VisionHeatmaps.jl

heatmap(x::AbstractArray)
heatmap(x::AbstractArray, pipeline)
heatmap(x::AbstractArray, image)
heatmap(x::AbstractArray, image, pipeline)

Visualize 4D arrays as heatmaps, assuming the WHCN convention for input array dimensions (width, height, color channels, batch dimension).

heatmap(expl::Explanation)
heatmap(expl::Explanation, pipeline)
heatmap(expl::Explanation, image)   
heatmap(expl::Explanation, image, pipeline)

Visualize Explanation from XAIBase as a vision heatmap. Assumes WHCN convention (width, height, channels, batch dimension) for explanation.val. This will use the default heatmapping style for the given type of explanation.

heatmap(input::AbstractArray, analyzer::AbstractXAIMethod)
heatmap(input::AbstractArray, analyzer::AbstractXAIMethod, image)

Compute an Explanation for a given input using the XAI method analyzer and visualize it as a vision heatmap. This will use the default heatmapping style for the given type of explanation.

Pipeline(transforms...)

Heatmapping pipelines sequentially apply transforms.

AbstractTransform

Abstract supertype for all heatmapping transforms. Custom transforms must implement an apply method.

NormReduction()

Computes 2-norm over color channels

SumReduction()

Sums up color channels.

MaxAbsReduction()

Computes maximum(abs, x) over color channels

SumAbsReduction()

Computes sum(abs, x) over color channels

AbsSumReduction()

Computes abs(sum(x)) the color channels

FlipImage()

Permutes the width and height dimensions of an array. Assumes width and height are the leading directions in the array.

PermuteDims(dims...)

Permutes arrays according to the specified dimensions.

DropDims(dims...)

Drops specified singleton array dimensions.

PercentileClip()
PercentileClip(lower, upper)

Clip values outside of the specified percentiles of values. Bounds default to 0.001 and 0.999 (99.9% percentiles).

ExtremaColormap(name::Symbol)
ExtremaColormap(name::Symbol, colorscheme)

Apply a sequential colorscheme from ColorSchemes.jl, turning an array of values into an image. Defaults to :batlow.

CenteredColormap(name::Symbol)
CenteredColormap(name::Symbol, colorscheme)

Apply a divergent colorscheme from ColorSchemes.jl, turning an array of values into an image. Defaults to :berlin.

ResizeToImage()
ResizeToImage(method)

Resize the heatmap to match the image dimensions. In some cases, this is needed for overlays. Defaults to Lanczos(1) from Interpolations.jl.

AlphaOverlay()
AlphaOverlay(alpha)

Overlays a heatmap on top of an image. The opacity alpha of the heatmap defaults to 0.6.