Core Layers

DL4J equivalents and API reference for Keras core layers — Dense, Flatten, Dropout, Reshape, Merge, Permute, and more.

Keras Core Layer Import

Core layers are implemented in the layers/core package of deeplearning4j-modelimport.

Support Summary

Keras Layer	DL4J Equivalent	Supported
Dense	DenseLayer	Yes
Activation	ActivationLayer	Yes
Dropout	DropoutLayer	Yes
Flatten	CnnToFeedForwardPreProcessor / RnnToFeedForwardPreProcessor	Yes
Reshape	ReshapePreProcessor	Yes
Merge	MergeVertex	Yes
Permute	PermutePreProcessor	Yes
RepeatVector	RepeatVector	Yes
Lambda	SameDiffLambda	Yes
ActivityRegularization	—	No
Masking	MaskZeroLayer	Yes
SpatialDropout1D	DropoutLayer (spatial 1D)	Yes
SpatialDropout2D	DropoutLayer (spatial 2D)	Yes
SpatialDropout3D	DropoutLayer (spatial 3D)	Yes

---

KerasPermute

source

Imports a Permute layer from Keras as a DL4J input preprocessor.

Constructor

public KerasPermute(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• layerConfig — dictionary containing Keras layer configuration

isInputPreProcessor

public boolean isInputPreProcessor()

Returns true; Permute is implemented as an input preprocessor in DL4J.

getInputPreprocessor

public InputPreProcessor getInputPreprocessor(InputType... inputType)
        throws InvalidKerasConfigurationException

Returns the DL4J InputPreProcessor corresponding to the Permute configuration.

getOutputType

public InputType getOutputType(InputType... inputType) throws InvalidKerasConfigurationException

Returns the output InputType after permutation.

---

KerasFlatten

source

Imports a Keras Flatten layer as a DL4J CnnToFeedForwardInputPreProcessor or RnnToFeedForwardPreProcessor, depending on the preceding layer type.

Constructor

public KerasFlatten(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

isInputPreProcessor

public boolean isInputPreProcessor()

Returns true; Flatten is implemented as an input preprocessor.

getInputPreprocessor

public InputPreProcessor getInputPreprocessor(InputType... inputType)
        throws InvalidKerasConfigurationException

Selects the correct DL4J preprocessor based on inputType.

getOutputType

public InputType getOutputType(InputType... inputType) throws InvalidKerasConfigurationException

---

KerasReshape

source

Imports a Keras Reshape layer as a DL4J input preprocessor.

Constructor

public KerasReshape(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

isInputPreProcessor / getInputPreprocessor / getOutputType

Same pattern as KerasFlatten above.

---

KerasMerge

source

Imports a Keras Merge layer as a DL4J graph vertex. Supports add, multiply, subtract, average, maximum, and concatenate modes. The dot mode is not supported.

Constructor

public KerasMerge(Integer kerasVersion) throws UnsupportedKerasConfigurationException

getOutputType

public InputType getOutputType(InputType... inputType)

---

KerasDense

source

Imports a Keras Dense layer as a DL4J DenseLayer.

Constructor

public KerasDense(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

getDenseLayer

public DenseLayer getDenseLayer()

Returns the DL4J DenseLayer.

getOutputType

public InputType getOutputType(InputType... inputType) throws InvalidKerasConfigurationException

getNumParams

public int getNumParams()

Returns the number of trainable parameters (weights + bias).

setWeights

public void setWeights(Map<String, INDArray> weights) throws InvalidKerasConfigurationException

---

KerasDropout

source

Imports a Keras Dropout layer as a DL4J DropoutLayer.

Constructor

public KerasDropout(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

getDropoutLayer

public DropoutLayer getDropoutLayer()

getOutputType

public InputType getOutputType(InputType... inputType) throws InvalidKerasConfigurationException

---

KerasMasking

source

Imports a Keras Masking layer as a DL4J MaskZeroLayer.

---

KerasSpatialDropout

source

Imports Keras SpatialDropout1D, SpatialDropout2D, and SpatialDropout3D layers as DL4J spatial dropout variants.

---

KerasLambda

source

Imports a Keras Lambda layer. The Lambda layer is mapped to a SameDiff-based lambda function in DL4J. The function body is serialized separately.

---

KerasRepeatVector

source

Imports a Keras RepeatVector layer, which repeats the input a fixed number of times along a new axis.

---

Notes

• ActivityRegularization has no direct DL4J equivalent and will throw UnsupportedKerasConfigurationException at import time. Apply regularization directly in the layer configuration instead.

• The Permute, Flatten, and Reshape layers are all implemented as InputPreProcessor instances in DL4J rather than as trainable layers.