ly/3RXrE1Y ● End-To-End Computer Vision: Build and Deploy a Video Summarization API bit.ly/3tyQX0M ● Computer Vision on Edge: Real Time Number Plate Recognition on an Edge Device bit.ly/4dYodA7 ● Machine Learning Entrepreneur: How to start your entrepreneurial journey as a freelancer and content creator bit.ly/4bFLeaC Learn to create AI-based prototypes in the Computer Vision School! www.computervision.school 😃🚀🎓

[@DhanushPalla]

in classification how to find f1score,accuracy prevision etc yolov8 pls explain

[@MahDudeIDC]

Its not just how you relay and explain information, its your whole personality and energy that just makes your videos feel relaxed and easier to understand. 10/10 mad props. Thank you so much Phillipe, you're amazing !

[@ComputerVisionEngineer]

Thank you so much for your support! 🙂🙌

[@Paula-ek4ed]

you are the best i swear to god. I am truly going to mention in the acknowledgements section of my master thesis

[@ComputerVisionEngineer]

Thank you for your support! Glad the content was useful and it helped you in your master thesis! 🙂🙌

[@moditagrawal1846]

Your videos are amazing! Clear explanations, easy to follow, and super helpful. Thanks for all you do! Can you please bring a video explaining Anomalib too?

[@ComputerVisionEngineer]

Thank you! I will try to do a video about it. 🙌

[@hernanbernal5784]

what a content! the community really will appreciate some content in spanish! thanks and regards from argentina :)

[@ComputerVisionEngineer]

Thank you! I may create a Spanish speaking channel in the future with these topics. Regards! 🙌

[@rataash_x]

Dude!! you are a very very very very rare gem.

[@ComputerVisionEngineer]

😂😂 Thank you for your support! 🙌

[@Py7Val]

Thank you very much. Very useful and clear explanations!!

[@cyberhard]

100% agree, you need to test the model on real unseen data.

[@欧阳狮子]

I'm a Chinese university student, and I think this is very comprehensive

[@ComputerVisionEngineer]

Thank you! Glad you enjoyed it! 😃

[@ИльдарАлтынбаев-г1ь]

Thank you very much for your work, pal

[@harveyp.1949]

Your channel is awesome, thank you so much!!!!

[@ComputerVisionEngineer]

Thank you for your support! 😃🙌

[@oscartejedorgarcia5897]

Eres un crack!!! muchas gracias por los videos

[@mardeenosman8979]

Thank youuuu the video is just perfect, clear, detailed enough to cover all the needed parts. So thank you so much I have 2 questions please; Can we use YOLOV8 for instance segmentation/ What is the format you exported the duck mask with and to which file format you have changed it? Thanks again ❤

[@MuhammadAdnan-tq3fx]

brother you did a great work. its really helpful content especially for the beginner. I watch your whole video, your way of teaching and explaining is very nice. Now i will implement this by watching your video. At the end i have two question for you first, how many images we can select for annotation in this tool? and 2nd question how we can download the open image v7 dataset for specific category ?

[@ComputerVisionEngineer]

Thank you! I am not aware of any limit on the number of images you can annotate using CVAT, please take a look at its specs to make sure! About how to download images using open image v7, the instructions are listed in this github repository github.com/computervisioneng/train-yolov8-custom-dataset-step-by-step-guide. I may make a video about that in the future. 🙌

[@michael3226]

Great video ❤

[@ComputerVisionEngineer]

😊 Thank you! 🙌

[@gbengaomoyeni4]

Thank you Felipe. Your videos are always top-notch. Please can you make video on Online Exam proctor, i.e preventing online examination malpractices and cheating through computer vision? We would like you to make the video ASAP

[@myetis1990]

eventually high quality content detected :) thank you sir It would be great to see high-performance low-power applications, for autonomous robots and drones. you did not mention fps performance but it is crucial for real-time applications, is there any optimization method for increasing fps?

[@rafaeldesantis7580]

Bro could you do a video about video classification ? It would be awesone, there is no good content about it on internet…

[@ComputerVisionEngineer]

Hey, ok noted, I will try to make a video about video classification! 🙌

[@rohitchan007]

​@@ComputerVisionEngineer You can do video classification on the word-level American sign language recognition(WLASL) dataset.

[@BertelSchmitt]

A video is a series of still images. To classify objects in a video, run each frame of the video through your classifier.

[@ahmedmkaouar3776]

Thank you very much for your efforts and how to add an email or sms notification when the object is detected?

[@eduardo9626]

Thanks for this video

[@ComputerVisionEngineer]

You are welcome! 😃🙌

[@giuliaghinassi1734]

Hi Felipe, first of all thank you for the amazing tutorial! I'm a complete newbie and I'm learning from scratch. I need to ask you thou where did you get the file "google_colab_config.yaml", because I can't wrap my head around it

[@kanyimdels.3gp]

Awesome

[@marouenbensaid9495]

@computerVisionEngineer I would like to ask you how the class background appears in a confusion matrix if you have only trained one class or one object with one data set thank you for your help

[@xichen2617]

Thank you for your video. Do you know how to fix class imbalance issue using loss function in yolov8?

[@СтаниславВедмедев-г6т]

This neural net only trainig or testing? I don't see images for testing

[@ahmedm2702]

Hey, great content there I have been watching your videos and learning alot, do you have any tutorials on how to train Faster RCNN using a custom dataset locally? Thanks

[@ComputerVisionEngineer]

Hey thank you! Glad you find the videos useful! 💪💪 You can train a Fast R-CNN on your custom data using detectron2, take a look at my video on detectron2 here kzbin.info/www/bejne/f2iyZayjiKmWfdk. Go to detectron2 model zoo (github.com/facebookresearch/detectron2/blob/main/MODEL_ZOO.md) and look for the one that says 'Fast R-CNN'. 😃🙌

[@Сергій-й5ь8с]

nice video

[@SpeerKING]

Hey very nice Video, i am very new to this topic. Where do you get the .txt files in the labels folder?

[@eranfeit]

Very nice , As for the segmentation , How can I generate this label folder of txt files if the images is not binary , but contain for example 3 different segmented objects in the same images ?

[@ComputerVisionEngineer]

Different class ids for the different categories. The first value of each line in the txt file is the class id.

[@VishalNaik-gq4vl]

i have downloaded the csv files , now how to convert it to yml or how to use for this model?

[@SA-oj3bo]

Hi, do you have a video for yolo beginners where they can learn how to use the existing model? I need to make a solution where an ipcamera is connected to a rp5 or other computer to detect if there are both a person+bike on the image. Could you help? thx

[@shivakumarbhatkere1807]

Very good video, really appreciate the effort. I am doing a project to detect the car's fuel pannel is open or closed, how many images should i train the model for?

[@ComputerVisionEngineer]

Thank you! It is hard to say how many images you need to train that model, try with a few hundreds and go from there! 🙌

[@mobinmohammadi6562]

Hello. I have several objects in one photo for segmentation from one class. Now, when I want to convert the production mask to a binary image (0,1), it only displays one object of that class, if you have to display all the objects of that class. My code is as follows. results = model(image_1,classes=2, show=True) for result in results: for j, mask in enumerate(result.masks.data): mask = mask.numpy()*25 mask = cv2.convertScaleAbs(mask) cv2.imshow('mask', mask) I want all the objects of a class to be displayed in the binary image. please guide me.

[@simplesecomumcidadao7592]

What if we wanted to do this with C++ instead of Python???

[@sholehuye2737]

maaf pak, izin tanya setelah data berhasil di latih untuk dapat digunakan sebagai tracking objek langkah selanjutnya bagaimana pada tutorial pertama custom dataset alpacha.. 🙏

[@sluggy6074]

64gb of ram, task manager shows 15% of memory being used. I run training it stays at 25% then BOOM goes through the ceiling and crashes. I tried bringing the batch to 1 but it doesn't change anything, same exact result. Wild Edit: okay, command line worked without modifying batch or anything. It trained in like 10 seconds. Don't know why it's in such a rush

[@moses5407]

Can you suggest one it more free, annotated human post datasets suitable for use with this approach?

[@ComputerVisionEngineer]

Sure, have you tried the COCO-WholeBody dataset?

[@cwp132]

That's a good tutorial. I have a question. If there is coffee when several foods or drinks are classified into large categories in yolo, is it possible to classify them in coffee? For example, can we do object detection like espresso, americano, cafe latte, etc?

[@ComputerVisionEngineer]

Yes, it is possible to do object detection using espresso, americano, cafe latte, etc. as labels. It all depends on the data, if the drinks are served in transparent cups I would say it is possible. 💪🙌

[@lordeeveetheking4195]

Great video! I was just wondering how were you able to get the predictions for the prediction image? I've been running more data and epochs, but I still don't get any predictions... do you have any tips?

[@ComputerVisionEngineer]

If you are not getting good predictions with your model, in most cases the problem is the data. How many images are you using to train the model?

[@JeremyBirba]

Hello, thank you so much for this video ! I trained my model, but the labels can't be found and I do not know why. The labels have the same names as the images, they are in .txt , yolo format and in the folder "labels". And I have used the ultimate path. Do you have any idea why ?

[@abhisheknegi2888]

Yes I am encountering the same problem. Please pass the feedback if you get any solution.

[@JeremyBirba]

@@abhisheknegi2888 I have found a solution. In my case, I have the "images" and "labels" folders. In both of them, I have "train", "val" and "test". Then I put all the labels in the three folders present in the "labels" folder. I hope my explanation is clear. Good luck

[@024_habeeb7]

Please help, I downloaded the data from OID toolkit, but the labels containing .txt files has the name of the class instead of a float/int and hence during training it throws an error of cannot convert string to int. What to do in such case

[@MaiNguyen-vp6mt]

Thanks for sharing a great video. I have a question if I have 4 classes: person, car, truck, dog. In which I only want to get the keypoint of the person class and the rest of the classes will not have the keypoint, how do I need to build the data to train? Looking forward to your answer.

[@ComputerVisionEngineer]

Do you want to detect keypoints for the person class and only bounding boxes for the remaining classes? You could apply 2 models; a keypoint detector for the person class and an object detector for the remaining 3 classes.

[@河南省智障收容所副所]

yes @@ComputerVisionEngineer

[@河南省智障收容所副所]

how to do it@@ComputerVisionEngineer

[@milindchaudhari1676]

Hiii, great video, but i want to seek a help from u in my master's thesis work as im planning to use yolo for peach fruit detection, I've taken around 250 images having fruits using iphone but the fruits are overlapping as well as occluded by each other and leaves and no one is guiding me how to deal with annotating in such condition, also i have to increase my datatset using augmentation..., So please guide me with this, if i could somehow connect with u on e-mail or something i would share my dataset with u for reference. It would an immense help from u... Waiting for ur positive response 🌝!!!

[@mobinmohammadi6562]

Great, great, only in segmentation we can change the color of the mask? Because the color of my object is very similar to the color of the mask and this is not very interesting in the display.

[@ComputerVisionEngineer]

Do you mean when making inferences? Yes, you can choose any color. 🙌

[@mobinmohammadi6562]

@@ComputerVisionEngineer when i test the training network. It defines the border and the object with one color, for example, in your training video, you defined the ducks with pink color (in the output). Can this color be changed?

[@olanrewajuatanda533]

Please, did you use annotation for the videos you used to test the trained model? I wished you really showed that part. Also, does it make any difference using a totally different data to test the model? Also, is there a reason why you used the same training set as your validation set? Did you make use of the output of the training for the prediction?

[@ComputerVisionEngineer]

Hey, are you referring to the object detection tutorial?

[@olanrewajuatanda533]

@@ComputerVisionEngineer yes, I am.

[@ComputerVisionEngineer]

@@olanrewajuatanda533 I downloaded the dataset from the open images dataset v7. In that tutorial I used the same set for both training and validation just for simplicity, it is obviously better to use a separate set for validation.

[@olanrewajuatanda533]

@@ComputerVisionEngineer yes, I am referring to the object detection tutorial. Please, can you drop a snippet of how the test video can output the number of objects it detected in the video and the inference as well?

[@AliTarhini-c9h]

The weights folder is empty anybody know why? Plz help

[@ipass-q4m]

Hi , for the object detection part , can you make a video on how you download the data from open images dataset , i tried the steps but have error . Thank you in advance

[@ComputerVisionEngineer]

I will try to make a video about that 🙌

[@PelletsEaseFrance]

Nice video ! Is a CVAT pro account needed? I can only export the masks and labels but not the pictures.

[@ComputerVisionEngineer]

Thank you! CVAT pro account is not needed for this tutorial. In order to save the images you need to check the box 'save images'. 🙌

[@PelletsEaseFrance]

@@ComputerVisionEngineer If I check the box 'save images' I've this error: Could not export dataset for the task 210656 Error: exporting task dataset with images is not available. Upgrade the account to extend the limits. I'm trying image recognition to identify mathematical curves 🙂

[@ComputerVisionEngineer]

@@PelletsEaseFrance how many images are them?

[@PelletsEaseFrance]

@@ComputerVisionEngineer 225, I think is not a lot. And It's not possible to add additional images in the CVAT task detail page? If I've annotated 225 images and I want to add 200 other, how do?

[@programmingkitten1326]

Is this method actually used? Annotating by hand? Because for actually training a simple model you need at least 10k of images and for more complex ones up to a million

[@ComputerVisionEngineer]

Yes, sure, manually annotation of images is actually used. There are some technics to 'automate' the annotation process, but they usually require some degree of manual intervention too.

[@programmingkitten1326]

​@ComputerVisionEngineer I don't think manually annotating is used anywhere for a real project. Next time definitely show how to automate the process by generating the dataset. For example get 100 alpaca images and 100 background images and make 10000 combinations with them for the model to train

[@mariebocquelet1711]

Hi, I have a question. When we fill the config.yaml file, why don't we also indicate the path to the labels ? Thank you in advance :)

[@ComputerVisionEngineer]

By the way yolov8 works, the labels will be loaded from a directory called 'labels'. If not mistaken there is no way to change this from the config file. 🙌

[@bikramjitm]

Can anyone explain why confidence score is coming "0" after training ( 3 epochs ) and NO bounding box in coming in validation prediction? @ComputerVisionEngineer

[@ComputerVisionEngineer]

It is probably an issue with your data. How many images are you using? Are you sure the annotations are correct?

[@bikramjitm]

@@ComputerVisionEngineer i have done step by step exactly same as u have done in this video ( object detection ). I have downloaded the same dataset as of yours

[@ibbsinn]

Hey i am having issues, after training the model when i apply it to the video it detects nothing, like no object detection. i am using a dataset of 100 images

[@ComputerVisionEngineer]

Hey, you may need to use more images, 100 images may not be enough to train an object detector. Also, how many epochs are you using? Make sure you are not overfiting or underfiting. 🙌

[@juruns]

hello, I'm have a question, if do more than one class in CVAT, how I use the code "mask to polygon"? because apparently the binary segmentation just use in two classes, therefore 3 classes will appear (2 CVAT classes and background)

[@ComputerVisionEngineer]

If your data is comprised of more than one class, you would need to make some edits in the script. If you need assistance you could try creating an issue in the github repository. 🙌

[@kuliahovide7597]

Could i custom model (architecture) YOLOv8 yaml? In YOLOv5 we could custom model yaml. Thanks

[@ComputerVisionEngineer]

Hey, not sure if it is possible with yolov8, I will take a look. 🙌

[@rn7404]

bro, can you explain me what is the difference between classification and detection?

[@michael3226]

Classification is the process by which you want to categorize or group something based on some features or properties. Example: imagine you have a soccer ball and a tennis ball and you want to classify it. Obviously, a soccer ball is bigger than a tennis ball, so SIZE is what you used in grouping them. Detection on the other hand is the process of identifying if a particular thing is present in a time space. Example: imagine you want to know whether a soccer ball or tennis ball or both balls are present on a pitch. This does not group the balls, it tries to determine whether they are present or not. I hope that helps

[@ComputerVisionEngineer]

Image classification focuses on assigning a single label to an entire image, while object detection detects and localizes multiple objects within an image, providing both object presence and spatial information. 💪🙌

[@ZakariaDali1]

but how to collect or to download this data from Open Images ??

[@ComputerVisionEngineer]

Take a look at the github repository of object detection tutorial. You will find instructions regarding how to download the data from open images dataset v7.

[@nguyenthanhnhanng5662]

You look really similar to Pep Guardiola - Manchester City manager. Has anyone told you that?

[@ComputerVisionEngineer]

Nop, you are first one! I looked at a few images, I kinda look like him when he was younger, you are right! 😄

[@rabieelhakouni1167]

hi sir , i have a question , how i can optimize the loss function in classification task

[@ComputerVisionEngineer]

Do you want to change the loss function in yolov8 classification?

[@rabieelhakouni1167]

@@ComputerVisionEngineer yes sir

[@sanjaykumar-l3p3z]

How we can train the pre trained model?

[@ComputerVisionEngineer]

Hey, not sure if I understand your question, what do you mean?

[@mrsidi9535]

Brother please help me I have a problem I have a problem I have been trying to solve for three months and I could not solve it please I want to contact you on Facebook or Instagram or Telegram

[@not_amanullah]

what's the problem