Code[Github] abstract We introduce IVD-SEG, a dataset encompassing defect images from 43 different industrial products, totaling 5686 images, spanning tasks that include binary and multiclass segmentation. Within IVD-SEG, we meticulously propose 12 sub-datasets, including two newly developed datasets by our team. Serving as a large-scale standardized industrial defect image dataset, all images are unified to a 256 × 256 size, accompanied by semantic segmentation annotations. This standardization facilitates users unfamiliar with industrial product defects to utilize the dataset, allowing them to focus on exploring algorithmic performance on the IVD-SEG dataset. To our knowledge, the dataset we propose is currently the most comprehensive and voluminous compilation of industrial defect images, thereby contributing to the advancement of relevant research in industrial image analysis. Simultaneously, our dataset supports research and education in various fields, including computer vision and machine learning. We conducted benchmark tests on IVD-SEG using several baseline methods, including representative CNN and ViT networks.