Elaine0 / Anomaly-Detection

NOTE: The work on this project is still in progress.

Anomaly Detection

This project aims at developping a Deep Learning model using an unsupervided method to detect surface anomalies on images.

Overview

The method being used in this project is inspired to a great extent by the papers MVTec AD — A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection and Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders. The method is devided in 3 steps: training, finetuning and testing.

Dataset

The dataset being used is the MVTec dataset.

Prerequisites

Dependencies

Libraries and packages used in this project:

• tensorflow-gpu 2.1.0
• Keras 2.3.1
• ktrain 0.13.0
• scikit-image 0.17.2
• opencv-python 4.2.0.34
• pandas 1.0.3
• numpy 1.18.1
• matplotlib 3.1.3

Download the Dataset

1. Download the mvtec dataset here and save it to a directory of your choice (e.g in /Downloads)
2. Extract the compressed files.
3. Create a folder named mvtec in the project directory.
4. Move the extracted files to the mvtec folder.

Directory Structure using mvtec dataset

For the scripts to work propoerly, it is required to have a specific directory structure. In the case of using the mvtec dataset, here is an example of how the directory stucture should look like:

├── bottle
│   ├── ground_truth
│   │   ├── broken_large
│   │   ├── broken_small
│   │   └── contamination
│   ├── test
│   │   ├── broken_large
│   │   ├── broken_small
│   │   ├── contamination
│   │   └── good
│   └── train
│       └── good
├── cable
│   ├── ground_truth
│   │   ├── bent_wire
│   │   ├── cable_swap
│   │   ├── combined
│   │   ├── cut_inner_insulation
│   │   ├── cut_outer_insulation
│   │   ├── missing_cable
│   │   ├── missing_wire
│   │   └── poke_insulation
│   ├── test
│   │   ├── bent_wire
│   │   ├── cable_swap
│   │   ├── combined
│   │   ├── cut_inner_insulation
│   │   ├── cut_outer_insulation
│   │   ├── good
│   │   ├── missing_cable
│   │   ├── missing_wire
│   │   └── poke_insulation
│   └── train
│       └── good
...

---

Directory Structure using your own dataset

To train with your own dataset, you need to have a comparable directory structure. For example:

├── class1
│   ├── test
│   │   ├── good
│   │   ├── defect
│   └── train
│       └── good
├── class2
│   ├── test
│   │   ├── good
│   │   ├── defect
│   └── train
│       └── good
...

---

Training (train.py)

Description

The method uses a Convolutional Auto-Encoder (CAE). There are two proposed variants:

• CAE proposed in the paper MVTec AD — A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection
• CAE that uses Keras's inception_resnet_v2 CNN-model pretrained on imagenet as the Encoder. The Decoder is inspired by the paper Anomaly Detection and Localization in Images using Guided Attention. NOTE: This model seems not to work properly at the moment and is still being tested.

During training, the CAE trains exclusively on defect-free images and learns to reconstruct (predict) defect-free training samples.

Usage

usage: train.py [-h] -d [-a] [-c] [-l] [-b] [-i]

optional arguments:

-h, --help show this help message and exit

-d , --input-dir directory containing training images

-a , --architecture architecture of the model to use for training: 'resnet', 'mvtec' or 'mvtec2'

-c , --color color mode for preprocessing images before training: 'rgb' or 'grayscale'

-l , --loss loss function to use for training: 'mssim', 'ssim' or 'l2'

-b , --batch batch size to use for training

-i, --inspect generate inspection plots after training

Example usage:

python3 train.py -d mvtec/capsule -a mvtec2 -b 8 -l ssim -c grayscale

NOTE 1: There is no need for the user to pass a number of epochs since the training process implements an Early Stopping strategy.

NOTE 2: There is a total of 3 models implemented in this project: resnet, mvtec and mvtec2. Resnet seems not to be working properly at the moment and needs further investigation/testing.

NOTE 3: While mvtec and mvtec2 are two slightly different variants of the same model, we recommend opting for mvtec2, as it has been tested extensively.

Finetuning (finetune.py)

This script approximates a good value for minimum area and threshold pair of parameters that should be used during testing to obtain good classification results. It relies on 10% of the defect-freee validation images and 20% of the defect and defect-free test images.

Usage

usage: finetune.py [-h] -p [-m] [-t]

optional arguments:

-h, --help show this help message and exit

-p , --path path to saved model

-m , --method method for generating resmaps: 'ssim' or 'l2'

-t , --dtype datatype for processing resmaps: 'float64' or 'uint8'

Example usage:

python3 finetune.py -p saved_models/mvtec/capsule/mvtec2/ssim/13-06-2020_15-35-10/CAE_mvtec2_b8_e39.hdf5 -m ssim -t float64

Testing (test.py)

This script classifies test images using the threshold and the minimum defect area that have been previously determined by finetuning.

Usage

usage: test.py [-h] -p

optional arguments:

-h, --help show this help message and exit

-p , --path path to saved model

Example usage:

python3 test.py -p saved_models/mvtec/capsule/mvtec2/ssim/13-06-2020_15-35-10/CAE_mvtec2_b8_e39.hdf5

Project Organization

├── mvtec                       <- folder containing all mvtec classes.
│   ├── bottle                  <- subfolder of a class (contains additional subfolders /train and /test).
|   |── ...
├── autoencoder                 <- directory containing modules for training: autoencoder class and methods as well as custom losses and metrics.
├── processing                  <- directory containing modules for preprocessing images and before training and processing images after training.
├── results                     <- directory containing finetuning and test results.
├── readme.md                   <- readme file.
├── requirements.txt            <- requirement text file containing used libraries.
├── saved_models                <- directory containing saved models, training history, loss and learning plots and inspection images.
├── train.py                    <- training script to train the auto-encoder.
├── finetune.py                 <- approximates a good value for minimum area and threshold for classification.
└── test.py                     <- test script to classify images of the test set using finetuned parameters.

---

Authors

• Adnene Boumessouer - (https://github.com/AdneneBoumessouer)

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Paul Bergmann, the main author of the paper which this project relies on.
• François Chollet, author of the Keras deep learning library.
• Aurélien Géron, autor of the great book Hands on Machine Learning with Scikit-Learn, Keras and Tensorflow.
• Arun S. Maiya, author of the ktrain library: a wrapper for TensorFlow Keras that makes deep learning and AI more accessible and easier to apply.
• Adrian Rosebrock, author of the website pyimagesearch.