"WELD"

"This dataset provides a collection of high-resolution weld surface images for industrial defect and anomaly detection. Each image is paired with a corresponding ground truth file that accurately delineates defective regions, enabling pixel-level evaluation. The dataset covers multiple typical weld defect types encountered in real manufacturing scenarios and reflects realistic variations in texture, illumination, and defect appearance.The dataset is designed to support research on industrial visual inspection, anomaly detection, and zero-shot or weakly supervised learning methods, where abnormal samples may be scarce or unavailable during training. By providing consistent image\u2013ground truth pairs, this dataset facilitates fair benchmarking for defect localization and detection tasks. It can serve as a valuable resource for both academic research and practical industrial applications."