YOLOv12

This page describes how to use the YOLOv12 implementation by the original authors with LightlyTrain.

Note

YOLOv12 is a custom fork of a specific version of the Ultralytics package. For this reason, YOLOv12 is not fully integrated with LightlyTrain and has to be installed manually.

YOLOv12 Installation

Important

Be aware that this will overwrite any current installation of ultralytics, so it is best to do in a new virtual environment.

Please install YOLOv12 directly from GitHub through:

pip install git+https://github.com/sunsmarterjie/yolov12

In case you are facing a version mismatch issue using CUDA and FlashAttention:

FlashAttention is not available on this device. Using scaled_dot_product_attention instead.

you can fix it by running the following commands:

pip install flash-attn --no-build-isolation

You can verify the results by:

python -c "import flash_attn; print('FlashAttention version:', flash_attn.__version__)"

and a successful installation will give you:

FlashAttention version: <some-version>

See this GitHub issue for more information.

Pretrain and Fine-tune a YOLOv12 Model

Pretraining or fine-tuning a YOLOv12 model is the same as doing so with any supported Ultralytics model. The only difference is that the config file is named yolov12.yaml instead of yolo12.yaml in the official Ultralytics releases.

Below we will provide the minimum scripts for pretraining and fine-tuning:

Pretrain

Python

import lightly_train

if __name__ == "__main__":
    lightly_train.train(
        out="out/my_experiment",                # Output directory.
        data="my_data_dir",                     # Directory with images.
        model="ultralytics/yolov12s.yaml",      # Pass the YOLO model.
    )

Command Line

lightly-train train out="out/my_experiment" data="my_data_dir" model="ultralytics/yolov12s.yaml"

Fine-tune

Python

from pathlib import Path

from ultralytics import YOLO

if __name__ == "__main__":
    # Load the exported model.
    model = YOLO("out/my_experiment/exported_models/exported_last.pt")

    # Fine-tune with ultralytics.
    data = Path("my_data_dir/config.yaml").absolute()
    model.train(data=data)

Command Line

yolo detect train model=out/my_experiment/exported_models/exported_last.pt data="my_data_dir"

Supported Models

The following YOLOv12 model variants are supported:

• ultralytics/yolov12n.yaml

• ultralytics/yolov12n.pt

• ultralytics/yolov12s.yaml

• ultralytics/yolov12s.pt

• ultralytics/yolov12m.yaml

• ultralytics/yolov12m.pt

• ultralytics/yolov12l.yaml

• ultralytics/yolov12l.pt

• ultralytics/yolov12x.yaml

• ultralytics/yolov12x.pt

Check the Ultralytics page for more information on our support of earlier YOLO variants.