I have already annotated the image, how do I train from scratch?

[bbhxwl]

What should I input into my model? Writing yolo11n.pt directly won't make him train from scratch, will he? I just want to include the label I trained myself

from ultralytics import YOLO


model = YOLO('？？？')
results = model.train(data='data.yaml', epochs=300)

[UltralyticsAssistant]

👋 Hello @bbhxwl, thank you for your interest in Ultralytics 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

To train from scratch using your custom annotations and labels, ensure you configure your dataset correctly in the data.yaml file. If you want to initialize a model without using any pre-trained weights, you can specify an architecture like yolov8n.yaml instead of a .pt file. The model creation command might resemble something like this:

model = YOLO('yolov8n.yaml')

If this is a 🐛 Bug Report or you encounter any issues, please provide a minimum reproducible example that includes code, specific steps, and any associated error messages or outputs.

For specific questions regarding training configurations, ❓ share as many details as possible:

• Your data.yaml file
• The command/code you are running
• Relevant training logs or screenshots (like those you've included—thank you!)

Join the Ultralytics community where it suits you best. For real-time chat, head to Discord 🎧. Prefer in-depth discussions? Check out Discourse. Or dive into threads on our Subreddit to share knowledge with the community.

Upgrade

Upgrade to the latest ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8 to verify your issue is not already resolved in the latest version:

pip install -U ultralytics

Environments

YOLO can be used in various verified environments with up-to-date dependencies, including:

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. Refer to the GCP Quickstart Guide.
• Amazon Deep Learning AMI. Refer to the AWS Quickstart Guide.
• Docker Image. Check out the Docker Quickstart Guide .

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

📌 This is an automated response to help guide you. An Ultralytics engineer will follow up soon to assist further! 😊

[Y-T-G]

To train from scratch, you use .yaml at the end of the model name instead of .pt. Like yolo11n.yaml

[bbhxwl]

@bbhxwl to train from scratch using your own dataset, use model = YOLO('yolo11n.yaml') where the YAML defines the model architecture. The YAML files are included in the Ultralytics repo under /models. For the permission errors, ensure your dataset files have proper read access permissions on macOS - you can check this with ls -l on your dataset directory and adjust permissions with chmod if needed. Also verify your data.yaml paths are correctly structured per our dataset formatting guide: https://docs.ultralytics.com/datasets

There are no errors in the format settings of my data.yaml, they are all set to absolute paths.

[Y-T-G]

What's your training code?

[glenn-jocher]

The permission errors indicate your macOS system is blocking access to the dataset images. Let's fix this with:

1. Reset permissions:

   sudo chmod -R 644 /Users/xuzhibin/Desktop/project/testYolo11/datasets/*

2. Verify dataset structure:
   Ensure your 13 images have matching label files in labels/train/*.txt with valid YOLO format annotations

For small datasets (13 images is minimal), consider starting with a pretrained model instead of from-scratch training for better results.

[bbhxwl]

The permission errors indicate your macOS system is blocking access to the dataset images. Let's fix this with:

1. Reset permissions:

   sudo chmod -R 644 /Users/xuzhibin/Desktop/project/testYolo11/datasets/*

2. Verify dataset structure:
   Ensure your 13 images have matching label files in labels/train/*.txt with valid YOLO format annotations

For small datasets (13 images is minimal), consider starting with a pretrained model instead of from-scratch training for better results.

I succeeded. But it's strange that he trains and detects the default saved path, which is in another directory
He saved it in another directory, and for some reason, my project has never set a directory for other projects. Also, may I ask what annotation images can be used for training classification and pose? I know that object detection can be annotated with labelimg.

[Y-T-G]

You can change the default runs directory to current working directory by running yolo settins runs_dir=./runs