One-for-All: Proposal Masked Cross-Class Anomaly Detection
Authors: Xincheng Yao, Chongyang Zhang, Ruoqi Li, Jun Sun, Zhenyu Liu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that the proposed PMAD can outperform current state-of-the-art models significantly under the multiand especially cross-class settings. |
| Researcher Affiliation | Collaboration | Xincheng Yao1, Chongyang Zhang1,2*, Ruoqi Li1, Jun Sun1, Zhenyu Liu3 1School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China 2Mo E Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University, Shanghai 200240, China 3Ningbo HTVision Digital Technology Co.,Ltd, Ningbo 315000, China {i-Dover, sunny zhang, nilponi, junsun}@sjtu.edu.cn1, lzy409911075@163.com3 |
| Pseudocode | Yes | The procedure of blockwise masking is summarized in Algorithm 1 in Appendix. |
| Open Source Code | Yes | Code will be publicly available at https://github.com/xcyao00/PMAD. |
| Open Datasets | Yes | We evaluate the proposed approach on two widely used industrial anomaly detection datasets: the MVTec AD (Bergmann et al. 2019a) and BTAD (Mishra et al. 2021). |
| Dataset Splits | No | The paper describes training and testing sets (e.g., '3629 images for training and 1725 images for testing' for MVTec AD), but does not explicitly mention a separate validation set or its split details. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory specifications, or cloud computing resources used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers (e.g., 'PyTorch 1.9', 'Python 3.8') needed to replicate the experiment. |
| Experiment Setup | Yes | We mainly follow the hyperparameters in (Bao, Dong, and Wei. 2021) to train the reconstruction model. All training hyperparameters are listed in Appendix. Because the sizes of anomalies in different classes are generally different, the same mask ratio for all classes cannot achieve the optimal results. We select a suitable mask ratio for each class through extensive experiments (see Appendix for details). |