A unified vision-language model for cross-product defect detection in glove manufacturing
Fig 2
This figure demonstrates the model’s inference results on three distinct samples: a normal blue nitrile glove, a torn blue nitrile glove, and a white PVC glove with an oil stain.
The first case highlights the model’s robustness; its ability to correctly identify the rolled cuff of the normal glove, a feature often misclassified as a tear by other models, shows superior handling of gloves with special morphologies.