Fabric Defect

A custom dataset of fabric defects was constructed, utilizing images captured from garments with various imperfections. The dataset comprises 2468 high-quality images and each image to ensure quality. The dataset contains images of six types of fabric defects. The dataset has Button-hike (307 images), Broken-button (545 images), Hole (556 images), Color-defect (384 images), Foreign-yarn(334 images), and Sewing-error (317 images). To address potential limitations arising from a finite dataset size, data augmentation techniques were employed. This augmentation process enhanced the model's ability to generalize to unseen variations in fabric defects. The images were manually annotated with defect type and bounding box information. The image annotation contains a class label, bounding box center (x,y), height, and width. For this research, datasets of the defects were collected from various garment manufacturers with their consent. However, Data collection is an ongoing process. Researchers interested in the dataset can contact us for access.本研究构建了定制化面料缺陷数据集，所用图像均采集自存在各类瑕疵的成衣制品。该数据集共包含2468张高质量图像，所有图像均经过质量核验以保障整体品质。数据集涵盖六大类面料缺陷图像，分别为纽扣突起（Button-hike，307张）、纽扣破损（Broken-button，545张）、破洞（Hole，556张）、颜色瑕疵（Color-defect，384张）、异物纱线（Foreign-yarn，334张）与缝纫误差（Sewing-error，317张）。为缓解数据集规模有限可能带来的性能局限，本研究采用了数据增强技术，该流程可提升模型对未见面料缺陷变体的泛化能力。所有图像均由人工标注缺陷类别与边界框（bounding box）信息，标注内容包含类别标签、边界框中心点坐标(x,y)、高度与宽度。本研究的缺陷数据集已征得多家成衣制造商同意后完成采集，目前数据采集工作仍在持续推进中。对该数据集有使用需求的研究人员可联系我方获取访问权限。