torchvision4ad 0.1.1

torchvision for anomaly detection

torchvision for Anomaly Detection

You can use the MVTec Anomaly Detection Dataset.

Installation

pip:

$ pip install torchvision4ad

From source:

$ python setup.py install

Usage

You can use one of the MVTec AD Dataset names {'bottle', 'cable', 'capsule', 'carpet', 'grid', 'hazelnut', 'leather', 'metal_nut', 'pill', 'screw', 'tile', 'toothbrush', 'transistor', 'wood', 'zipper'}.

from torchvision4ad.datasets import MVTecAD


root = 'mvtec_ad'
dataset_name = 'bottle'
mvtec_ad = MVTecAD(root, dataset_name, train=True, download=True)
for (img, target) in mvtec_ad:
    ...

Of course, you can also give a function/transform takes in an PIL image and returns a transformed version.

import torchvision.transforms as transforms

from torchvision4ad.datasets import MVTecAD


transform = transforms.Compose([transforms.Resize([64, 64]),
                                transforms.ToTensor()])
mvtec_ad = MVTecAD('mvtec_ad', 'bottle', train=True, transform=transform, download=True)