Fig 1.
Illustration of complex challenges in real-world industrial inspection.
(a) Four images demonstrating variations in lighting conditions due to different equipment deployment environments and complex backgrounds like stained backdrops. (b) Examples of various types of damage and tearing defects. (c) Two images showing unstable glove poses during capture, caused by the shaking of glove molds.
Table 1.
The format of SFT and RFT dataset.
Table 2.
Comparison of mAP and Precision across different models.
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.
Table 3.
Comparison of mAP and Precision across different models.