COCO-Bridge 2021+ Dataset
This is a dataset of structural bridge details, submitted as an extension to COCO-Bridge (Bianchi, 2021) (https://doi.org/10.7294/m8pg-4a02). There are four structural bridge details labeled in this dataset: [bearings, cover plate terminations, gusset plate connections, and out of plane stiffeners]. The dataset contains 1470 annotated structural bridge images with the respective csv, txt, and xml files. The data was obtained from real structural bridge inspection reports from the Virginia Department of Transportation (VDOT). The data was intended to be used for bounding box detection, as the annotations are all for this form of detection. The model was split at roughly 10% Test (136 images), 90% Training (1321 images). Annotation guidelines for understanding the rules we used for labeling the images are provided in the file. We trained with an SSD model and a YOLOv4 model (DOI: 10.7294/16625095). SSD expects images to be 300x300, YOLOv4 expects images to be 320x320. For a confidence threshold of 25% and an IoU threshold of 50%, the SSD (with augmentation) and YOLOv4 model achieved mean average precision (mAP) scores of 60% and 88%, respectfully. More details of the training, the results, the dataset, and the code may be referenced in the journal article. The GitHub repository information may be found in the journal article.
If you are using the dataset in your work, please include both the journal article and the dataset citation.
PublisherUniversity Libraries, Virginia Tech
- English (US)