API

Feature Visualization

• ActivationMaximization

Attentions

• Vanilla Saliency and SmoothGrad

Class Activation Map

• GradCAM
• GradCAM++
• ScoreCAM
• LayerCAM

Utilities

• Model Modifiers
• Scores
• Functions

Model Visualization Class

class tf_keras_vis.ModelVisualization(model, model_modifier=None, clone=True)

Bases: ABC

Visualization class that analyze the model for debugging.

Parameters:

• model – A tf.keras.Model instance. When model_modifier is NOT None, this model will be cloned with tf.keras.models.clone_model function and then will be modified by model_modifier according to needs.

• model_modifier – A tf_keras_vis.utils.model_modifiers.ModelModifier instance, a function or a list of them. We recommend to apply tf_keras_vis.utils.model_modifiers.ReplaceToLinear to all visualizations (except tf_keras_vis.scorecam.Scorecam) when the model output is softmax. Defaults to None.

• clone – A bool that indicates whether or not it clones the model. When False, the model won’t be cloned. Note that, although when True, the model won’t be clone if model_modifier is None. Defaults to True.

abstract __call__()

Analyze the model.

Raises:

NotImplementedError – This method must be overwritten.

Returns:

Visualized image(s) or something(s).

Return type:

ndarray | list