ฝัง
- เผยแพร่เมื่อ 17 มิ.ย. 2024
- YOLOv8 is the latest installment of the highly influential YOLO (You Only Look Once) architecture. YOLOv8 was developed by Ultralytics, a team known for its work on YOLOv3 and YOLOv5.
Following the trend set by YOLOv6 and YOLOv7, we have at our disposal object detection, but also instance segmentation, and image classification. The model itself is created in PyTorch and runs on both the CPU and GPU. As with YOLOv5, we also have a number of various exports such as TF.js or CoreML.
In this video, I'll take you through a step-by-step tutorial on Google Colab, and show you how to train your own YOLOv8 object detection model.
Chapters:
0:00 Introduction
0:51 Overview
3:09 Setting up the Python environment
5:36 New API: CLI vs. Python SDK
8:51 Prepare the YOLOv8 object detection dataset
12:29 Train YOLOv8 model on custom dataset
13:54 YOLOv8 model evaluation
16:47 YOLOv8 model inference on images and videos
18:44 YOLOv8 model deployment and inference via hosted API
19:58 Conclusion
Resources:
🌏 Roboflow: roboflow.com
🌌 Roboflow Universe: universe.roboflow.com
📝 How to Train YOLOv8 Object Detection on a Custom Dataset Blogpost: blog.roboflow.com/how-to-trai...
📓How to Train YOLOv8 Object Detection on a Custom Dataset Notebook: colab.research.google.com/git...
⭐ YOLOv8 repository: github.com/ultralytics/ultral...
📄 YOLOv8 docs: v8docs.ultralytics.com
📓 Learn more about YOLOv8 and other Computer Vision models with Roboflow Notebooks: github.com/roboflow/notebooks
🎬 Automatically Label Computer Vision Data: • Image Labeling API | A...
🆕 What's New in YOLOv8 Architecture: blog.roboflow.com/whats-new-i...
💯 RF100 Dataset blog.roboflow.com/roboflow-100
Stay up to date with the projects I'm working on at github.com/roboflow and github.com/SkalskiP! ⭐ - วิทยาศาสตร์และเทคโนโลยี
Epic, waiting on the next parts. Cheering for Roboflow & Ultralytics teams !
Thank you Roboflow!! Always keeps us updated🤝🤝
Thank you for wanting to be updated!
Good diction, i'm currently learning English, and I find your pronunciation much easier to understand compared to most people, not sure why. Great video!
Finally is really simple to use for industrial projects!
Exactly!
Great video! Definitely useful to train your own dataset since yolov8 was originally trained on COCO so it may not work for special applications!
Great video! very helpful to get started with Yolov8
Very simple and useful… Thank you so much
Stay tuned, we will soon post YOLOv8 instance segmentation too ;)
Thank you Brother , The Roboflow is just Amazing and super easy to use.
You are the best. Roboflow 💙
It was soooo helpful! Thank you!
I love to hear that!
Nicely explained!
Thank you!
This video has made my DL work so much easier! Thx for the great tutorial on YoloV8 and connecting it to Roboflow workflow😊😊
Hi! It's peter from the video. Thats what I wanted to hear! 💜
That is a great tutorial! Thanks sir.
Thanks a lot!
good job Roboflow and Ultralytics team... I want similar videos in docker... Thank you
easy tutorial to follow, thanks!
Awesome!
thank you for great video :)
Thanks a lot for watching :)
Jest trochę ten polski akcent 😄, świetnie wytłumaczone!
thank you, useful, great content
Thanks a lot!
@@Roboflow welcome sir, getting errors at code, please solve sir
@@RAZZKIRAN I'm happy to help. Could you please create a thread on ours discussions page: github.com/roboflow/notebooks/discussions/categories/q-a ?
Thanks for your efforts. How to generate a confusion matrix if the training is stopped due to no improvements in loss?
When you train a model, the weights are saved in the runs folder correct? We do not have to upload to roboflow?
Also using v5 you used to be able to put custom data sets in a certain place in the directory structure which you reference with a data.yaml file. Is this not the case anymore?
Where are the datasets stored now? Is it no longer on the local machine in the directory? Do we have to use roboflow and upload there?
Thanks
thank you so much
wow thx!
Thanks for the video! Does yolo segment things in geometries that aren’t rectangles? For instance, if you wanted to segment different planar surfaces on a roof from aerial imagery.
if you want to seperate the teams would you do that in the labelling (annotation) when preparing the dataset or later in the algorithm based on for example the jersey color? Thanks for the great video!
Yes, you would need to do the annotation, but the model will predict poorly on new teams.
hi, thank you for the documentation.
I have a problem about predict images. i trained my model and predict image grayscale but i come into view error : ValueError: axes don't match array.
What should I do? I must predict image grayscale.
great !
I trained our network with custom dataset. The training period lasted a long time. I want to test the performance of my test set with the network I trained at another time. Is there any other solution than retraining the network?
what is the best.pt file? ı just downloaded it and closed everything else. did ı save my model? can ı use it?
Hey, in the video at 11:26, u said that u have posted the links in the description but i couldn't find that link. Can u please check for that link?
It will be very helpful for me.
do we need to specify device=none parameter while training to access GPU even after changing GPU settings under Runtime bar in Colab? When i execute training even after changing to GPU settings under runtime tab, i get device=none mentioned in the output of training
How can I prepare dataset_params if I have a dataset structured as follows: Vid1/images and labels, Vid2/images and labels, and so on up to Vid100? The dataset consists of multiple videos, with each video stored in its own folder.
hi! thank you for the great video. how can i write down the confusion matrix summary?
can we run live inferencing on yolov8 models without using ultralytics library like we used to in previous version of yolov5? I want to setup the codebase for just running inferencing without using the ultralytics library.
Hey, for my dataset its taking so much time to train the model and i am running out of GPU limits, earlier i was getting error in training but i added batch size 8 ,now its training the model but taking too much time and GPU. Can u suggest what should i do?
How did you added the labels at the video? When I try using a test video it does the process and shows the classification but the video remain the same
I used my trained model path "project.version(dataset.version).deploy(model_type="yolov8", model_path=f'{HOME}/runs/detect/train3/')"
nice
Hello.. I trained a model on YOLOv8 and it worked very well. I had a question. Like I wanted to make some changes in the Predict file when I was running yolov8 in my system locally on CLI. I wanted to integrate a alarm system when any thing is detected. So I wanted the location of the Predict file. I have download the ultralytics repo in my pc.
Hi 👋It is Peter from video! If you want to do changes, I encourage you not to install via pip but clone repo and install it the old way.
Hi, thanks for the great knowledge and information you have provided. Please could you help use the custom model created to create an app/UI using Streamlit or Flask with Webcamp?
hello,
Can we add the object trained with the custom dataset to the other 80 object YOLO weights? As a single weight of 80+1. Can we increase the weight of the existing 80 objects?
thanks.
normally yolo weight consists of 80 objects.
Can we add new objects to objects of this weight by training with custom datasets?
Sorry sir, permission to ask if the code that runs on google collab can be run on pycham
Could you explain how to edit the bounding box to visualize them with a better appearance?
Did you try our supervision pip package? We offer custom annotators for bounding boxes there.
Cool
on 18:08 you mentioned that you downloaded the result and played the video. Do you mind sharing how to do that?
at 13:45 you can utilize gpu by typing "device=0" so it can train faster
Doesn’t it train by default on GPU if it is accessible?
Can this work with raspberry pi? (Pi4b with thermal camera.
Hi, I have trained my model with 10 classes, but in case if I want to detect only a specific class by getting the input from the user what should be modified. I tried passing it as an argument and it worked out. But when I tried using
# integer input
class_to_be_detected = int(input())
# print type
print(type(class_to_be_detected))
when I tried pass 'class_to_be_detected' in class argument I am getting error as 'TypeError: new(): invalid data type 'str''
Please help me with the same
Hey. I annotated some 60 odd images. but once I am done, i can't see the "submit for review" option at the top left. what do i do?
Awesome video and I can hardly wait to train a few models for V8. Only issue I saw however is that the datasets I had for v5 models won't work for v8? Am I missing something or some of the training images and labels format chnaged ? I tried to run a training session for v8 using a dataset exactly as I had it for v5 but it throws an error about the lables. Something along the lines that labels are not actually available ? Can you please give me a short hint on what do I have to do to train a model in v8 using the very same dataset I used for v5? I'd like to compare the behaviur of V8 to V5 and this would the best way of doing it. Obviously if I load the images into roboflow and annotate using that there are no issues at all all works fine. Awesome work and thank you for the video .
There is no change when it comes to actual labels - the txt files are exactly the same. However YOLOv8 team decided to change path management logic. Your dataset contain also a data.yaml file. And inside you’ll find train, test and val paths. If I’m not mistaken you need to change them to train/images, test/images and valid/images respectively. In short to relative paths from dataset root directory to image subdirectories.
@@Roboflow Okay. Thanks . I'll try that and see what happens 🙂
Hello, thanks for the amazing tutorial. the older version of roboflow is working fine but I could not figure out why is the the roboflow 1.0.1 or later throwing following error just by importing it.
TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'
Could you please create new issue here: github.com/roboflow/notebooks/issues and give me a bit more detail?
Hello, I collected data with video to use in my project. Can I use this video to train my model or do I need to train the model using only photos? If I can use it, how should I label the data in this video? I would appreciate it very much if you could help me on this issue.
What is the best way to resize images to 640x640? Is it stretching or resizing keepeing proporions?
Hello, how can I assess the yolov8 model with test dataset where I can get Recall , Precision, mAP, confusion matrix, curvs, and accurecy.
Thanks!
I tested the model on some images I can see the results in text but the bounding boxes on the pictures won't save.
Hi it is Peter from the video. I just updated our notebook for object detection. Could you check one more time? The issue should be fixed now :)
I have uploaded video for object detection. However it takes a very long time to download the analyzed video. How can i shorten this time? Or how can i watch without downloading?
What algo you are using with yolo for image detection
Thanks for the fantastic video. My prediction picture data won't store in /runs/detect/predict at the moment I run with mode=predict. What should I do? I repeatedly ran my code, but it failed to save. Your advice is essential. thank you in advance.
Try to use save=True parameter
can we use this model directly in our machine after the training ? and how ?
Please create a video on how to utilize multi GPU in a single node.
00:05 Yellow V8 is the latest object detection model that fine tunes much faster than its predecessors
02:29 Yellow V3 and Yellow V5 repositories have almost 45,000 stars on GitHub and will solve previous issues in the Yellow V8 project.
04:58 Importing Yolo from ultralytics and running inference
07:36 Creating a dataset for training the YOLO model using Roboflow
10:16 Use Cinema to label images and create a dataset for training
13:00 The training has been completed and the results are satisfactory.
15:41 Training the models could take longer and yield better results.
18:04 Yellow V8 model can be trained and deployed for inference using a single line of code.
20:16 Comparing yellow V8 to previous object detection models
Crafted by Merlin AI.
Hey man! Still trying to figure out where you drag and dropped the images from.If anybody knows,do let me know. Thanks in advance :)
Thank you RoboFlow!
Please how to show in this case number of detected objects on the test image?
We have this exact example in the supervision readme. Take a look: github.com/roboflow/supervision?tab=readme-ov-file#-quickstart
10/10
do you have a tutorial on how to use roboflow?
Hi Mate When I run the code in the last section it's says https is forbidden
and How Can I run video and detect object like you ? or Can I Detect Live with Web Cam ?
So I’m new to all of this, I’m assuming this is all in python, is there any way to turn this from .pt files to .pb files
can you show with the coordinate please at the bounding box . I am very need that
Hi, is yolo using Standardscale for image detection
How can I make yolov8 detect only the highest confidence interval class for a given class? IE. filter all the players and just display the player with most confidence interval?
Iam new to YOLO and object detection....
What is the relationship between YOLO and PyTorch? This video didn't mention PyTorch but in other tutorials on YOLO, PyTorch was brought up but it was not clear how they integrate or if that is even needed...Any help is greatly appreciated!
Hi it is Peter from video! PyTorch is a general framework used to build neural networks. YOLO is one of neural networks that is written in PyTorch.
@@SkalskiP oh! YOLO is written in pytorch! Didn't know that... So no other separate processing by pytorch is needed, thanks!
while running inference on a custom dataset i get this error: TF-TRT Warning: Could not find TensorRT.
After then it detects properly. I just don't get the visual of its detection!
Hey so when I'm running the video, the model is detecting the objects and the count frame by frame. But I'm not able to see the video to see which objects it detect. Can you please help me out?
Do you pass show=True ?
Amazing video! i had a doubt though. You said we could directly upload the images along with the annotations. I uploaded images (png files), but I am unable to upload the annotation files (.json files). any suggestions?
Hi! It is Peter from the video. Sorry I didn'. noticed your comment previously. Are you still experiencing those problems?
@@SkalskiP unfortunately yes. i have a dataset with images and their corresponding .json files annotating bounding boxes for the objects to be classified. since this is a large dataset, I cannot manually annotate boxes like you showed in this video. hence I am trying to upload the .json files, but I'm unsure as to why this doesn't work
@@thilakcm1527 and this is standard COCO json?
@@SkalskiP yes
@@SkalskiP any clue on what i can do?
is there any detailed blog or tutorial for yolov8 classification. with custom dataset. am having problem with data parameter in model.train
Me too! It raises FileNotFound: None error
is there custom train for pose estimation?
There is no submit for review option now what to do to annotate the objects for rest all images
Can we convert annotated JPEG images into VOC data format?
what is imgsz here in *****!yolo task=detect mode=train model=yolov8s.pt data={dataset.location}/data.yaml epochs=25 imgsz=800 plots=True*****. Does my value of imgsz depend on something according to my inputs?
Hey what about pose detection training??
Is there something wrong with your code, or did recent ultralytics version change break the code?
Do you experience some problems now?
guys after inferencing I am not getting where the results are storing. And after inferencing also it is not showing. Please help me through this
Thank you very much it worked very well with my own custom dataset, I have a question how do you download the result on the video?
You mean from Colab?
I just had save and it worked, yes from collab
Could you send the video? predict part is not coming.
@@omerkaya5669 could you be a bit more specific?
@@Roboflow %cd {HOME}
!yolo task=detect mode=predict model={HOME}/runs/detect/train2/weights/last.pt conf=0.1 source=/content/25km.mp4
I run this code, but I can't see where the video is saved.
Hi, thank you for the lovely video.
Although I am getting this error when I initiate training: FileNotFoundError:
Dataset '/content/project_name/data.yaml' not found ⚠, missing paths ['/content/datasets/project_name/valid/images']
Hi 👋🏻 could you create a bug report here: github.com/roboflow/notebooks? Please provide us with as many details as possible.
This happens to a lot of the files for some reason, it also did it for 'predict3' and required a change to ''predict2'. I'm not sure whether this is intentional or not : |
@@Roboflow For me this occurred because the 'train' file (/runs/detect/train/weights/best.pt) was displaced for some reason. All the weights had saved to 'train3'rather than the preestablished location within the given code. Try and find where the files are saving and change the file destinations (i.e., from /runs/detect/train/weights/best.pt to /runs/detect/train3/weights/best.pt).
Did you use pre-train model on COCO dataset and just updated the weights or did you actually created a new instance YOLOv8?
Hi! If I understand your question correctly, you ask if we used transfer learning or not?
@@Roboflow No no, it was something else, but now I am encountering the same error as Rachaer CR ERROR-Error executing job with overrides
I also used my own dataset.
Please help
@@mohammadhaadiakhter2869 could you create a issue here: github.com/roboflow/notebooks/issues? It would be great if you could include link to your version of notebook.
@@Roboflow Just one more thing
from ultralytics import YOLO
model = YOLO("yolov8n.pt")
import cv2
import numpy as np
cap=cv2.VideoCapture(0)
while cap.isOpened():
ret,frame=cap.read()
results=model(frame)
cv2.imshow('frame',np.squeeze(results.render))
if cv2.waitKey(10)& 0xFF==ord('q'):
break
cap.release()
cv2.destroyAllWindows()
I am encountering an OS error, any idea how to solve it?
@@mohammadhaadiakhter2869 is that your help request here: github.com/roboflow/notebooks/discussions/44
👌🏾
is there a way to download the model?
hello sir i have one probelm i train it on 100 epochs but when it perform detection on video or test images .. then detection have been completed but the predict folder cannot make ..means the output video or test images are no where
Hi it is Peter from the video. I just updated our notebook. Could you check if you still experience that issue?
Is this working good in real time rtsp stream fetched from CCTV cameras??
Hi it is Peter from video! 👋Honestly I haven't checked that out, but it sounds like a very good topic for next tutorial video.
@@SkalskiP yes please do that for the next video.
@@afrahthahir7179just keep in mind that it works good mostly on big size objects
Hi, do I need to a paid account or free acount will work as well?
hello sir after detection there is no predict directory where my video and test images are stored... detection on video perform and completed successfully but predict directory is not
Hi it is Peter from video. I just pushed slightly updated version of our notebook. I think you should not experience that problem now. Could you take a look?
hello, i have some problems on confusion matrix part where the font size that shows number in that confusion matrix is too small and quite hard to read. is there any way that i can make the font size on that confusion matrix bigger? is there any modification on the code?
No. But you can use supervision confusion matrix: github.com/roboflow/supervision
After extracting the downloaded dataset zip file (from my roboflow account), to which folder (following this video) should I upload the valid, train and test folders and, the data.yaml and README text files?
Dou you use manual download or pip package?
@@Roboflow I used manual download from the roboflow online app. I click the "Export Dataset" button. Then select "YOLO8" format. Then select "download zip to computer" option and finally click the "Continue" button. Trouble is, the snippet I copied/pasted into my Colab crashed when ran because it does not include a workspace name. :(
With the roboflow deplyement, is it possible to get the weighs or smh so it can run offline on the computer? because using the api we need to be online. Or do i need to like dowloads the weights from the google collabs training?
Did you find an answer?
I also Wanna know
It always glitching on my machine because of dependencies conflict: numpy, pandas versions for v8, and pycocotools for yolo nas
The option to export in yoloV8 format is now available. Do I still have to export the annotated dataset to yoloV5?
Good question. It is better if you'll export in YOLOv8 format.
Hi, how to add more than 3 person in a project
Thanks for your video ! When I tried to train the polar panels dataset on roboflow, I got the following error. How can I fix it? Thanks !!
Sizes of tensors must match except in dimension 1. Expected size 1364 but got size 0 for tensor number 1 in the list. Sentry is attempting to send 2 pending error messages
Do you think YOLOv8 is able to differentiate between individual raccoons with a dataset containing 130 individuals and 7500 pictures? Each individual would be one class.
Just try this experiment. At least, the model would be able to differentiate between some of them. I would be happy to see the results!
😍😍
I have finished training my custom dataset using yolov8 and have successfully tested some images, so how can I get the coordinates of the bounding box of those images?
Hello, did you find a way?