Gitiles
Code Review Sign In
review.mlplatform.org / ml / armnn / d998a1cfbcc7b4ed9922532ffe19c51283fcae68 / . / docs / 05_02_deserializer_serializer.dox
blob: 84324d89d50601c25da6d9f644103653c18b23a7 [file] [log] [blame]
Jan Eilers31a7c892021-01-29 14:18:13 +0000[diff] [blame]1/// Copyright (c) 2021 ARM Limited and Contributors. All rights reserved.
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]2///
3/// SPDX-License-Identifier: MIT
4///
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]5
6namespace armnn
7{
8/**
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]9@page serializer Serializer
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]10@tableofcontents
11
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]12The `armnnSerializer` is a library for serializing an Arm NN network to a stream.
13
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]14@section serializersupport Supported Layers
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]15
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]16This reference guide provides a list of layers which can be serialized by the Arm NN SDK.
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]17
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]18@subsection serializersupportflully Fully supported
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]19
20The Arm NN SDK Serializer currently supports the following layers:
21
22- Activation
23- Addition
24- ArgMinMax
25- BatchToSpaceNd
26- BatchNormalization
mathad01b392e982021-04-07 12:07:30 +0100[diff] [blame]27- Cast
Simon Obute51f67772021-09-03 15:50:13 +0100[diff] [blame]28- ChannelShuffle
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]29- Comparison
30- Concat
31- Constant
32- Convolution2d
Matthew Sloyanb63a3112021-09-08 13:05:51 +0100[diff] [blame]33- Convolution3d
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]34- DepthToSpace
35- DepthwiseConvolution2d
36- Dequantize
37- DetectionPostProcess
38- Division
39- ElementwiseUnary
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]40- Fill
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]41- Floor
42- FullyConnected
43- Gather
44- Input
45- InstanceNormalization
46- L2Normalization
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]47- Logical
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]48- LogSoftmax
49- Lstm
50- Maximum
51- Mean
52- Merge
53- Minimum
54- Multiplication
55- Normalization
56- Output
Matthew Sloyan2e5d0b22021-10-21 14:05:31 +0100[diff] [blame]57- Pad (Constant, Symmetric, Reflect)
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]58- Permute
59- Pooling2d
Tamas Nyirid998a1c2021-11-05 14:55:33 +0000[diff] [blame^]60- Pooling3d
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]61- Prelu
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]62- QLstm
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]63- Quantize
64- QuantizedLstm
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]65- Rank
Jan Eilers1121c6b2021-02-19 09:38:03 +0000[diff] [blame]66- Reduce
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]67- Reshape
68- Resize
Keith Davis3ae3f972021-05-21 16:33:48 +0100[diff] [blame]69- Shape
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]70- Slice
71- Softmax
72- SpaceToBatchNd
73- SpaceToDepth
74- Splitter
75- Stack
76- StandIn
77- StridedSlice
78- Subtraction
79- Switch
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]80- Transpose
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]81- TransposeConvolution2d
Narumol Prangnawarata0162e12021-07-23 14:47:49 +0100[diff] [blame]82- UnidirectionalSequenceLstm
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]83
84More machine learning layers will be supported in future releases.
85
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]86@subsection serializersupportdeprecated Deprecated layers
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]87
88Some 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:
89
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]90- Abs will deserialize as ElementwiseUnary
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]91- Equal will deserialize as Comparison
92- Greater will deserialize as Comparison
93- Merger will deserialize as Concat
94- ResizeBilinear will deserialize as Resize
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]95- Rsqrt will deserialize as ElementwiseUnary
96<br/><br/><br/><br/>
97
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]98@page deserializer Deserializer
99@tableofcontents
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]100
101The `armnnDeserializer` is a library for loading neural networks defined by Arm NN FlatBuffers files
102into the Arm NN runtime.
103
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]104@section deserializersupport Supported Layers
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]105
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]106This reference guide provides a list of layers which can be deserialized by the Arm NN SDK.
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]107
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]108@subsection deserializersupportfully Fully supported
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]109
110The Arm NN SDK Deserialize parser currently supports the following layers:
111
112- Abs
113- Activation
114- Addition
115- ArgMinMax
116- BatchToSpaceNd
117- BatchNormalization
mathad01b392e982021-04-07 12:07:30 +0100[diff] [blame]118- Cast
Simon Obute51f67772021-09-03 15:50:13 +0100[diff] [blame]119- ChannelShuffle
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]120- Concat
121- Comparison
122- Constant
123- Convolution2d
124- DepthToSpace
125- DepthwiseConvolution2d
126- Dequantize
127- DetectionPostProcess
128- Division
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]129- ElementwiseUnary
130- Fill
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]131- Floor
132- FullyConnected
133- Gather
134- Input
135- InstanceNormalization
136- L2Normalization
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]137- Logical
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]138- LogSoftmax
139- Lstm
140- Maximum
141- Mean
142- Merge
143- Minimum
144- Multiplication
145- Normalization
146- Output
147- Pad
148- Permute
149- Pooling2d
150- Prelu
151- Quantize
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]152- QLstm
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]153- QuantizedLstm
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]154- Rank
Jan Eilers1121c6b2021-02-19 09:38:03 +0000[diff] [blame]155- Reduce
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]156- Reshape
James Ward590f3f72020-12-02 10:20:48 +0000[diff] [blame]157- Resize
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]158- ResizeBilinear
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]159- Slice
160- Softmax
161- SpaceToBatchNd
162- SpaceToDepth
163- Splitter
164- Stack
165- StandIn
166- StridedSlice
167- Subtraction
168- Switch
Jan Eilers1121c6b2021-02-19 09:38:03 +0000[diff] [blame]169- Transpose
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]170- TransposeConvolution2d
Narumol Prangnawarata0162e12021-07-23 14:47:49 +0100[diff] [blame]171- UnidirectionalSequenceLstm
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]172
173More machine learning layers will be supported in future releases.
174
Jan Eilers1ed388f2021-01-28 14:45:09 +0000[diff] [blame]175@subsection deserializersupportdeprecated Deprecated layers
Ryan OSheaf3a43232020-02-12 16:15:27 +0000[diff] [blame]176
177Some 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:
178
179- Equal will deserialize as Comparison
180- Merger will deserialize as Concat
181- Greater will deserialize as Comparison
182- ResizeBilinear will deserialize as Resize
183
184**/
185}