MMIO: 80K-Sample Industrial Defect Dataset for Zero-Shot LVLM Benchmarking

MMIO is an 80K+ sample multi-modal industrial dataset covering 6 super-categories and 18 subcategories, paired with a Refined Text-Visual Prompt method for zero-shot defect detection without user-provided segmentation prompts. It's the first large-scale pre-training dataset targeting open-mode industrial zero-shot learning.