MvtecAD unsupervised Anomaly Detection

Last update: Feb 25, 2022

Related tags

Overview

MvtecAD unsupervised Anomaly Detection

This respository is the unofficial implementations of DFR: Deep Feature Reconstruction for Unsupervised Anomaly Segmentation

Result of 500 epochs trained model

Selects latent sizes of Autoencoder by PCA

Classes	latent size	Segmentation AUC	Detection AUC
bottle	116	97.2771%	99.8413%
cable	557	95.5101%	84.8951%
capsule	162	98.8928%	97.3275%
carpet	245	97.9116%	90.5297%
grid	145	97.2484	79.5322%
hazelnut	459	98.5848%	100%
leather	325	98.8649%	95.4484%
metal_nut	380	96.127%	97.263%
pill	292	98.0543%	94.108%
screw	283	99.3001%	92.0066%
tile	557	89.4887%	91.7388%
toothbrush	243	98.6729%	91.3889%
transistor	333	83.9157%	89.0833%
wood	364	91.7027%	98.9474%
zipper	115	95.6663%	83.2983%
mean		95.8141%	92.3606

How to run

requirements

pytorch scikit-learn matplotlib numpy pandas PIL wget

Train

python main.py --mode          train      
               --data_dir_head [Datapath] 
               --BATCH_SIZE    [BATCH_SIZE] 
               --LR            [Learning Rate] 
               --EPOCH         [Epochs] 
               --backbone      [Feature map of Conv in VGG19]
               --latent_dim    [Latent size of CAE] 
               --classes       [Default is all]

Download 500 Epochs Finetuned Models

Here provide the model of each classes in Drophox

python main.py --mode download

Evaluate the ROC-AUC of Test Set

python main.py --mode        evaluation    
               --classes     [Default is all]

Inference the model

python main.py --mode           inference    
               --heatmap_path   [Input path] 
               --heatmap_item   [Class of input] 
               --heatmap_gt     [GT path Default is None]
               --device         [cpu or cuda]
               --device         [Output path ]

Example run in main.py

if __name__ == "__main__":  
    cfg = config()
    cfg.mode = "inference"
    cfg.heatmap_path = 'mvtecad_unsupervise/bottle/test/broken_small/001.png'
    cfg.heatmap_item = 'bottle'
    cfg.heatmap_gt = 'mvtecad_unsupervise/bottle/ground_truth/broken_small/001_mask.png'
    cfg.device = 'cpu'
    cfg.heatmap_export = 'validate/Inferece.png'

validate/Inferece.png is

Code Reference

https://github.com/YoungGod/DFR

https://www.kaggle.com/danieldelro/unsupervised-anomaly-segmentation-of-screw-images

MvtecAD unsupervised Anomaly Detection

Related tags

Overview

MvtecAD unsupervised Anomaly Detection

Result of 500 epochs trained model

Selects latent sizes of Autoencoder by PCA

How to run

requirements

Train

Download 500 Epochs Finetuned Models

Evaluate the ROC-AUC of Test Set

Inference the model

Code Reference

Owner

Industrial knn-based anomaly detection for images. Visit streamlit link to check out the demo.

A simple editor for captions in .SRT file extension

D2LV: A Data-Driven and Local-Verification Approach for Image Copy Detection

Zalo AI challenge 2021 task hum to song

Medical image analysis framework merging ANTsPy and deep learning

Team Enigma at ArgMining 2021 Shared Task: Leveraging Pretrained Language Models for Key Point Matching

ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators

Source code for paper "Deep Superpixel-based Network for Blind Image Quality Assessment"

PyTorchCV: A PyTorch-Based Framework for Deep Learning in Computer Vision.

People movement type classifier with YOLOv4 detection and SORT tracking.

Source code for our Paper "Learning in High-Dimensional Feature Spaces Using ANOVA-Based Matrix-Vector Multiplication"

E2EC: An End-to-End Contour-based Method for High-Quality High-Speed Instance Segmentation

FB-tCNN for SSVEP Recognition

Tensorflow implementation of the paper "HumanGPS: Geodesic PreServing Feature for Dense Human Correspondences", CVPR 2021.

Proposal, Tracking and Segmentation (PTS): A Cascaded Network for Video Object Segmentation

Robust, modular and efficient implementation of advanced Hamiltonian Monte Carlo algorithms

Revisiting Global Statistics Aggregation for Improving Image Restoration

A Pytorch reproduction of Range Loss, which is proposed in paper 《Range Loss for Deep Face Recognition with Long-Tailed Training Data》

Accurate identification of bacteriophages from metagenomic data using Transformer

Making self-supervised learning work on molecules by using their 3D geometry to pre-train GNNs. Implemented in DGL and Pytorch Geometric.