Anomaly.Paddle

简介

基于 Paddle 的异常检测方法复现

Logs&Tricks

参考repo:

• PaDiM-Anomaly-Detection-Localization-master
• PaDiM-Paddle
• anomalib

复现论文 |Including paper:

• PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization
• Semi-orthogonal Embedding for Efficient Unsupervised Anomaly Segmentation
• PatchCore: Towards Total Recall in Industrial Anomaly Detection
• Modeling the Distribution of Normal Data in Pre-Trained Deep Features for Anomaly Detection and so on...

数据集|Datasets:

• MVTec AD
• KolektorSDD
• mSTC

评估指标|Metrics:

• Image AUROC
• Pixel AUROC
• PRO score (Per Region Overlap Score)

快速开始

第一步：克隆本项目

# clone this repo
git clone git@github.com/ultranity/Anomaly.Paddle.git
cd Anomaly.Paddle

第二步：安装依赖

pip install -r requirements.txt

后续使用：训练/验证/导出模型，TIPC测试等

见各算法单独文档

• PaDiM
• OrthoAD
• PatchCore

代码结构

Anomaly.Paddle
├── datasets #数据集定义
├── logs #复现日志
├── output #输出目录
├── test_tpic # TPIC自动化测试
├── eval.py # 模型验证
├── export_model.py #模型导出
├── infer.py # 模型推理脚本
├── net.py # resnet网络结构补充（wide_resnet50_2）
├── OrthoAD.md # OrthoAD说明文件
├── PaDiM.md # PaDiM说明文件
├── predict.py # 模型单独预测
├── README.md # 主说明文件
├── train.py # 模型训练
├── utils.py # 工具函数
└── requirements.txt #库依赖

说明

在此非常感谢CuberrChen贡献的PaDiM-Paddle项目，提高了本repo复现论文的效率。

感谢百度 AIStudio 提供的算力支持

LICENSE

本项目的发布受Apache 2.0 license许可认证。

参考链接与文献

• Simonjan, Jennifer and Unluturk, Bige D. and Akyildiz, Ian F. PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization
• Jin-Hwa Kim, Do-Hyeong Kim, Saehoon Yi, Taehoon Lee. Semi-orthogonal Embedding for Efficient Unsupervised Anomaly Segmentation
• Modeling the Distribution of Normal Data in Pre-Trained Deep Features for Anomaly Detection
• anomalib
• PaDiM-Anomaly-Detection-Localization-master
• Semi-Orthogonal Embedding for Efficient Unsupervised Anomaly Segmentation(github.com)
• PaDiM-Paddle
• Knowledge_Distillation_AD_Paddle
• DFR