ZD001
Description
A fabric defect dataset was independently collected and constructed, with all original images sourced from the fabric inspection workshops of textile enterprises. A flaw imaging control system, based on computer vision, was developed in conjunction with industrial field equipment. The system integrates key technologies such as multi-light source fusion, high-precision visual sensing modules, edge computing platforms, and visual interaction. It is capable of multi-angle, high-quality, and high-resolution defect image acquisition and digital storage. For clarity, the dataset created is named ZD001_FD, which was used for training the IFD-GAN algorithm and validating its industrial application. The ZD001_FD dataset includes eight categories of cross-scale fabric defects: warp-direction defects, such as broken end, misdraw, and reediness; weft-direction defects, including thin place and thick place; and localized defects, such as hole, cotton ball, and kinky filling. The conventional approach to dividing test sets has limitations, as it often involves simplistic scenarios that fail to capture the complexity and diversity of real-world industrial applications, which are influenced by numerous interfering factors . To address these shortcomings, test sets were collected and created under various industrial scenarios to replicate edge-testing conditions, thereby assessing the model's adaptability and generalization capabilities. Due to the flexible and high-fidelity cross-scale defect generation capability of IFD-GAN, the expansion and inter-class balancing of the ZD001_FD dataset were successfully achieved, resulting in enhanced diversity. The augmented dataset is named ZD001_FD_IG.