핵심 개념
RealNet introduces innovative anomaly detection methods with synthetic anomaly generation, feature selection, and reconstruction residuals.
초록
Self-supervised feature reconstruction methods face challenges in synthesizing realistic anomalies.
RealNet introduces Strength-controllable Diffusion Anomaly Synthesis (SDAS) for diverse anomaly samples.
Anomaly-aware Features Selection (AFS) and Reconstruction Residuals Selection (RRS) enhance anomaly detection.
RealNet outperforms state-of-the-art methods on benchmark datasets.
Synthetic Industrial Anomaly Dataset (SIA) facilitates anomaly synthesis.
통계
"RealNet achieves state-of-the-art performance while addressing computational cost limitations suffered by previous methods."
"RealNet achieves an Image AUROC of 99.65% and a Pixel AUROC of 99.03% on the MVTec-AD dataset."
"RealNet achieves an Image AUROC of 96.3% on the MPDD dataset, surpassing the current best performance by 10.2%."
인용구
"RealNet fully exploits the discriminative capabilities of large-scale pre-trained CNNs while reducing feature redundancy and pre-training bias."
"Our results demonstrate significant improvements in both Image AU-ROC and Pixel AUROC compared to the current state-of-the-art methods."