FabriD:Fabric Defect Detection Dataset
Description
This dataset presents a comprehensive collection of real-world fabric images developed to support research in computer vision, image classification, fabric defect detection & segmentation/instance segmentation, and intelligent textile recognition. The dataset consists of four commonly defective and one defect-free image category: Needle Mark (467) Hole (459) Stains (407) Missing Yarn (454) Defect-free (624). The overall total image count in this dataset is 2407. Images were captured under diverse real-world conditions, including variations in illumination, viewing angles, texture patterns, wrinkles, folds, and background complexity, making the dataset suitable for developing robust deep learning models. All images were manually annotated using the Roboflow annotation platform specifically for image classification, object detection, and instance segmentation. Bounding boxes, polygons, and segmentation. Bounding boxes, polygons, and segmentation masks were created for precise defect localization and boundary segmentation. Instead, each image was assigned a single class label corresponding to the dominant fabric type. The annotation process included careful manual verification to ensure label consistency and minimise annotation errors. This dataset provides a valuable benchmark for the textile industry and the research community by enabling the development of automated fabric recognition systems for smart manufacturing, quality inspection, inventory management, e-commerce product categorisation, and intelligent textile applications.
Files
Institutions
- University of Information Technology and SciencesDhaka Division, Dhaka