maco018/YOLOv12_traffic-delineator

Custom YOLOv12 Model for Detecting Traffic Delineators

This is a fine-tuned YOLOv12 model trained on a custom dataset with over 4100 labeled images to detect traffic delineators.

Data details

• Link to original dataset: https://app.roboflow.com/baschenk/traffic-delineators-detection/14
• 4131 originally annotated labels
• with augmentation: train set: 22918 images; valid set: 412 images; test set: 408 images

Model details

• Base model: YOLOv12x
• Classes: 1 (traffic delineator)
• Framework: PyTorch with Ultralytics

How to use

Prerequisites

Optional

If you want to run Torch with CUDA, make sure to get the correct version from: https://pytorch.org/get-started/locally/

Required

pip install ultralytics

Run inference

    from ultralytics import YOLO 
     
    model = YOLO(f'YOLOv12_traffic-delineator.pt')    
    results = model('test.jpg')  
    results[0].show()

Training Details

• Epochs: 300
• Input: Images were resized to 640x640
• Augmentations: Crop, Grayscale, Hue, Saturation, Brightness, Exposure, Blur, Noise
• GPU: The training was done on a GeForce RTX 3060 12 GB

Performance

Metric	Value
mAP50	0.607
mAP50-95	0.339
Precision	0.874
Recall	0.419

Citation

@misc{yolov12x_delineator_2025,
  title={A fine-tuned YOLOv12 model for traffic delineator detection},
  author={Pascal Schenk, Max Rädler, Mark Colley},
  year={2025},
  howpublished={\url{https://huggingface.co/maco018/YOLOv12_traffic-delineator}},
}