Global

Methods

activationDifference(imgActivations, targetActivations, transformF, activLossF)

Difference between activations of two input images.
Parameters:
Name Type Default Description
imgActivations *
targetActivations *
transformF * null
activLossF *
Source:

activationModification(model, originalImage, activationModDict, activationLossF)

Optimize for specified activation modifications.
Parameters:
Name Type Description
model *
originalImage *
activationModDict *
activationLossF *
Source:

alertDialog()

Displays alert dialog with message and close callback.
Source:

App()

All routes go here. Don't forget to import the components above after adding new route.
Source:

baseImageLaplacianPyramid(shape, imgData, nLevels, decorrelate)

Returns image-initialized laplacian pyramid as sum function and individual layer variables.
Parameters:
Name Type Description
shape * image shape
imgData * initial image
nLevels * pyramid layers
decorrelate * decorrelate colors
Source:

channel()

Visualize a single channel.
Source:

deepdream()

Maximize 'interestingness' at some layer. See Mordvintsev et al., 2015.
Source:

fillCanvasPixelsWithGreyAndAlpha(canvasPixels, greyData, width, height, channel, mult)

Replicates grey pixel array into rgba canvas pixel array.
Parameters:
Name Type Description
canvasPixels * rgba pixel array
greyData * single channel pixel array
width *
height *
channel * channel offset in case input contains multiple concatenated channels
mult * multiply src pixel data (e.g. to convert from normalized 0-1 to 0-255)
Source:

fillCanvasPixelsWithRgbAndAlpha(canvasPixels, rgbData, width, height, channel, mult)

Writes rgb pixel array into rgba canvas pixel array.
Parameters:
Name Type Description
canvasPixels * rgba pixel array
rgbData * rgb pixel array
width *
height *
channel * channel offset
mult * multiply src pixel data (e.g. to convert from normalized 0-1 to 0-255)
Source:

gaussianKernel3x3(ch)

Computes gaussian 3x3 kernel
Parameters:
Name Type Description
ch * channels
Source:

getActivationsForLayers(model, image, layers)

Returns list of activation tensors for each layer.
Parameters:
Name Type Description
model *
image *
layers *
Source:

getAuxModel(model, outputs)

Returns new model with specified outputs.
Parameters:
Name Type Description
model *
outputs *
Source:

getGrayFromCanvasRgba(canvasPixels, grayData, width, height)

Puts red channel of rgba pixel array into single channel pixel array.
Parameters:
Name Type Description
canvasPixels * rgba pixel array
grayData * single channel pixel array
width *
height *
Source:

getImageData(img)

Returns ImageData object from image element.
Parameters:
Name Type Description
img * html image element
Source:

getImgDataFromFile(file, cb)

Passes ImageData object from image file to callback
Parameters:
Name Type Description
file * image file
cb * callback
Source:

getLayerOutputs(model, layers)

Returns output layer tensors.
Parameters:
Name Type Description
model *
layers *
Source:

getRgbFromCanvasRgba(canvasPixels, rgbData, width, height)

Converts rgba pixel array to rgb pixel array.
Parameters:
Name Type Description
canvasPixels * rgba pixel array
rgbData * rgb pixel array to write into
width *
height *
Source:

gramStyle(array)

Returns gram matrix of input tensor, normalized by length of flat length.
Parameters:
Name Type Description
array *
Source:

imgLaplacianPyramid(imgArray, w, h, ch, batch, decorrelate, nLevels)

Returns input function and variables in naive pixel space parametrization with image initialization.
Parameters:
Name Type Description
imgArray * input image
w * width
h * height
ch * channels
batch * batchsize
decorrelate * decorrelate colors
nLevels * number of pyramid layers
Source:

inverseDecorrelate(t)

Applies inverse decorrelation: can be used to preserve color values when the input is initalized with an image and decorrelation is applied afterwards in the computation graph.
Parameters:
Name Type Description
t * Input tensor
Source:

laplacianPyramid(shape, sd, nLevels)

Returns random-initialized laplacian pyramid as sum function and individual layer variables. Contains outcommented code for fading in pyramid layers.
Parameters:
Name Type Description
shape * tensor shape
sd * standard deviation
nLevels * pyramid layers
Source:

linearDecorrelateColor(t)

Applies linear decorrelation to input tensor.
Parameters:
Name Type Description
t * Input tensor
Source:

loadJSON(file, callback)

Reads json file content from URL and passes it to callback.
Parameters:
Name Type Description
file *
callback *
Source:

loadJSONFromLocalFile(file, callback)

Reads json file content from local file and passes it to callback.
Parameters:
Name Type Description
file *
callback *
Source:

makeArray(tensorOrArray)

Returns array of tensors from tensor or array of tensors
Parameters:
Name Type Description
tensorOrArray *
Source:

makeLaplacianPyramidFromImgData(shape, imgData, nLevels, invDec)

Returns image-initialized laplacian pyramid as summed image and individual layers.
Parameters:
Name Type Description
shape * image shape
imgData * initial image
nLevels * pyramid layers
invDec * inverse decorrelate
Source:

meanL1Loss(g1, g2)

Returns mean of absolute differences.
Parameters:
Name Type Description
g1 *
g2 *
Source:

meanL2Loss(g1, g2)

Returns mean of squared differences.
Parameters:
Name Type Description
g1 *
g2 *
Source:

naiveFromImage(imgArray, w, h, ch, batch, decorrelate)

Returns input function and variables with naive pixel space parametrization of an existing image.
Parameters:
Name Type Description
imgArray Uint8Array Input image data
w * width
h * height
ch * channels
batch * batchsize
decorrelate * decorrelate colors
Source:

neuron()

Visualize a single neuron of a single channel. Defaults to the center neuron. When width and height are even numbers, we choose the neuron in the bottom right of the center 2x2 neurons. Odd width & height: Even width & height: +---+---+---+ +---+---+---+---+ | | | | | | | | | +---+---+---+ +---+---+---+---+ | | X | | | | | | | +---+---+---+ +---+---+---+---+ | | | | | | | X | | +---+---+---+ +---+---+---+---+ | | | | | +---+---+---+---+
Source:

output(model, options)

Maximize output / class activation.
Parameters:
Name Type Description
model *
options *
Source:

pixelImage(shape, sd, initVal)

Returns randomly initialized input image variables.
Parameters:
Name Type Description
shape *
sd *
initVal *
Source:

randImage(w, h, ch, batch, sd, decorrelate, alpha)

Returns input function and variables with naive pixel space parametrization with random initialization.
Parameters:
Name Type Description
w * width
h * height
ch * channels
batch * batchsize
sd * standard deviation
decorrelate * decorrelate colors
alpha * use alpha channel
Source:

randLaplacianPyramid(w, h, ch, batch, sd, decorrelate, nLevels)

Returns input function and variables with laplacian pyramid parametrization and random initialization.
Parameters:
Name Type Description
w * width
h * height
ch * channels
batch * batchsize
sd * standard deviation
decorrelate * decorrelate colors
nLevels * number of pyramid layers
Source:

render() → {*}

Renders this react component
Source:
Returns:
the components contents
Type
*

showImage(fileReader, cb)

Creates image element from FileReader.
Parameters:
Name Type Description
fileReader *
cb *
Source:

spatial(model, options)

Maximize single "pixel" location
Parameters:
Name Type Description
model *
options *
Source:

style(model, contentImg, styleImg, contentLrs, styleLrs)

Experimental style objective
Parameters:
Name Type Description
model *
contentImg *
styleImg *
contentLrs *
styleLrs *
Source:

toValidRgb(t, decorrelate, sigmoid, normalize)

Transforms the optimization parameter values into valid RGB space and applies optional additional functions.
Parameters:
Name Type Description
t * Input tensor
decorrelate boolean Decorrelate colors
sigmoid boolean Apply sigmoid
normalize boolean Normalize to 1.0
Source: