The layers that ArmNN SDK Serializer currently supports.

This reference guide provides a list of layers which can be serialized currently by the Arm NN SDK.

Fully supported

The Arm NN SDK Serializer currently supports the following layers:

• Activation
• Addition
• BatchToSpaceNd
• BatchNormalization
• Concat
• Constant
• Convolution2d
• DepthwiseConvolution2d
• Dequantize
• DetectionPostProcess
• Division
• Equal
• Floor
• FullyConnected
• Gather
• Greater
• L2Normalization
• Lstm
• Maximum
• Mean
• Merge
• Minimum
• Multiplication
• Normalization
• Pad
• Permute
• Pooling2d
• Prelu
• Quantize
• Reshape
• ResizeBilinear
• Rsqrt
• Softmax
• SpaceToBatchNd
• SpaceToDepth
• Splitter
• StridedSlice
• Subtraction
• Switch

More machine learning layers will be supported in future releases.

Note: Merger layer has been renamed Concat. Serializations of the old format with Merger layers will deserialize to Concat layers to maintain backward compatibility.