dronefreak/rdd2022-rfdetr-medium

RF-DETR Medium Finetuned on RDD2022 Road Damage

Fine-tuned RF-DETR Medium object detector on the RDD2022 Road Damage 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.

---

Detection Showcase

---

Performance

Metric	Score (%)
mAP@50	65.08
mAP@50-95	36.02
Precision	71.18
Recall	55.98
F1 Score	62.67
Parameters	33.7M
FLOPs	N/A (not published upstream)

---

Evaluation Protocol

Metrics reported in this model card are computed on the RDD2022 Road Damage test split, using DetectionBench's standard evaluation pipeline (detectionbench-evaluate).

---

RDD2022 Road Damage Model Zoo

Every model DetectionBench has trained and evaluated on RDD2022 Road Damage 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
RF-DETR Medium	65.08	36.02	71.18	55.98
RF-DETR Small	64.71	35.73	65.69	59.41
YOLOv8m	62.03	34.08	65.61	57.42
YOLOv8s	61.45	33.53	64.55	57.18
YOLOv26s	61.27	33.3	64.42	57.13
YOLOv26m	61.24	33.38	63.7	57.27
RF-DETR Nano	60.85	33.22	65.49	54.3
YOLOv8n	58.8	32.05	62.03	56.08

---

Per-Class Performance

Class	mAP@50	mAP@50-95
longitudinal_crack	59.57	31.99
transverse_crack	59.86	30.22
alligator_crack	66.22	35.37
pothole	74.68	46.48

---

Evaluation Visualizations

This model was evaluated with Supervision's detection metrics, which report mAP/Precision/Recall directly but don't produce PR-curve, F1-curve, or confusion-matrix plot images the way Ultralytics' validator does. See the Performance table above for Precision/Recall/F1 and the per-class table above for the full per-class mAP breakdown.

---

Dataset

This model was trained on RDD2022 Road Damage. For the full dataset description, provenance, license, and citation, see the dataset card:

https://huggingface.co/datasets/dronefreak/RDD2022

Classes

• longitudinal_crack
• transverse_crack
• alligator_crack
• pothole

---

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/rdd2022-rfdetr-medium",
    filename="checkpoint_best_total.pth"
)

model = rfdetr.RFDETRMedium(pretrain_weights=weights)

Run Inference

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

---

Training Configuration

Setting	Value
Dataset	RDD2022 Road Damage
Framework	RF-DETR
Training Toolkit	DetectionBench
Epochs (configured max)	50
Epochs (actually trained)	24
Early Stopping Patience	10
Batch Size	5
Resolution	576
Optimizer	adamw
Learning Rate	0.0001
Seed	42

---

Repository Contents

checkpoint_best_total.pth
metrics.csv
config.json
rdd2022_rfdetr-medium_showcase.jpg
README.md

---

Related Resources

• RDD2022 Road Damage dataset card on Hugging Face
• DetectionBench -- reproducible benchmarks for modern object detectors on real-world datasets
• RDD2022 paper preprint (arXiv:2209.08538)
• RDD2022 project repository (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

• Not comparable to the official CRDDC2022 leaderboard: the challenge test set has no public labels, so the test split here is a held-out 15% slice (70/15/15 split) of the publicly-labelled images, merged across countries -- scores are only comparable between the models listed in this card's Model Zoo.
• Class imbalance: longitudinal_crack (44.0%) is the most common class, while pothole (18.1%) and alligator_crack (17.9%) are the rarest of the four -- per-class accuracy differs noticeably between them.
• Four-class taxonomy only: the source data's 5th "other" bucket (block cracks, road repairs and country-specific codes, ~6.5k boxes) was dropped to match the four damage types the CRDDC2022 challenge scores, so those damage types are not detected.
• Sparse, thin targets: about a third of images contain no in-taxonomy damage (clean-road frames), with 1.5 boxes per image on average, and cracks are thin structures that are easily lost when large frames (some over 4000 pixels wide) are downscaled to the model's input size.
• Uneven country and imaging-setup mix: the images come from six countries and several capture setups (smartphone, dashboard camera, drone) in very different proportions, so performance can vary substantially by country and generalization to unseen regions or damage conventions is untested.
• Share-alike data: the RDD2022 images are CC BY-SA 4.0 -- see the Dataset section above for attribution and the dataset card for the full terms.

---

Citation

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

1. The RDD2022 Road Damage dataset (see below)
2. The original RF-DETR Medium 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

@article{arya2022rdd2022,
  title = {RDD2022: A multi-national image dataset for automatic Road Damage Detection},
  author = {Arya, Deeksha and Maeda, Hiroya and Ghosh, Sanjay Kumar and Toshniwal, Durga and Sekimoto, Yoshihide},
  journal = {arXiv preprint arXiv:2209.08538},
  year = {2022}
}

@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}
}

Other architectures compared against on RDD2022 Road Damage in this model card:

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}
}