Repository dedicated to storing models trained on LARD_V2 with Ultralytics framework, and provided under AGPL-3.0 License

Repository

Models are divided into 3 categories:

• yolo_v8_models/ folder contain all the models trained from a yolov8 architecture with different label configurations, as described in our paper (only on IN_ODD or on both IN_ODD and IN_EXTENDED_ODD).
• yolo_v11_models/ folder contains all the models trained from a yolov11 architecture with different data source configuration, as described in our paper: in single-source configurations, in leave-one-out configurations, or on the complete dataset.
• piano_calibration_model/ folder contains the model trained on the piano detection task used during the calibration process of the different simulators, as described in our paper.

Training reproductibility

The models provided here are trained with Ultralytics framework using the yaml file provided. The training is based on the command line interface of Ultralytics, for example:

yolo train data=LARD_V2.yaml model=models/yolo11n.pt

We used different type of folders from different data sources (GES, BingMaps, ArcGIS, XPlane and Flight Simulator) and specific tags (IN_ODD and IN_EXTENDED_ODD) for the training of the different models described in our paper.

Data

Except for the models dedicated to piano_calibration, all the models provided here were trained on LARD_V2 which is a publicly available dataset for runway detection.

• The LARD V2 dataset is available on HuggingFace at huggingface.co/datasets/DEEL-AI/LARD_V2
• The repository for generating new image data is available in open source at github.com/deel-ai/LARD

Finally, the dataset for piano_calibration is also available on HuggingFace at huggingface.co/datasets/DEEL-AI/Runway_Thresholds

Paper

The paper will be available in the proceedings of the ERTS 2026 conference and soon on Arxiv.

Licence

AGPL: GNU Affero General Public inherited from Ultralytics

• The provided pretrained YOLO weights are under AGPL-3.0 (per Ultralytics’ licensing statement). Please keep the LICENSE with any redistribution.
• Our libraries (provided at https://github.com/deel-ai) are independent. If you package/distribute a combined application that includes Ultralytics (AGPL) code, the distribution must comply with AGPL.