Unsupported keras layer type MyCrossLayer

DL4J
1

I customized keras layer by myself in python keras. When I read .h5 file call the error.

SLF4J: Failed to load class “org.slf4j.impl.StaticLoggerBinder”.
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See SLF4J Error Codes for further details.
Exception in thread “main” org.deeplearning4j.nn.modelimport.keras.exceptions.UnsupportedKerasConfigurationException: Unsupported keras layer type MyCrossLayer. Please file an issue at Issues · eclipse/deeplearning4j · GitHub.
at org.deeplearning4j.nn.modelimport.keras.utils.KerasLayerUtils.getKerasLayerFromConfig(KerasLayerUtils.java:334)
at org.deeplearning4j.nn.modelimport.keras.KerasModel.prepareLayers(KerasModel.java:218)
at org.deeplearning4j.nn.modelimport.keras.KerasModel.(KerasModel.java:164)
at org.deeplearning4j.nn.modelimport.keras.KerasModel.(KerasModel.java:96)
at org.deeplearning4j.nn.modelimport.keras.utils.KerasModelBuilder.buildModel(KerasModelBuilder.java:307)
at org.deeplearning4j.nn.modelimport.keras.KerasModelImport.importKerasModelAndWeights(KerasModelImport.java:172)
at Formatter.main(Formatter.java:12)

2

WechatIMG38
WechatIMG381297×807 32.5 KB

3

Hi, you need to register an equivalent layer in dl4j in order for the model import to work. You can find an overview here: https://jrmerwin.github.io/deeplearning4j-docs/model-import-keras.html

4

Can you give me some '‘register an equivalent layer’‘ example.I can’t find in this overview

5

You can find some simple examples of defining a SameDiffLambdaLayer here: https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-modelimport/src/test/java/org/deeplearning4j/nn/modelimport/keras/e2e/KerasLambdaTest.java#L47

There’s also this guide with a little more information and background on defining custom layers: https://deeplearning4j.konduit.ai/keras-import/custom-layers