shanexf/cardd-damage

CarDD — YOLO detection format

Car damage photos from CarDD (Wang et al., CarDD: A New Dataset for Vision-based Car Damage Detection, IEEE T-ITS 2023), converted to Ultralytics YOLO box labels.

• 4000 images, 6 classes: dent, scratch, crack, shattered_glass, broken_lamp, flat_tire
• images without any box removed
• split: original CarDD split

split	images	dent	scratch	crack	shattered_glass	broken_lamp	flat_tire
train	2816	1806	2560	651	475	494	225
val	810	501	728	177	135	141	62
test	374	236	307	70	71	69	32

Use

from huggingface_hub import snapshot_download
import yaml
p = snapshot_download("shanexf/cardd-damage", repo_type="dataset", local_dir="cardd")
d = yaml.safe_load(open("cardd/data.yaml")); d["path"] = p
yaml.safe_dump(d, open("cardd/data_local.yaml", "w"))
from ultralytics import YOLO
YOLO("yolo11m.pt").train(data="cardd/data_local.yaml", epochs=50, imgsz=640)

Licence

CarDD is distributed by its authors under their licensing agreement (https://cardd-ustc.github.io/). Use is subject to that agreement; do not redistribute without permission.

@article{wang2023cardd,
  title={CarDD: A New Dataset for Vision-Based Car Damage Detection},
  author={Wang, Xinkuang and Li, Wenjing and Wu, Zhongcheng},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  year={2023}
}