Unsupervised Anomaly Detection via Masked Diffusion Posterior Sampling

Authors: Di Wu, Shicai Fan, Xue Zhou, Li Yu, Yuzhong Deng, Jianxiao Zou, Baihong Lin

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Exhaustive experiments on MVTec and BTAD datasets demonstrate that MDPS can achieve state-of-the-art performance in normal image reconstruction quality as well as anomaly detection and localization. 5 Experiments In this section, we compare our MDPS with other UAD methods, and conduct ablation studies to validate the designs.
Researcher Affiliation Academia Di Wu1 , Shicai Fan1,2 , Xue Zhou1,2 , Li Yu1,2 , Yuzhong Deng1 , Jianxiao Zou1,2 , Baihong Lin2 1School of Automation Engineering, University of Electronic Science and Technology of China(UESTC) 2Shenzhen Institute for Advanced Study, UESTC
Pseudocode Yes Algorithm 1 Masked Diffusion Posterior Sampling
Open Source Code No Source code will be available at https://github.com/Kevin BHLin/.
Open Datasets Yes We conduct all experiments on the MVTec and BTAD Datasets. The MVTec dataset is an industrial AD benchmark [Bergmann et al., 2021]... The BTAD dataset contains approximately 2500 real-world industrial images of three products [Mishra et al., 2021]...
Dataset Splits No The paper describes the number of normal and anomaly samples in the MVTec dataset and states that models learn from normal samples, but it does not specify explicit training, validation, and test dataset splits or percentages.
Hardware Specification Yes For each category of normal samples in MVTec/BTAD, we train a UNet denoiser ϵθ(xt, t) separately within 2000 epochs using an Adam optimizer (learing rate: 1e-4, weight decay: 5e-2) based on a single Ge Force RTX 3090 GPU.
Software Dependencies No The paper mentions using a 'U-net architecture' and 'DDIM' but does not provide specific version numbers for these or other software libraries/dependencies.
Experiment Setup Yes For each category of normal samples in MVTec/BTAD, we train a UNet denoiser ϵθ(xt, t) separately within 2000 epochs using an Adam optimizer (learing rate: 1e-4, weight decay: 5e-2)... In the training process, the batchsize is set to 8, and the timestep of DDIM is set to be 1000. After training, we utilize the trained denoiser for the proposed MDPS, and let T = 200, N = 10, ρ = 100. In our paper, S is set to 500.