CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection

Authors: Xiaolei Wang, Xiaoyang Wang, Huihui Bai, Eng Gee Lim, Jimin Xiao

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our method achieves competitive performance on the MVTec AD and Vis A datasets, demonstrating its effectiveness. We conduct comprehensive experiments to demonstrate the efficacy of our approach, yielding a notable performance gain over singlemodal methods.
Researcher Affiliation Collaboration 1Xi an Jiaotong-Liverpool University 2University of Liverpool 3Dinnar Automation Technology 4Beijing Jiaotong University EMAIL, EMAIL, EMAIL
Pseudocode No The paper includes a figure (Figure 2) illustrating the framework, but it is a high-level diagram and not a structured pseudocode or algorithm block.
Open Source Code Yes Code https://github.com/cvddl/CNC
Open Datasets Yes Datasets MVTec AD (Bergmann et al. 2019) is the most widely used industrial anomaly detection dataset... Vis A (Zou et al. 2022) is a challenging AD dataset...
Dataset Splits Yes The training set consists of 3629 images with anomaly-free samples. The test dataset includes 1725 normal and abnormal images.
Hardware Specification Yes All experiments are conducted on a single NVIDIA Tesla V100 32GB GPU.
Software Dependencies No The implementation is based on Pytorch. The publicly available CLIP model (VITL/14@336px) is the backbone of our method. We select the Adam optimizer to train our model.
Experiment Setup Yes For both datasets, we set temperature coefficient τ = 0.001 and batch size to 8 with learning rate 0.001 to train the whole model. The number of experts is set to 5 with top K = 2 gated scores in the Mo E. Next, we set the epoch to 250 and 200 for MVTec AD and Vis A with the same ϑ = 5, respectively.