Average Activation
Classes Histogram
Top 10 classes activating the current filter the most in terms of average score and their corresponding spikes
Images Gradients
Images from the train dataset activating the current filter the most in terms of average score. The corresponding gradients can also be displayed.
Histogram Top
Average spikes corresponding to each of the images above (from top-left to bottom-right)
Max Activation
Classes Histogram
Top 10 classes activating the current filter the most in terms of maximum score and their corresponding spikes
ImagesGradients
Images from the train dataset activating the current filter the most in terms of maximum score. The corresponding gradients can also be displayed.
ImagesGradients
Cropped versions of the images above. It corresponds to the location in the original image to which the filter responds maximally.
Histogram Top
max spikes corresponding to each of the images above (from top-left to bottom-right)
Filter Visualization
Artificial image computed by gradient descent inducing the greatest activation for the current filter
Torchlurk