Find research datasets worth reusing

Pixel-wise Gaussian AD visualization (ResNet18, MVTecAD)

<p>A multivariate Gaussian model is fitted with the pixel-wise feature vectors [1] (all pixel positions confounded) extracted from a pre-trained convolutional neural network using the maximum likelihood estimator (empirical mean and empirical covariance matrix). From the eigendecomposition of the empirical covariance matrix, a whitening transformation is applied to the feature vectors, which yields a new vector of the same number of &quot;components&quot; (cf. [2]).&nbsp;</p> <p>This dataset shows visualizations of the component maps (cf. [2]) of all the categories in the MVTec-AD dataset. All the layer blocks&nbsp;[3] from a ResNet18 [4] are&nbsp;used as feature map extractors for the model described above.&nbsp;Each combination of category and layer is trained separately.</p> <p>The visualizations are composed of heatmaps of the square of the component maps; their sum (the Mahalanobis distance of each pixel, yielding an anomaly score map), and (for images containing defects, a.k.a. &quot;anomalies&quot;) the ground truth annotation provided in the MVTec-AD dataset.&nbsp; &nbsp;</p> <p>For more details, refer to our paper.</p> <p>---</p> <p>[1] Pixel-wise feature vectors: given a stacked feature map of a 2D image (i.e. multiple 2D images, or &quot;channels,&quot; containing information about the 2D input image), extract the values of all channels for a single pixel position.</p> <p>[2] The axes of the whitened [feature] vectors are referred to as &quot;components&quot; instead of &quot;features&quot; as a reminder that they come from a projection of the feature vectors along the eigenvectors, or eigen-<em>components</em>, of the covariance matrix. As the whitened vectors are in the same 2D grid structure as their original feature maps, we refer to them as &quot;component maps.&quot;&nbsp;</p> <p>[3] &quot;layer1&quot;, &quot;layer2&quot;, &quot;layer3&quot;, &quot;layer4&quot;.</p> <p>[4] From torchvision version &quot;0.15.2&quot;, model weights &quot;ResNet18_Weights.IMAGENET1K_V1&quot;.</p> <p>&nbsp;</p> <p>TODO ref to paper</p>