YOLOv8 Architecture Detailed Explanation - A Complete Breakdown
ฝัง
- เผยแพร่เมื่อ 27 ต.ค. 2023
- Hey AI Enthusiasts! 👋 Join me on a complete breakdown of YOLOv8 architecture.
In this captivating video, I'll be your guide as we explore the intricacies of YOLOv8 architecture, one of the latest and most powerful object detection models. We'll unravel its secrets, dissect its components, and demystify how it achieves mind-blowing real-time object detection. 🕵️♂️
Prepare to be amazed as we delve into:
1. The unique YOLOv8 convolutional block
2. The new C2f block
3. The bottlenecks
4. The spatial pyramid pooling fast (SPPF)
Join me for this exciting journey, where we'll decode YOLOv8 together! 🎥 Don't forget to hit that subscribe button and ring the notification bell to stay updated on YOLO. Let's geek out together! 🤓
Do you want to know how to easily PRUNING and MODIFYING YOLOv8 architecture?
And how to greatly IMPROVE SPEED up to 4x and ACCURACY up to +21 mAP by modifying YOLOv8, click this link
👉 bit.ly/Improve-YOLOv8
👉 bit.ly/Improve-YOLOv8
#yolov8architecture #yolov8 #objectdetection #artificialintelligence #computervision #deeplearning #yolo - วิทยาศาสตร์และเทคโนโลยี
How to greatly IMPROVE SPEED up to 4x and ACCURACY up to +21 mAP by modifying YOLOv8, click this link
👉 bit.ly/Improve-YOLOv8
thank you so much for detailed explanation!
You are welcome
Thank you it was so simple and so informative!
no problem
Thank you so much Dr. Hidayatullah. This was beautifully explained. Just wow
You're welcome. I am glad it is useful
Nice work doctor really appreciated
Thank you, alhamdulilah
Thanks for neat, but impactful explanation
You're welcome!
Very clear explanation thanks
you're welcome
thank you so much for detailed explanation,
and i have a question,in yolo v8 is image divided into grid cell before entering the CNN layer or after the CNN layer?
to my knowledge: before
Hi sir, Thank you so much for this explanation but could you please explain what exactly is this 'max output channel'?
Great video!
thanks
is it possible for yolo to train data whose channels>3 ?
Thanks alot for the good video! How can we access those images of the structure of YOLOv8? I need to include them in my report and cite them if you have a website :)
Thank you for the video it is very usefull, but i have a question. Shouldn't there be another track for the confidence prediction in the detect block? Or where does this value come from? Is it already in the Cls?
please refer to glenn jocher (YOLOv8 author) answer at this link github.com/ultralytics/ultralytics/issues/4149
Thank you for the detailed explanation. Could I ask what software you used to draw the architecture?
yes. We used app.diagrams.net/
May I ask why the feature map with higher height x width specializes detecting small objects? Does it have something to do with the channel? (9:42)
Not with the channel, but with the resolution.
Thank you for the explanation! Do you have any other explanation for the YOLOv8 segmentation model?
or maybe in your udemy course that explain the YOLOv8 segmentation architecture
Not yet. But you can follow this thread if you need it now github.com/ultralytics/ultralytics/issues/1289
is there any discount coupom for your course in udemy ?
Yes. You can use this coupon www.udemy.com/course/yolo-performance-improvement-masterclass/?couponCode=BLACK-FRIDAY
Your explanation was amazing!, Do you have any tutorials on how to implement pruning on YOLO8?
Yes. I have a special tutorial on architecture pruning for YOLOv8. Plus TensorRT optimization and Openvino Quantization to greatly boost your YOLOv8 speed! And many more.
You can check it here: www.udemy.com/course/yolo-performance-improvement-masterclass/?referralCode=A87DA906397E1027C6C5
any discount ?
@@ccss4892 ok then if you want. You can use this link www.udemy.com/course/yolo-performance-improvement-masterclass/?couponCode=4U-SUBSCRIBER
I dont understand it fully but thanks anyways Dr.Hidayatullah
hello sir, can i use the images here for a paper?, i will include the references. Thank you
Go ahead!
Does YOLOv8 only accept images of size 640 x 640 during training? What if I want to use 3840 x 2160 image?
YOLOv8 accept any size during training as well as during inference. If you set the parameter imgsz into 640, YOLOv8 will resize your image (what ever the size) into 640x640
@@Dr.Priyanto.Hidayatullah Thanks for the quick response. After the model is trained using 640x640 img, in my case I get a very pixelated inferred image so I want to use the actual size. The problem I am facing is that I want to train on RGB image of size 3840x2160, but during training CUDA quickly runs out of memory. For reference, I am training on ~16 GB GPU.
I know this might be a silly question but I would like to know if there are any metrics that would allow me to access how much memory it would required in this case.
@bipinkoirala2962 you can use patch training and for memory limitation you can play with batch size
This is more comprehensive answers to your question
github.com/ultralytics/ultralytics/issues/1658
Do you need that resolution for inference? Or could you use a lower resolution (downscale) and the scale the results back up to your original resolution (upscale)?
how to create new blocks to improve the accuracy , for detecting small objects or adding new blocks like GAM , how do we decide where to add ???
You can edit the yaml file and add the blocks there.
Adding new kind of blocks required you to edit the source code.
Where to add? That is challenging question. I have not found any resource saying where exactly the right place to add a block. I once asked my prof. He said: "You have to try and see the result." iterate this process.
@@Dr.Priyanto.Hidayatullah thank you sir
@@dalinsixtus6752 no problem
@@Dr.Priyanto.Hidayatullah sir i need to change the color of bounding boxes during runtime if certain conditions are satisfied.is there any way to use built in function rather than changing the ultralytics source code .
I understand your question. However, I am not sure there is such function.@@dalinsixtus6752
can you tell me number of layers used in each block like in backbone, neck and head?
For this, you have to go to the source code and count it. I have not do the counting.
@@Dr.Priyanto.Hidayatullah your lecture is very good thanks for sharing
@@hinamohsin7561 you are very welcome :)
Izin bertanya. Apakah dalam implementasinya, Yolov8 bisa dikombinasikan dengan arsitektur lain?
Menurut saya bisa. Pas mengerjakannya harus sabar dan telaten
@@Dr.Priyanto.Hidayatullah Apakah ada harga promo untuk mengikuti kelasnya di Udemy Prof?
@@dosenswasta www.udemy.com/course/yolo-performance-improvement-masterclass/?couponCode=C00946B7B921C18E4ED9 Ini saya buatkan. Kuponnya tinggal berlaku 2-3 HARI lagi saja.
Mohon diperhatikan, kupon ini untuk kursus peningkatan performa YOLOv8 yang memang membahas cukup dalam tentang arsitektur YOLOv8 dan modifikasinya
ada kelas khusus penggunaan apple silicon? untuk trianing YOLO
Kalau khusus Apple, di kelas kami belum ada. Tapi pengguna Apple bisa menggunakan Google colab. Kami sediakan file untuk dieksekusi di google colabnya (yolov7 dan YOLOv8). Bisa di searching di udemy YOLOv8, lalu cari yang instruktur nya saya.
Sir please share architecture diagram .
Where is the difference between the YOLOv8 object detection architecture and classification architecture?
Based on architecture file github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/models/v8/yolov8-cls.yaml
In the classification, there is no SPPF block in the backbone. After that in the head, there is only a classify block
is this an explanation of how feature extraction from yolov8 works?
in one POV, yes
Tq so much man💖
you're welcome brother
@@Dr.Priyanto.Hidayatullah can you explain loss functions of yolov8
@@vigneshvicky6720 the complete list of loss functions in YOLOv8 is here: docs.ultralytics.com/reference/utils/loss/
a friendly explanation can be found here:
www.linkedin.com/pulse/losses-weights-yolov8-dsaisolutions-x1ggf/