GLAD: Towards Better Reconstruction with Global and Local Adaptive Diffusion Models for Unsupervised Anomaly Detection (2406.07487v3)

Abstract: Diffusion models have shown superior performance on unsupervised anomaly detection tasks. Since trained with normal data only, diffusion models tend to reconstruct normal counterparts of test images with certain noises added. However, these methods treat all potential anomalies equally, which may cause two main problems. From the global perspective, the difficulty of reconstructing images with different anomalies is uneven. Therefore, instead of utilizing the same setting for all samples, we propose to predict a particular denoising step for each sample by evaluating the difference between image contents and the priors extracted from diffusion models. From the local perspective, reconstructing abnormal regions differs from normal areas even in the same image. Theoretically, the diffusion model predicts a noise for each step, typically following a standard Gaussian distribution. However, due to the difference between the anomaly and its potential normal counterpart, the predicted noise in abnormal regions will inevitably deviate from the standard Gaussian distribution. To this end, we propose introducing synthetic abnormal samples in training to encourage the diffusion models to break through the limitation of standard Gaussian distribution, and a spatial-adaptive feature fusion scheme is utilized during inference. With the above modifications, we propose a global and local adaptive diffusion model (abbreviated to GLAD) for unsupervised anomaly detection, which introduces appealing flexibility and achieves anomaly-free reconstruction while retaining as much normal information as possible. Extensive experiments are conducted on three commonly used anomaly detection datasets (MVTec-AD, MPDD, and VisA) and a printed circuit board dataset (PCB-Bank) we integrated, showing the effectiveness of the proposed method.

The paper "GLAD: Towards Better Reconstruction with Global and Local Adaptive Diffusion Models for Unsupervised Anomaly Detection" explores advancements in using diffusion models for unsupervised anomaly detection. The primary goal is to improve how these models reconstruct images by introducing global and local adaptation mechanisms.

Key Contributions:

1. Challenge with Uniform Treatment of Anomalies: Diffusion models traditionally treat all anomalies equally during image reconstruction, which can lead to uneven reconstruction performance across different anomaly types. This uniform approach fails to consider the complexity and variability of anomalies in the data.
2. Global Perspective Adaptation: The authors propose customizing the denoising steps for each sample. By predicting a unique denoising step, the model evaluates the discrepancy between the image content and the priors from the diffusion model, allowing for more tailored and accurate reconstruction.
3. Local Perspective Adaptation: Within a single image, reconstructing abnormal regions differs from normal regions. The paper suggests that standard Gaussian noise predictions deviate in abnormal areas. To address this, synthetic abnormal samples are introduced during training, encouraging the model to overcome the limitations of standard Gaussian distributions.
4. Spatial-Adaptive Feature Fusion: During inference, a spatially adaptive feature fusion scheme is used to enhance performance further, enabling more precise reconstruction by focusing on the discrepancies between anomaly and normal regions within images.

Experimental Validation:

The model, named Global and Local Adaptive Diffusion (GLAD), is extensively tested on several datasets commonly used in anomaly detection:

• MVTec-AD
• MPDD
• VisA
• PCB-Bank: A printed circuit board dataset integrated by the authors.

Results from these experiments demonstrate the effectiveness of the GLAD model in achieving anomaly-free reconstructions while effectively retaining normal image information. This indicates that the proposed global and local adaptations are beneficial in handling diverse and complex anomalies in unsupervised settings.

Overall, the paper introduces a novel approach to enhancing the performance of diffusion models in anomaly detection by tailoring the reconstruction process both at a global and local level. This method provides increased flexibility and promises better handling of anomalies in varied datasets.