The armnnSerializer is a library for serializing an Arm NN network to a stream.

Supported Layers

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

The Arm NN SDK Serializer currently supports the following layers:

Activation
Addition
ArgMinMax
BatchMatMul
BatchToSpaceNd
BatchNormalization
Cast
ChannelShuffle
Comparison
Concat
Constant
Convolution2d
Convolution3d
DepthToSpace
DepthwiseConvolution2d
Dequantize
DetectionPostProcess
Division
ElementwiseUnary
Fill
Floor
FullyConnected
Gather
GatherNd
Input
InstanceNormalization
L2Normalization
Logical
LogSoftmax
Lstm
Maximum
Mean
Merge
Minimum
Multiplication
Normalization
Output
Pad (Constant, Symmetric, Reflect)
Permute
Pooling2d
Pooling3d
Prelu
QLstm
Quantize
QuantizedLstm
Rank
Reduce
Reshape
Resize
ReverseV2
ScatterNd
Shape
Slice
Softmax
SpaceToBatchNd
SpaceToDepth
Splitter
Stack
StandIn
StridedSlice
Subtraction
Switch
Transpose
TransposeConvolution2d
UnidirectionalSequenceLstm

More machine learning layers will be supported in future releases.

Deprecated layers

Some layers have been deprecated and replaced by others layers. In order to maintain backward compatibility, serializations of these deprecated layers will deserialize to the layers that have replaced them, as follows:

Abs will deserialize as ElementwiseUnary
Equal will deserialize as Comparison
Greater will deserialize as Comparison
Merger will deserialize as Concat
ResizeBilinear will deserialize as Resize
Rsqrt will deserialize as ElementwiseUnary