https://github.com/KLIVIS/IVD-SEG/blob/main/README.md
We introduce IVD-SEG, a large-scale standardized industrial defect image dataset similar to MNIST. The dataset includes 12 subsets of images that are all unified to a size of 256×256 with semantic segmentation annotations for users to use when unfamiliar with industrial product defects. The IVD-SEG dataset covers defect images of 43 different products in industrial scenes, including fabrics, roads, metals, and magnetic tiles, among others. There are a total of 5686 images, covering both binary and multi-class segmentation tasks. In the absence of easy access to industrial defect images, we propose the first large-scale industrial defect dataset containing multiple product categories, which helps advance research on industrial defect detection and segmentation technology. Additionally, as image segmentation is a universal task, our dataset supports research and education in multiple fields, such as computer vision and machine learning. We benchmark several baseline methods on IVD-SEG, including representative CNN and VIT networks.
Aavilable at Zenodo