Search papers, labs, and topics across Lattice.
The paper introduces SubspaceAD, a training-free anomaly detection method for few-shot industrial inspection that leverages frozen DINOv2 features. It addresses the question of whether complex training procedures are necessary for anomaly detection given the strong feature representations of vision foundation models. SubspaceAD achieves state-of-the-art performance by fitting a PCA model to patch-level DINOv2 features extracted from a few normal images and detecting anomalies based on reconstruction residuals, outperforming existing methods on MVTec-AD and VisA datasets.
Forget finetuning: this training-free anomaly detection method using DINOv2 and PCA achieves state-of-the-art results in few-shot settings.
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. Code and demo are available at https://github.com/CLendering/SubspaceAD.
