SubspaceAD：基于子空间建模的无训练少样本异常检测
SubspaceAD: Training-Free Few-Shot Anomaly Detection via Subspace Modeling
用冻结的 DINOv2 patch 特征拟合 PCA 子空间、以重建残差作为异常分数，为少样本/单样本工业异常检测提供了一个无需训练、无需 memory bank 的强基线，可直接迁移到其他需要无监督打分的一类分类场景。
原文依据 a Principal Component Analysis (PCA) model is fit to these features to estimate the low-dimensional subspace of normal variations
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论在工业检测中检测视觉异常，通常每个类别只能使用少量正常图像进行训练。近期的少样本方法借助基础模型特征取得了很好的效果，但通常依赖记忆库、辅助数据集，或对视觉语言模型进行多模态微调。因此我们提出疑问：在视觉基础模型的特征表示之下，这样的复杂性是否必要。为回答这一问题，我们提出 SubspaceAD，一种无训练方法，仅包含两个简单阶段。首先，由冻结的 DINOv2 主干网络从少量正常图像中提取块级特征。其次，对这些特征拟合主成分分析（PCA）模型，以估计正常变化的低维子空间。在推理时，通过相对于该子空间的重建残差检测异常，从而产生可解释且具有统计依据的异常分数。尽管方法简单，SubspaceAD 在无需训练、无需提示词微调、无需记忆库的情况下，在单样本与少样本设定下均取得了最优性能。在单样本异常检测设定下，SubspaceAD 在 MVTec-AD 数据集上取得 98.0% 的图像级与 97.6% 的像素级 AUROC，在 VisA 数据集上分别取得 93.3% 与 98.3%，超过了此前的最优结果。
英文原文摘要 1351 字符
Detecting visual anomalies in industrial inspection often requires training with only a few normal images per category. Recent few-shot methods achieve strong results employing foundation-model features, but typically rely on memory banks, auxiliary datasets, or multi-modal tuning of vision-language models. We therefore question whether such complexity is necessary given the feature representations of vision foundation models. To answer this question, we introduce SubspaceAD, a training-free method, that operates in two simple stages. First, patch-level features are extracted from a small set of normal images by a frozen DINOv2 backbone. Second, a Principal Component Analysis (PCA) model is fit to these features to estimate the low-dimensional subspace of normal variations. At inference, anomalies are detected via the reconstruction residual with respect to this subspace, producing interpretable and statistically grounded anomaly scores. Despite its simplicity, SubspaceAD achieves state-of-the-art performance across one-shot and few-shot settings without training, prompt tuning, or memory banks. In the one-shot anomaly detection setting, SubspaceAD achieves image-level and pixel-level AUROC of 98.0% and 97.6% on the MVTec-AD dataset, and 93.3% and 98.3% on the VisA dataset, respectively, surpassing prior state-of-the-art results.