Class: LucidRenderer

LucidRenderer()

Encapsulates state and methods for various feature visualization techniques.

Constructor

new LucidRenderer()

Source:

Classes

LucidRenderer

Members

canOptimize

Check if renderer has optimization target layer set.
Source:

compileInput

Generates input parametrization and initializes regularization transform chain.
Source:

compileLossF

Prepares loss function for current objective.
Source:

compileObjective

Prepares optimization for current objective.
Source:

dispose

Disposes renderer.
Source:

getActivationMaps

Returns activation tensors for currently selected layer.
Source:

getActivationShape

Returns activation tensor shape for currently selected layer.
Source:

getCentralNeuronCoords

Get the 2D coordinates of the central neuron for a specific layer.
Source:

getChannelNumber

Returns number of channels of current layer.
Source:

getClassPrediction

Returns class prediction of current input image.
Source:

getObjectiveF

Returns objective function depending on specified type.
Source:

initObjectiveParamsForModel

initObjectiveParamsForModel - Initializes parameters according to model to provide some fitting default parameters for convenience: - layer: last Conv layer - neuron: central neuron for initial layer
Source:

initTransformF

Builds input transform chain
Source:

optimize

Do one optimization step
Source:

resizeStyleImage

Resizes style image to fit input image dimensions.
Source:

setChannel

Sets target channel. Can be changed interactively during optimization.
Source:

setClass

Sets target class. Can be changed interactively during optimization.
Source:

setContentImage

Set content image for style transfer.
Source:

setFeatureMapLayer

Sets layer to output featuremaps for.
Source:

setInputParams

setInputParams - Sets input params and causes re-compilation of model, not possible during running optimization.
Source:

setJitter

Set input jitter
Source:

setLayer

Sets layer for optimization objective. Can't change during optimization. Resets target neuron to central neuron.
Source:

setLearningRate

Set optimizer learning rate
Source:

setModel

Assigns model to the renderer.
Source:

setNegative

Set negative optimization objective status
Source:

setNeuron

Sets target neuron. Can be changed interactively during optimization.
Source:

setObjectiveType

Sets the optimization objective type.
Source:

setStyleImage

Set style image for style transfer.
Source:

setStyleLayers

Sets layers that should be considered for style loss.
Source:

startOptimization

Stars optimization loop.
Source:

stopOptimization

Stops optimization loop.
Source:

Methods

deprocessFeatureMap()

Returns feature map tensor in 0 - 255 color range.
Source: