README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path: train/**
      - split: test
        path: test/**

Defect VLM Dataset

This dataset contains paired structural defect images and English natural-language descriptions.

Fields:

• image: the defect image
• text: the generated English defect description

Splits:

• train: 884 images
• test: 211 images

The descriptions were generated from defect category and extent annotations while preserving the original train/test split.