YOLOv9: How to Train on Custom Dataset from Scratch with Ultralytics

Inside my school and program, I teach you my system to become an AI engineer or freelancer. Life-time access, personal help by me and I will show you exactly how I went from below average student to making $250/hr. Join the High Earner AI Career Program here 👉 www.nicolai-nielsen.com/aicareer (PRICES WILL INCREASE SOON)
You will also get access to all the technical courses inside the program, also the ones I plan to make in the future! Check out the technical courses below 👇
_____________________________________________________________
In this video 📝 we are going to create the whole computer vision training pipeline on a custom dataset. We will take raw images, auto label them with roboflow and export it into a Google Colab notebook. We are then going to train the new YOLOv9 model with Ultralytics in just a few lines of code. Then we download the model and see how to use it in a custom Python script.
If you enjoyed this video, be sure to press the 👍 button so that I know what content you guys like to see.
_____________________________________________________________
🛠️ Freelance Work: www.nicolai-nielsen.com/nncode
_____________________________________________________________
💻💰🛠️ High Earner AI Career Program: www.nicolai-nielsen.com/aicareer
⚙️ Real-world AI Technical Courses: (www.nicos-school.com)
📗 OpenCV GPU in Python: www.nicos-school.com/p/opencv...
📕 YOLOv7 Object Detection: www.nicos-school.com/p/yolov7...
📒 Transformer & Segmentation: www.nicos-school.com/p/transf...
📙 YOLOv8 Object Tracking: www.nicos-school.com/p/yolov8...
📘 Research Paper Implementation: www.nicos-school.com/p/resear...
📔 CustomGPT: www.nicos-school.com/p/custom...
_____________________________________________________________
📞 Connect with Me:
🌳 linktr.ee/nicolainielsen
🌍 My Website: www.nicolai-nielsen.com/
🤖 GitHub: github.com/niconielsen32
👉 LinkedIn: / nicolaiai
🐦 X/Twitter: / nielsencv_ai
🌆 Instagram: / nicolaihoeirup
_____________________________________________________________
🎮 My Gear (Affiliate links):
💻 Laptop: amzn.to/49LJkTW
🖥️ Desktop PC:
NVIDIA RTX 4090 24GB: amzn.to/3Uc7yAM
Intel I9-14900K: amzn.to/3W4Z5Cb
Motherboard: amzn.to/4aR6wBC
32GB RAM: amzn.to/3Jt2XVR
🖥️ Monitor: amzn.to/4aLP8hh
🖱️ Mouse: amzn.to/3W501GH
⌨️ Keyboard: amzn.to/3xUGz5b
🎙️ Microphone: amzn.to/3w1F1WK
📷 Camera: amzn.to/4b4Ryr9
_____________________________________________________________
Timestamps:
0:00 Intro
0:42 Upload Dataset
2:33 Auto Label
5:50 Generate & Export Dataser
9:00 Train in Collab
18:20 Inference in Python Script
20:37 Outro
Tags:
#YOLOv9 #objectdetection #computervision

[@NicolaiAI]

Join My AI Career Program www.nicolai-nielsen.com/aicareer Enroll in My School and Technical Courses www.nicos-school.com

[@miguelvalverde7873]

Hey. Do you have any idea why YOLOV9-t (the tiny model) weights are not available?

[@gatharajayaweera3169]

Thank you for this amazing video this is very useful.

[@NicolaiAI]

Thanks a ton for watching!

[@gatharajayaweera3169]

@@NicolaiAI can I know how to integrate this to a live video captured from an esp32-cam? Do you have any videos on that?

[@harshbhatt8250]

I am working on FSC 147 dataset and I will combine SAM and YOLO for counting and segmentation tasks, for trial of custom YOLO I was trying it on CarPK dataset and was facing some errors, but your video just released on the right time and I was able to solve ther problems I was facing with trial. The results I acheived were pretty good and therefore my professor agreed to my idea of combining SAM + YOLO for counting and segmentation tasks. I know YOLO can perform both segmentation and counting task, but we want to use SAM for counting, and my idea was a to add a layer of YOLO/ CNN model to accurately predict the objects. Thank you so much for this.

[@LokeshLokesh-sf7so]

Papa papa aree q

[@farhanonymous8666]

Hello Harsh, I am also kinda working on a similar project. Right now, I am using only YOLO for both segmentation and counting tasks but I want to use SAM for counting purposes and I really need help on this. Can we please connect somewhere and discuss it?

[@khouloudbelkhodja4586]

hello , How to resume YOLOv9 training an after interruption?

[@tamilgenius7684]

how to evaluate the yolo model in google colab notebook

[@oguzkurtoglu]

Can you make video about semantic segmentation?

[@chvitaalston7110]

Hey!The following error occurred while I was running, what should I do? RuntimeError: An attempt has been made to start a new process before the current process has finished its bootstrapping phase. This probably means that you are not using fork to start your child processes and you have forgotten to use the proper idiom in the main module: if __name__ == '__main__': freeze_support() ... The "freeze_support()" line can be omitted if the program is not going to be frozen to produce an executable.

[@nachiadhi]

Also after training custom dataset , the best. pt file, when I give any image(I.e things that I didn’t train) it’s still detecting it (I.e it’s detecting a bike as a car) ps I trained only car and not bike.. so I suppose my result image should not be predicted… please send help

[@NicolaiAI]

Comes down to the dataset you are training on

[@jigglejogglers2606]

Great video guide. Question though, is YOLOv9 possible to run on a video live stream instead of just uploading single images?

[@NicolaiAI]

Yeah for sure! Have tons of videos around that as well

[@jigglejogglers2606]

@@NicolaiAI Ive been playing around with using my GPU to speed up the video streaming perhaps? Could you make a guide that perhaps itulizes Nvidia CUDA to run these algorithms for object detection?

[@binhphan2928]

You can instruct yolov8 pruning with torch-pruning ?

[@binhphan2928]

@NicolaiAI

[@AlirezaFazeli-i7y]

How can I use gelan-c or gelan-e in Python? not cli

[@Mica_No]

what about live video detection?

[@ignis.valorant]

loving your video! is it possible to just train the model locally and not on google colab?

[@NicolaiAI]

Yup you can do it locally as well. Exact same code. And it will use the hardware available. Either GPU or CPU directly

[@akhilsajeevan1192]

Hi i have a doubt, if i have 4 classes and and train yolo v8 model with my custom data set . and i want to add 2 more new classes in the trained yolo model with new classes without loosing the weight of the first trained model. How to do that ? show me step by step procedure in simple steps. Do freezing can be done? If so show me that technique also .

[@NicolaiAI]

Show me show me. But no you can’t do that. You have to retrain

[@akhilsajeevan1192]

@@NicolaiAI when retraining i have lost weight of first traiing. The model now only knows the classes of last training. I don't want like that i want model to have both weights of first training and the last traing also. Is there any solutions for that?

[@rololop34]

What is the point of training a yolov9 model if the foundation model can already do almost perfect predicitons? Just knowledge distillation?

[@NicolaiAI]

Depends on what classes you want to detect. The pre trained model is only able to detect 80 different classes from the coco dataset

[@rololop34]

@@NicolaiAI Sorry, I meant the foundation model from roboflow, which annotated the cars.

[@NicolaiAI]

Ohh in that way. Those foundation models are too large to run in real-time and too expensive for the task. They are way too overkill and requires significantly more processing power which is unnecessary @@rololop34

[@rololop34]

@@NicolaiAI Thank you for answering. I just read on the ultraytics docs that YOLOv8 is approx. 866x faster than SAM-b on CPU.

[@yameenkhan127]

how about video and live footage

[@Mrsmith0119]

How to use this model results for practical use like autonomous robots

[@NicolaiAI]

In what way do you want to use it for? Any specific objects?

[@Mrsmith0119]

@@NicolaiAI I have certain types of obstacles and want to apply obstacles avoidance and by also identifying them and want to feed the results obtained from prediction to computer or microprocessor for performing tasks

[@NicolaiAI]

@Mrsmith0119 definitely check out the yolo world model as well here on my channel

[@nachiadhi]

Well even I want to deploy this model on a practical bot !

[@errrbrrr3821]

how to deploy these models into a raspberry pi or any edge device?

[@Mrsmith0119]

Same question i also want to ask

[@Player-oz2nk]

Yeah same here!

[@NicolaiAI]

Will definitely do it on both raspberry pi and jetson nano. Ultralytics have some nice guides on their documentation but I should definitely cover it since it looks like a lot of people want to see that. Thanks a lot for commenting!

[@nachiadhi]

Yeah pls do it

[@HarisKhan-ph9jl]

You have written yolov9 in the video and you are training yolov8. 😔 😔

[@NicolaiAI]

Nope 14:21 it’s yolov9 and training right after. Yolov8 from the code example from Ultralytics and then changed the model name

[@HarisKhan-ph9jl]

@@NicolaiAI Thanks for your quick response! You are absolutely right. I just watched the full video, and I apologize for the misunderstanding. I'm planning to run it on Kaggle and would love your assistance with that. Thank you so much! I just subscribed to your channel-keep up the great work!

[@NicolaiAI]

@HarisKhan-ph9jl thanks a ton!

[@LukasSmith827]

[@tamilgenius7684]

how to evaluate the yolo model in google colab notebook

[@tamilgenius7684]

how to evaluate the yolo model in google colab notebook