PatchCore anomaly detection

Unofficial implementation of PatchCore(new SOTA) anomaly detection model

Original Paper : Towards Total Recall in Industrial Anomaly Detection (Jun 2021)
Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Schölkopf, Thomas Brox, Peter Gehler

https://arxiv.org/abs/2106.08265
https://paperswithcode.com/sota/anomaly-detection-on-mvtec-ad

notice(21/06/18) :
This code is not yet verified. Any feedback is appreciated.
updates(21/06/21) :

I used sklearn's SparseRandomProjection(ep=0.9) for random projection. I'm not confident with this.
I think exact value of "b nearest patch-features" is not presented in the paper. I just set 9. (args.n_neighbors)
In terms of NN search, author used "faiss". but not implemented in this code yet.
sample embeddings/carpet/embedding.pickle => coreset_sampling_ratio=0.001

updates(21/06/26) :

A critical issue related to "locally aware patch" raised and fixed. Score table is updated.

Usage

# install python 3.6, torch==1.8.1, torchvision==0.9.1
pip install -r requirements.txt

python train.py --phase train or test --dataset_path .../mvtec_anomaly_detection --category carpet --project_root_path path/to/save/results --coreset_sampling_ratio 0.01 --n_neighbors 9'

# for fast try just specify your dataset_path and run
python train.py --phase test --dataset_path .../mvtec_anomaly_detection --project_root_path ./

MVTecAD AUROC score (PatchCore-1%, mean of n trials)

Category	Paper (image-level)	This code (image-level)	Paper (pixel-level)	This code (pixel-level)
carpet	0.980	0.991(1)	0.989	0.989(1)
grid	0.986	0.975(1)	0.986	0.975(1)
leather	1.000	1.000(1)	0.993	0.991(1)
tile	0.994	0.994(1)	0.961	0.949(1)
wood	0.992	0.989(1)	0.951	0.936(1)
bottle	1.000	1.000(1)	0.985	0.981(1)
cable	0.993	0.995(1)	0.982	0.983(1)
capsule	0.980	0.976(1)	0.988	0.989(1)
hazelnut	1.000	1.000(1)	0.986	0.985(1)
metal nut	0.997	0.999(1)	0.984	0.984(1)
pill	0.970	0.959(1)	0.971	0.977(1)
screw	0.964	0.949(1)	0.992	0.977(1)
toothbrush	1.000	1.000(1)	0.985	0.986(1)
transistor	0.999	1.000(1)	0.949	0.972(1)
zipper	0.992	0.995(1)	0.988	0.984(1)
mean	0.990	0.988	0.980	0.977

Code Reference

kcenter algorithm :
https://github.com/google/active-learning
embedding concat function :
https://github.com/xiahaifeng1995/PaDiM-Anomaly-Detection-Localization-master

Unofficial implementation of PatchCore anomaly detection

Related tags

Overview

PatchCore anomaly detection

Usage

MVTecAD AUROC score (PatchCore-1%, mean of n trials)

Code Reference

Owner

Changwoo Ha

Implement of "Training deep neural networks via direct loss minimization" in PyTorch for 0-1 loss

Laplace Redux -- Effortless Bayesian Deep Learning

Official code for "Mean Shift for Self-Supervised Learning"

Tensorflow Repo for "DeepGCNs: Can GCNs Go as Deep as CNNs?"

Code for NAACL 2021 full paper "Efficient Attentions for Long Document Summarization"

Another pytorch implementation of FCN (Fully Convolutional Networks)

LQM - Improving Object Detection by Estimating Bounding Box Quality Accurately

Bare bones use-case for deploying a containerized web app (built in streamlit) on AWS.

Self-supervised Augmentation Consistency for Adapting Semantic Segmentation (CVPR 2021)

PyTorch implementation of HDN(Homography Decomposition Networks) for planar object tracking

Multitask Learning Strengthens Adversarial Robustness

this is a lite easy to use virtual keyboard project for anyone to use

Realtime Face Anti Spoofing with Face Detector based on Deep Learning using Tensorflow/Keras and OpenCV

PyTorch implementation of paper: HPNet: Deep Primitive Segmentation Using Hybrid Representations.

All supplementary material used by me while TA-ing CS3244: Machine Learning

Code for project: "Learning to Minimize Remainder in Supervised Learning".

S2-BNN: Bridging the Gap Between Self-Supervised Real and 1-bit Neural Networks via Guided Distribution Calibration (CVPR 2021)

Code for "Infinitely Deep Bayesian Neural Networks with Stochastic Differential Equations"

DC540 hacking challenge 0x00005a.

中文语音识别系列，读者可以借助它快速训练属于自己的中文语音识别模型，或直接使用预训练模型测试效果。