Stats Components#

Statistical functions.

class anomalib.models.components.stats.GaussianKDE(dataset=None)#

Bases: DynamicBufferMixin

Gaussian Kernel Density Estimation.

Parameters:: dataset (Tensor | None, optional) – Dataset on which to fit the KDE model. Defaults to None.

static cov(tensor)#

Calculate the unbiased covariance matrix.

Parameters:: tensor (torch.Tensor) – Input tensor from which covariance matrix is computed.
Return type:: Tensor
Returns:: Output covariance matrix.

fit(dataset)#

Fit a KDE model to the input dataset.

forward(features)#

Get the KDE estimates from the feature map.

Parameters:: features (torch.Tensor) – Feature map extracted from the CNN
Return type:: Tensor

Returns: KDE Estimates

class anomalib.models.components.stats.MultiVariateGaussian#

Bases: DynamicBufferMixin, Module

Multi Variate Gaussian Distribution.

fit(embedding)#

Fit multi-variate gaussian distribution to the input embedding.

Parameters:: embedding (torch.Tensor) – Embedding vector extracted from CNN.
Return type:: list[Tensor]
Returns:: Mean and the covariance of the embedding.

forward(embedding)#

Calculate multivariate Gaussian distribution.

Parameters:: embedding (torch.Tensor) – CNN features whose dimensionality is reduced via either random sampling or PCA.
Return type:: list[Tensor]
Returns:: mean and inverse covariance of the multi-variate gaussian distribution that fits the features.

Stats Components