Figure 12
From: Balancing complexity and accuracy for defect detection on filters with an improved RT-DETR
Confusion matrices demonstrating model generalization on the public NEU-DET dataset, which contains six defect categories. (a) The baseline RT-DETR model struggles with certain classes like Crazing. (b) The proposed RT-DETR-LGA model shows substantially improved accuracy and reduced confusion between classes, confirming its robust generalization capabilities.
