LibreYOLO9t-obb

Ultra-experimental LibreYOLO YOLO9-tiny oriented object detection weights fine-tuned for vehicle OBB detection.

These weights are ultra-experimental development weights. They were produced while validating LibreYOLO OBB training support and should not be treated as production or benchmark-official weights.

Source

Initialized from LibreYOLO9t detect weights. The YOLO9 implementation in LibreYOLO follows the permissive MultimediaTechLab/YOLO MIT-licensed lineage.

Fine-tuned on My First Project - v8, a Roboflow Universe UAV vehicle OBB dataset provided by a Roboflow user: https://universe.roboflow.com/bobo-48pem/my-first-project-ewwrm

This model was not trained on DOTA.

Dataset

Dataset license: Creative Commons Attribution 4.0 International (CC BY 4.0).

Classes:

• bike
• bus
• car
• other_vehicle
• taxi
• truck

Local development export metadata reported 932 source/export images, YOLOv8 Oriented Object Detection format, and Roboflow preprocessing/augmentation.

Metrics

UAV-OBB validation split, best epoch 5:

• mAP50: 0.605965
• mAP50-95: 0.365930
• mAP75: 0.403682
• precision: 0.242334
• recall: 0.700350

Modifications

The checkpoint is a lean LibreYOLO inference checkpoint produced from the best EMA training checkpoint. Optimizer, resume config, raw train model state, and EMA resume buffers were removed before upload.

License

These fine-tuned weights are released under CC BY 4.0 to preserve the attribution requirements of the training dataset. See LICENSE and NOTICE.