FabriD:Fabric Defect Detection Dataset

Published: 1 November 2026| Version 3 | DOI: 10.17632/vncdwps24r.3
Contributors:
,
,
,

Description

FabriD is a fabric defect detection and segmentation dataset comprising real-world fabric images designed to support research in computer vision, image classification, object detection, instance segmentation, and intelligent textile recognition. The dataset contains five image categories, including four fabric defect classes and one defect-free class: - Needle Mark: 467 images - Hole: 459 images - Stains: 407 images - Missing Yarn: 454 images - Defect-free: 620 images The dataset contains a total of 2,407 images. These images represent diverse visual conditions, including variations in illumination, viewing angles, fabric textures, wrinkles, folds, and background complexity. Such diversity makes the dataset suitable for developing and evaluating deep learning models for automated fabric inspection and defect analysis. The images were annotated manually using the Roboflow annotation platform to support image classification, object detection, and instance segmentation tasks. Depending on the annotation task, the dataset includes class labels, polygon annotations, and segmentation masks for defect localization and delineation. The annotations were manually reviewed to improve label consistency and reduce annotation errors. FabriD provides a resource for researchers and practitioners working on automated textile inspection and fabric defect analysis. Potential applications include quality control, smart manufacturing, textile production monitoring, and the development of intelligent fabric inspection systems. The dataset can also support comparative evaluations of computer vision and deep learning approaches for fabric defect detection and segmentation.

Files

Institutions

Categories

Computer Science, Artificial Intelligence, Machine Learning, Cotton Fabrics

Licence