Fine-tune YOLOv8 easily and for free
A simple way to fine-tune your model own YOLOv8 model for free.
Dataset
Firstly you must choose a subject. For the tutorial I decided to create a dataset on player detection in Minecraft (a video game) you can download it if you want.
So to begin take photos or screenshots of your subject, go to roboflow create an account or sign in, and create a new project.
Import all the images you need.
Click “save and continue”, wait and get to “annotate/Manual Labeling/assign images”, and finally “start annotating”.
Frame all your objects and do it for all the images.
shortcuts :
— “n” for annotating mode
— “arrow” for changing image
When you have annotated all the images you need, you can quit and add the images to the dataset.
(My arrows have the same quality as my English)
Now it’s almost finished you must just go to the “generate” tab.
Push the button “continue”*2 and “create”. Wait a little … and it’s good. You just have to export the dataset in YOLOv9 format.
Training
For the training you just need a Google account and an Ultralytics account.
Firstly you must go to the Ultralytics page and create a new dataset.
Now create a new model and attach to it the dataset, copy the code in the «google colab» tab. Go to Google Golab create a new notebook, choose T4 GPU, paste the code, and execute it. The model is training!
You can see the loss evolution on the Ultralytics hub. And after the training process you can download the model and use it in code.
You can see some examples of utilization.
I hope you have been interested in this arcticle and if this is the case you can clap it. =)
