Summary of Anomalysd: Few-shot Multi-class Anomaly Detection with Stable Diffusion Model, by Zhenyu Yan et al.
AnomalySD: Few-Shot Multi-Class Anomaly Detection with Stable Diffusion Model
by Zhenyu Yan, Qingqing Fang, Wenxi Lv, Qinliang Su
First submitted to arxiv on: 4 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a few-shot, multi-class anomaly detection framework called AnomalySD, which leverages the Stable Diffusion (SD) model for zero/few-shot inpainting to identify defective parts in industrial manufacturing. The SD model is fine-tuned using hierarchical text descriptions and foreground mask mechanisms to adapt it to the anomaly detection task. In the inference stage, a multi-scale mask strategy and prototype-guided mask strategy are used to accurately mask anomalous regions for inpainting. The anomaly score is estimated based on the inpainting result of all masks. The paper reports extensive experiments on the MVTec-AD and VisA datasets, achieving superior results in terms of AUROC (93.6%/94.8% on MVTec-AD and 86.1%/96.5% on VisA) under multi-class and one-shot settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps solve a big problem in manufacturing – finding defective parts. Right now, most methods need lots of normal data to work well. This can be expensive or hard to get because it’s private information. Another issue is that these methods usually require special models for each new object they want to detect, which gets very costly and hard to do. The paper proposes a new way using a Stable Diffusion model to help find anomalies without needing lots of normal data. They also come up with clever ways to fine-tune the model and make it better at detecting anomalies. The results show that this approach works really well on two big datasets, helping us identify defective parts more accurately. |
Keywords
» Artificial intelligence » Anomaly detection » Diffusion » Diffusion model » Few shot » Inference » Mask » One shot
