---
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.