| 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. | |