Surface Defect Saliency of Magnetic Tile
Vision-based detection on surface defects has long postulated in the magnetic tile automation process. In this work, we introduce a real-time and multi-module neural network model called MCuePush U-Net, specifically designed for the image saliency detection of magnetic tile. We show that the model exceeds the state-of-the-art, in which it both effectively and explicitly maps multiple surface defects from low-contrast images. Our model significantly reduces time cost of machinery from 0.5s per image to 0.07s, and enhances saliency accuracy on surface defect detection.
PDFCode
Datasets
Results from the Paper
Ranked #1 on Anomaly Detection on Surface Defect Saliency of Magnetic Tile (Segmentation AUROC metric, using extra training data)
Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Uses Extra Training Data |
Benchmark |
---|---|---|---|---|---|---|---|
Anomaly Detection | Surface Defect Saliency of Magnetic Tile | MCuePush (supervised) | Segmentation AUROC | 98.5 | # 1 |