dronefreak/visdrone-rfdetr-medium

RF-DETR Medium Finetuned on VisDrone-DET

Fine-tuned RF-DETR Medium object detector on the VisDrone-DET 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.

---

Usage

Install Dependencies

pip install rfdetr huggingface_hub

Load Model from Hugging Face

from huggingface_hub import hf_hub_download
import rfdetr

weights = hf_hub_download(
    repo_id="dronefreak/visdrone-rfdetr-medium",
    filename="checkpoint_best_total.pth"
)

model = rfdetr.RFDETRMedium(pretrain_weights=weights)

Run Inference

detections = model.predict("image.jpg", threshold=0.25)

---

Performance

Evaluated on the VisDrone-DET test split, using DetectionBench's standard evaluation pipeline (detectionbench-evaluate).

Metric	Score (%)
mAP@50	39.62
mAP@50-95	21.99
Precision	70.62
Recall	46.76
F1 Score	56.27
Parameters	33.7M
FLOPs	N/A (not published upstream)

---

VisDrone-DET Model Zoo

Every model DetectionBench has trained and evaluated on VisDrone-DET 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
YOLO26m	49.11	29.24	60.21	49.76
YOLO11m	48.03	28.56	59.34	48.3
YOLOv9m	47.62	28.52	59.26	48.4
YOLOv10m	46.73	27.72	59.23	47.35
YOLOv8m	45.47	26.94	57.86	46.39
YOLOv9s	45.38	26.95	56.95	46.19
YOLO26s	44.87	26.43	56.44	45.41
YOLOv10s	44.51	26.28	55.97	45.61
YOLO11s	43.64	25.88	54.48	45.01
YOLOv8s	43.47	25.77	56.09	44.51
YOLOv9t	40.67	23.73	52.84	41.78
YOLO26n	39.9	22.95	51.14	42.08
YOLOv10n	39.8	23.08	51.22	41.5
YOLOv8n	39.69	23.03	52.02	41.36
RF-DETR Medium	39.62	21.99	70.62	46.76
YOLO11n	39.52	23.0	51.49	41.02
RF-DETR Small	39.13	21.68	64.38	49.47
RF-DETR Nano	37.92	20.88	69.02	46.21

---

Per-Class Performance

Class	mAP@50	mAP@50-95
pedestrian	31.03	12.36
people	25.29	9.08
bicycle	18.6	7.41
car	75.62	46.26
van	42.95	26.97
truck	50.0	32.35
tricycle	27.39	13.83
awning-tricycle	23.19	13.06
bus	64.29	43.83
motor	37.89	14.7
others	0.0	0.0

This model was evaluated with Supervision's detection metrics, which report mAP/Precision/Recall directly but don't produce a confusion-matrix plot the way Ultralytics' validator does.

---

Dataset

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

https://huggingface.co/datasets/Voxel51/VisDrone2019-DET

Classes

• pedestrian
• people
• bicycle
• car
• van
• truck
• tricycle
• awning-tricycle
• bus
• motor
• others

---

Training Configuration

Setting	Value
Dataset	VisDrone-DET
Framework	RF-DETR
Training Toolkit	DetectionBench
Epochs (configured max)	100
Epochs (actually trained)	48
Early Stopping Patience	20
Batch Size	12
Resolution	640
Optimizer	adamw
Learning Rate	0.0001
Seed	42

---

Repository Contents

checkpoint_best_total.pth
metrics.csv
config.json
visdrone_rfdetr-medium_showcase.jpg
assets/demo_banner.mp4
assets/demo_banner_poster.jpg
README.md

---

Related Resources

• VisDrone-DET dataset card on Hugging Face
• DetectionBench -- reproducible benchmarks for modern object detectors on real-world datasets

---

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 (42.21%) and pedestrian (23.12%) account for two-thirds of all annotated boxes in the training set, while awning-tricycle (0.95%) and tricycle (1.40%) are rare -- the others class has zero annotated instances in the training set entirely and is effectively unusable (always 0 AP).
• Extreme small-object density: ~53 annotated boxes per image on average, with roughly 69% of boxes covering under 0.1% of the image area -- consistent with VisDrone's aerial small-object detection challenge (objects captured from significant altitude).
• The original authors license VisDrone under CC BY-NC-SA 3.0 -- non-commercial research use only (see the dataset's homepage); this applies to any model trained on it, not only the raw images.
• These RF-DETR checkpoints were trained/evaluated directly through DetectionBench. The YOLO/RT-DETR rows in the External VisDrone Model Zoo comparison below were trained via a separate companion codebase, not reproduced inside DetectionBench -- see that collection for their own training details and caveats.

---

Citation

If you use this model in your research, please consider citing the dataset and the model architecture:

@article{zhu2018vision,
  title={Vision meets drones: A challenge},
  author={Zhu, Pengfei and Wen, Longyin and Bian, Xiao and Ling, Haibin and Hu, Qinghua},
  journal={arXiv preprint arXiv:1804.07437},
  year={2018}
}

@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}
}