AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model

Authors: Teng Hu, Jiangning Zhang, Ran Yi, Yuzhen Du, Xu Chen, Liang Liu, Yabiao Wang, Chengjie Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our model significantly outperforms the state-of-the-art methods in generation authenticity and diversity, and effectively improves the performance of downstream anomaly inspection tasks.
Researcher Affiliation Collaboration Teng Hu1*, Jiangning Zhang2*, Ran Yi1 , Yuzhen Du1, Xu Chen2, Liang Liu2, Yabiao Wang2, Chengjie Wang1,2 1Shanghai Jiao Tong University 2Youtu Lab, Tencent {hu-teng, ranyi, Haaaaaaaaaa}@sjtu.edu.cn; {vtzhang, cxxuchen, leoneliu, caseywang, jasoncjwang}@tencent.com;
Pseudocode No No section or figure explicitly labeled 'Pseudocode' or 'Algorithm' was found.
Open Source Code Yes The code and data are available in https://github.com/sjtuplayer/anomalydiffusion.
Open Datasets Yes Dataset. we conduct experiments on the widely used MVTec (Bergmann et al. 2019) dataset.
Dataset Splits No The paper states, 'We employ one-third of the anomaly data with the lowest ID numbers as the training set, reserving the remaining two-thirds for testing.' A specific validation split is not explicitly mentioned.
Hardware Specification No No specific hardware details (like GPU models, CPU types, or cloud instance specifications) used for running experiments are provided in the paper.
Software Dependencies No The paper mentions software components like 'Latent Diffusion Model (LDM)', 'U-Net', 'Res Net-50', and 'Feature Pyramid Networks (FPN)' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes Implementation details. We assign k = 8 tokens for anomaly embedding, n = 4 tokens for spatial embedding, and k = 4 tokens for mask embedding.