dronefreak/uavdt-yolov9s

YOLOv9s Finetuned on UAVDT

Fine-tuned YOLOv9s object detector on the UAVDT benchmark dataset, trained and evaluated as part of DetectionBench -- a framework for reproducibly benchmarking modern object detectors with identical training recipes and evaluation metrics across multiple real-world datasets.

---

Performance

Metric	Score (%)
mAP@50	31.82
mAP@50-95	18.71
Precision	39.83
Recall	38.12
F1 Score	38.96
Parameters	7.3M
FLOPs	27.6B (at 640 px)

---

Evaluation Protocol

Metrics reported in this model card are computed on the UAVDT test split, using DetectionBench's standard evaluation pipeline (detectionbench-evaluate).

---

UAVDT Model Zoo

Every model DetectionBench has trained and evaluated on UAVDT so far, for full transparency -- see DetectionBench for the smaller, curated comparison set used on the project README.

Model	mAP@50	mAP@50-95	Precision	Recall
YOLOv26s	32.98	19.61	43.86	40.38
RF-DETR Nano	32.78	20.31	73.6	66.98
RF-DETR Small	32.62	20.21	73.83	71.63
YOLOv9s	31.82	18.71	39.83	38.12
YOLOv8m	31.42	18.8	40.27	37.79
YOLOv10m	30.12	17.33	40.13	35.68
YOLOv9t	29.42	17.03	35.75	36.47
YOLOv11s	29.1	17.16	34.32	37.31
YOLOv26n	28.88	16.79	33.14	35.66
YOLOv10s	28.85	16.48	36.53	33.16
YOLOv11n	28.56	16.3	38.04	32.26
YOLOv8n	27.8	15.34	35.42	33.61
YOLOv10n	27.17	15.16	33.3	31.21
YOLOv8s	27.12	15.33	34.65	31.87

---

Per-Class Performance

Class	mAP@50	mAP@50-95
car	72.93	40.38
truck	9.73	6.34
bus	12.8	9.42

---

Evaluation Visualizations

Precision-Recall Curve

F1 Curve

Confusion Matrix

Normalized Confusion Matrix

---

Dataset

This model was trained on UAVDT. For the full dataset description, provenance, license, and citation, see the dataset card:

https://huggingface.co/datasets/dronefreak/UAVDT

Classes

• car
• truck
• bus

---

Usage

Install Dependencies

pip install ultralytics huggingface_hub

Load Model from Hugging Face

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

weights = hf_hub_download(
    repo_id="dronefreak/uavdt-yolov9s",
    filename="best.pt"
)

model = YOLO(weights)

Run Inference

results = model.predict(
    source="image.jpg",
    conf=0.25
)

results[0].show()

---

Training Configuration

Setting	Value
Dataset	UAVDT
Framework	Ultralytics YOLO
Training Toolkit	DetectionBench
Epochs (configured max)	30
Epochs (actually trained)	15
Early Stopping Patience	8
Batch Size	auto (Ultralytics AutoBatch)
Image Size	1024
Optimizer	AdamW
Initial Learning Rate	0.0005
Seed	0

---

Repository Contents

best.pt
results.csv
args.yaml
BoxPR_curve.png
BoxF1_curve.png
BoxP_curve.png
BoxR_curve.png
confusion_matrix.png
confusion_matrix_normalized.png
val_batch0_pred.jpg
uavdt_yolov9s_showcase.jpg
assets/demo_banner.mp4
assets/demo_banner_poster.jpg
README.md

---

Related Resources

• UAVDT dataset card on Hugging Face
• DetectionBench -- reproducible benchmarks for modern object detectors on real-world datasets
• UAVDT paper preprint (arXiv:1804.00518)
• UAVDT project website (official data source)

---

Training Framework

This model was trained using DetectionBench, an open-source framework for benchmarking object detectors across multiple real-world datasets with a common pipeline.

Features include:

• A dataset-adapter registry for converting real-world datasets into a canonical format
• Identical training/evaluation recipes across model families (Ultralytics YOLO/RT-DETR, RF-DETR)
• Hardware profiling (latency, FPS, VRAM, parameters, FLOPs)
• One-command reproducibility via versioned Hydra configs

If you find this model useful, please consider starring the repository.

---

Known Limitations

• Severe class imbalance: car (94.6%) dominates the annotated boxes, while truck (3.1%) and bus (2.3%) are rare -- per-class accuracy on the minority classes is measured on comparatively few examples, and every model here scores far lower on them than on car.
• Very small objects: the median box covers only 0.14% of the image area (mean 0.26%), so this is a hard small-object regime and absolute mAP values are low for every architecture; the numbers are best read as a relative comparison between models, not as a production-quality detector.
• Video-derived, highly correlated frames: the ~40.7k labelled images come from 50 video sequences, so consecutive frames are near-duplicates. UAVDT's 50 tracking-only sequences have no detection labels and are excluded. The validation split is carved out of the training sequences by sequence (not by frame) to avoid leakage, but effective diversity is far lower than the image count suggests.
• Different density per split: instances per image are 15.7 (train), 28.0 (valid) and 22.7 (test), because the splits contain different sequences -- validation metrics are not directly predictive of test metrics.
• Research-use-only data: UAVDT is distributed "for research purpose only" with no redistribution grant, so the dataset is not mirrored here -- obtain it from the official source (see the Dataset section above) and check its terms before any use beyond research.

---

Citation

If you use this model in your research, please consider citing:

1. The UAVDT dataset (see below)
2. The original YOLOv9s architecture (see below)
3. The other model architectures shown in the Model Zoo/External Comparison tables above, if you reference their results
4. DetectionBench, the training/evaluation framework used to produce this checkpoint

@InProceedings{du2018unmanned,
  title={The Unmanned Aerial Vehicle Benchmark: Object Detection and Tracking},
  author={Du, Dawei and Qi, Yuankai and Yu, Hongyang and Yang, Yifan and Duan, Kaiwen and Li, Guorong and Zhang, Weigang and Huang, Qingming and Tian, Qi},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  year={2018}
}

@article{wang2024yolov9,
  title={YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information},
  author={Wang, Chien-Yao and Yeh, I-Hau and Liao, Hong-Yuan Mark},
  journal={arXiv preprint arXiv:2402.13616},
  year={2024}
}

Other architectures compared against on UAVDT in this model card:

RF-DETR

@inproceedings{robinson2026rfdetr,
  title     = {RF-DETR: Real-Time Detection Transformer},
  author    = {Robinson, Isaac and Robicheaux, Peter and Popov, Matvei and Ramanan, Deva and Peri, Neehar},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2026},
  url       = {https://arxiv.org/abs/2511.09554}
}

@article{oquab2023dinov2,
  title={DINOv2: Learning Robust Visual Features without Supervision},
  author={Oquab, Maxime and Darcet, Timoth{\'e}e and Moutakanni, Theo and Vo, Huy and Szafraniec, Marc and Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin and others},
  journal={arXiv preprint arXiv:2304.07193},
  year={2023}
}

YOLOv10

@article{wang2024yolov10,
  title={YOLOv10: Real-Time End-to-End Object Detection},
  author={Wang, Ao and Chen, Hui and Liu, Lihao and Chen, Kai and Lin, Zijia and Han, Jungong and Ding, Guiguang},
  journal={arXiv preprint arXiv:2405.14458},
  year={2024}
}

YOLOv11

No official YOLO11 research paper has been published by Ultralytics; the most commonly cited independent architectural analysis is used instead:

@article{khanam2024yolov11,
  title={YOLOv11: An Overview of the Key Architectural Enhancements},
  author={Khanam, Rahima and Hussain, Muhammad},
  journal={arXiv preprint arXiv:2410.17725},
  year={2024}
}

YOLOv26

@article{jocher2026yolo26,
  title={Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models},
  author={Jocher, Glenn and Qiu, Jing and Liu, Mengyu and Lyu, Shuai and Akyon, Fatih Cagatay and Kalfaoglu, Muhammet Esat},
  journal={arXiv preprint arXiv:2606.03748},
  year={2026}
}

YOLOv8

No official YOLOv8 research paper has been published by Ultralytics; this is their own recommended software citation instead:

@software{jocher2023yolov8,
  author = {Glenn Jocher and Ayush Chaurasia and Jing Qiu},
  title = {Ultralytics YOLOv8},
  version = {8.0.0},
  year = {2023},
  url = {https://github.com/ultralytics/ultralytics},
  license = {AGPL-3.0}
}

@software{Saksena_DetectionBench_2026,
  author = {Saksena, Saumya Kumaar},
  title = {DetectionBench: Reproducible Benchmarks for Modern Object Detectors on Real-World Datasets},
  url = {https://github.com/dronefreak/DetectionBench},
  year = {2026}
}