Pooling Layers

DL4J equivalents and API reference for Keras pooling layers — MaxPooling, AveragePooling, and GlobalPooling in 1D, 2D, and 3D variants.

Keras Pooling Layer Import

Pooling layer support is implemented in the layers/pooling package. All standard Keras pooling layers are supported.

Support Summary

Keras Layer	DL4J Equivalent	Supported
MaxPooling1D	Subsampling1DLayer (MAX)	Yes
MaxPooling2D	SubsamplingLayer (MAX)	Yes
MaxPooling3D	Subsampling3DLayer (MAX)	Yes
AveragePooling1D	Subsampling1DLayer (AVG)	Yes
AveragePooling2D	SubsamplingLayer (AVG)	Yes
AveragePooling3D	Subsampling3DLayer (AVG)	Yes
GlobalMaxPooling1D	GlobalPoolingLayer (MAX)	Yes
GlobalMaxPooling2D	GlobalPoolingLayer (MAX)	Yes
GlobalMaxPooling3D	GlobalPoolingLayer (MAX)	Yes
GlobalAveragePooling1D	GlobalPoolingLayer (AVG)	Yes
GlobalAveragePooling2D	GlobalPoolingLayer (AVG)	Yes
GlobalAveragePooling3D	GlobalPoolingLayer (AVG)	Yes

---

KerasPooling1D

source

Imports a Keras 1D pooling layer (MaxPooling1D or AveragePooling1D) as a DL4J Subsampling1DLayer.

Constructor

public KerasPooling1D(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• layerConfig — dictionary containing the Keras layer configuration

getSubsampling1DLayer

public Subsampling1DLayer getSubsampling1DLayer()

Returns the DL4J Subsampling1DLayer. The pooling type (MAX or AVG) is inferred from the Keras class name.

getOutputType

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

---

KerasPooling2D

source

Imports a Keras 2D pooling layer (MaxPooling2D or AveragePooling2D) as a DL4J SubsamplingLayer.

Constructor

public KerasPooling2D(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

getSubsampling2DLayer

public SubsamplingLayer getSubsampling2DLayer()

Returns the DL4J SubsamplingLayer.

getOutputType

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

---

KerasPooling3D

source

Imports a Keras 3D pooling layer (MaxPooling3D or AveragePooling3D) as a DL4J Subsampling3DLayer.

Constructor

public KerasPooling3D(Map<String, Object> layerConfig)
        throws InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

getSubsampling3DLayer

public Subsampling3DLayer getSubsampling3DLayer()

---

KerasGlobalPooling

source

Imports any Keras global pooling layer (GlobalMaxPooling1D/2D/3D, GlobalAveragePooling1D/2D/3D) as a DL4J GlobalPoolingLayer.

Constructor

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

getGlobalPoolingLayer

public GlobalPoolingLayer getGlobalPoolingLayer()

Returns the DL4J GlobalPoolingLayer. Pooling type and dimensionality are inferred from the Keras class name.

getInputPreprocessor

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

Returns the appropriate InputPreProcessor when the global pooling layer follows a layer with a different output format (e.g., following a convolutional layer).

getOutputType

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

---

KerasPoolingUtils

source

Utility class for pooling layer import.

mapPoolingType

public static PoolingType mapPoolingType(String className, KerasLayerConfiguration conf)
        throws UnsupportedKerasConfigurationException

Maps a Keras pooling class name (e.g., "MaxPooling2D") to the corresponding DL4J PoolingType enum value (PoolingType.MAX or PoolingType.AVG).

Parameters:

• className — the Keras layer class name string

• conf — the KerasLayerConfiguration instance

Returns: DL4J PoolingType

---

Notes

• All pooling layers use the pool_size, strides, and padding parameters from the Keras configuration.

• The padding value "same" and "valid" are both supported and translated to the equivalent DL4J padding modes.

• Global pooling layers collapse the spatial dimensions entirely; DL4J's GlobalPoolingLayer handles all three dimensionalities via the same class, parameterized by PoolingDimensions.