README.md

   1 # NNtrainer
   2
   3 [![Code Coverage](http://nnstreamer.mooo.com/nntrainer/ci/badge/codecoverage.svg)](http://nnstreamer.mooo.com/nntrainer/ci/gcov_html/index.html)
   4 ![GitHub repo size](https://img.shields.io/github/repo-size/nnstreamer/nntrainer)
   5 ![GitHub issues](https://img.shields.io/github/issues/nnstreamer/nntrainer)
   6 ![GitHub pull requests](https://img.shields.io/github/issues-pr/nnstreamer/nntrainer)
   7 <a href="https://scan.coverity.com/projects/nnstreamer-nntrainer">
   8   <img alt="Coverity Scan Build Status"
   9        src="https://scan.coverity.com/projects/22512/badge.svg"/>
  10 </a>
  11 [![DailyBuild](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/daily_build_test_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/)
  12
  13 NNtrainer is a Software Framework for training Neural Network models on devices.
  14
  15 ## Overview
  16
  17 NNtrainer is an Open Source Project. The aim of the NNtrainer is to develop a Software Framework to train neural network models on embedded devices which have relatively limited resources. Rather than training whole layers of a network from the scratch, NNtrainer finetunes the neural network model on device with user data for the personalization.
  18
  19 Even if NNtariner runs on device, it provides full functionalities to train models and also utilizes limited device resources efficiently. NNTrainer is able to train various machine learning algorithms such as k-Nearest Neighbor (k-NN), Neural Networks, Logistic Regression, Reinforcement Learning algorithms, Recurrent network and more. We also provide examples for various tasks such as Few-shot learning, ResNet, VGG, Product Rating and more will be added. All of these were tested on Samsung Galaxy smart phone with Android and PC (Ubuntu 18.04/20.04).
  20
  21 [ NNTrainer: Personalize neural networks on devices! ](https://www.youtube.com/watch?v=HKKowY78P1A), Samsung Developer Conference 2021 <br />
  22 [ NNTrainer: "On-device learning" ](https://www.youtube.com/embed/Jy_auavraKg?start=4035&end=4080), Samsung AI Forum 2021
  23
  24 ## Official Releases
  25
  26 |     | [Tizen](http://download.tizen.org/snapshots/tizen/unified/latest/repos/standard/packages/) | [Ubuntu](https://launchpad.net/~nnstreamer/+archive/ubuntu/ppa) | Android/NDK Build |
  27 | :-- | :--: | :--: | :--: |
  28 |     | 6.0M2 and later | 18.04 | 9/P |
  29 | arm | [![armv7l badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/tizen.armv7l_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/) | Available  | Ready |
  30 | arm64 |  [![aarch64 badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/tizen.aarch64_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/) | Available  | [![android badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/arm64_v8a_android_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/) |
  31 | x64 | [![x64 badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/tizen.x86_64_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/)  | [![ubuntu badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/ubuntu_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/)  | Ready  |
  32 | x86 | [![x86 badge](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/badge/tizen.i586_result_badge.svg)](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/)  | N/A  | N/A  |
  33 | Publish | [Tizen Repo](http://download.tizen.org/snapshots/tizen/unified/latest/repos/standard/packages/) | [PPA](https://launchpad.net/~nnstreamer/+archive/ubuntu/ppa) |   |
  34 | API | C (Official) | C/C++ | C/C++  |
  35
  36 - Ready: CI system ensures build-ability and unit-testing. Users may easily build and execute. However, we do not have automated release & deployment system for this instance.
  37 - Available: binary packages are released and deployed automatically and periodically along with CI tests.
  38 - [Daily Release](http://nnstreamer.mooo.com/nntrainer/ci/daily-build/build_result/)
  39 - SDK Support: Tizen Studio (6.0 M2+)
  40
  41 ## Maintainer
  42 * [Jijoong Moon](https://github.com/jijoongmoon)
  43 * [MyungJoo Ham](https://github.com/myungjoo)
  44 * [Geunsik Lim](https://github.com/leemgs)
  45
  46 ## Reviewers
  47 * [Sangjung Woo](https://github.com/again4you)
  48 * [Wook Song](https://github.com/wooksong)
  49 * [Jaeyun Jung](https://github.com/jaeyun-jung)
  50 * [Hyoungjoo Ahn](https://github.com/helloahn)
  51 * [Parichay Kapoor](https://github.com/kparichay)
  52 * [Dongju Chae](https://github.com/dongju-chae)
  53 * [Gichan Jang](https://github.com/gichan-jang)
  54 * [Yongjoo Ahn](https://github.com/anyj0527)
  55 * [Jihoon Lee](https://github.com/zhoonit)
  56 * [Hyeonseok Lee](https://github.com/lhs8928)
  57 * [Mete Ozay](https://github.com/meteozay)
  58
  59 ## Components
  60
  61 ### Supported Layers
  62
  63 This component defines layers which consist of a neural network model. Layers have their own properties to be set.
  64
  65  | Keyword | Layer Class Name | Description |
  66  |:-------:|:---:|:---|
  67  | conv1d | Conv1DLayer | Convolution 1-Dimentional Layer |
  68  | conv2d | Conv2DLayer |Convolution 2-Dimentional Layer |
  69  | pooling2d | Pooling2DLayer |Pooling 2-Dimentional Layer. Support average / max / global average / global max pooling |
  70  | flatten | FlattenLayer | Flatten layer |
  71  | fully_connected | FullyConnectedLayer | Fully connected layer |
  72  | input | InputLayer | Input Layer.  This is not always required. |
  73  | batch_normalization | BatchNormalizationLayer | Batch normalization layer |
  74  | activation | ActivaitonLayer | Set by layer property |
  75  | addition | AdditionLayer | Add input input layers |
  76  | attention | AttentionLayer | Attenstion layer |
  77  | centroid_knn | CentroidKNN | Centroid K-nearest neighbor layer |
  78  | concat | ConcatLayer | Concatenate input layers |
  79  | multiout | MultiOutLayer | Multi-Output Layer |
  80  | backbone_nnstreamer | NNStreamerLayer | Encapsulate NNStreamer layer |
  81  | backbone_tflite | TfLiteLayer | Encapsulate tflite as an layer |
  82  | permute | PermuteLayer | Permute layer for transpose |
  83  | preprocess_flip | PreprocessFlipLayer | Preprocess random flip layer |
  84  | preprocess_l2norm | PreprocessL2NormLayer | Preprocess simple l2norm layer to normalize |
  85  | preprocess_translate | PreprocessTranslateLayer | Preprocess translate layer |
  86  | reshape | ReshapeLayer | Reshape tensor dimension layer |
  87  | split | SplitLayer | Split layer |
  88  | dropout | DropOutLayer | Dropout Layer |
  89  | embedding | EmbeddingLayer | Embedding Layer |
  90  | rnn | RNNLayer | Recurrent Layer |
  91  | gru | GRULayer | Gated Recurrent Unit Layer |
  92  | lstm | LSTMLayer | Long Short-Term Memory Layer |
  93  | lstmcell | LSTMCellLayer | Long Short-Term Memory Cell Layer |
  94  | time_dist | TimeDistLayer | Time distributed Layer |
  95
  96 ### Supported Optimizers
  97
  98 NNTrainer Provides
  99
 100  | Keyword | Optimizer Name | Description |
 101  |:-------:|:---:|:---:|
 102  | sgd | Stochastic Gradient Decent | - |
 103  | adam | Adaptive Moment Estimation | - |
 104
 105 ### Supported Loss Functions
 106
 107 NNTrainer provides
 108
 109  | Keyword | Class Name | Description |
 110  |:-------:|:---:|:---:|
 111  | cross_sigmoid | CrossEntropySigmoidLossLayer | Cross entropy sigmoid loss layer |
 112  | cross_softmax | CrossEntropySoftmaxLossLayer | Cross entropy softmax loss layer |
 113  | constant_derivative | ConstantDerivativeLossLayer | Constant derivative loss layer |
 114  | mse | MSELossLayer | Mean square error loss layer |
 115
 116 ### Supported Activation Functions
 117
 118 NNTrainer provides
 119
 120  | Keyword | Loss Name | Description |
 121  |:-------:|:---:|:---|
 122  | tanh | tanh function | set as layer property |
 123  | sigmoid | sigmoid function | set as layer property |
 124  | relu | relu function | set as layer propery |
 125  | softmax | softmax function | set as layer propery |
 126  | weight_initializer | Weight Initialization | Xavier(Normal/Uniform), LeCun(Normal/Uniform),  HE(Normal/Unifor) |
 127  | weight_regularizer | weight decay ( L2Norm only ) | needs set weight_regularizer_param & type |
 128  | learnig_rate_decay | learning rate decay | need to set step |
 129
 130 ### Tensor
 131
 132 Tensor is responsible for calculation of a layer. It executes several operations such as addition, division, multiplication, dot production, data averaging and so on. In order to accelerate  calculation speed, CBLAS (C-Basic Linear Algebra: CPU) and CUBLAS (CUDA: Basic Linear Algebra) for PC (Especially NVIDIA GPU) are implemented for some of the operations. Later, these calculations will be optimized.
 133 Currently, we supports lazy calculation mode to reduce complexity for copying tensors during calculations.
 134
 135  | Keyword | Description |
 136  |:-------:|:---:|
 137  | 4D Tensor | B, C, H, W|
 138  | Add/sub/mul/div | - |
 139  | sum, average, argmax | - |
 140  | Dot, Transpose | - |
 141  | normalization, standardization | - |
 142  | save, read | - |
 143
 144 ### Others
 145
 146 NNTrainer provides
 147
 148  | Keyword | Loss Name | Description |
 149  |:-------:|:---:|:---|
 150  | weight_initializer | Weight Initialization | Xavier(Normal/Uniform), LeCun(Normal/Uniform),  HE(Normal/Unifor) |
 151  | weight_regularizer | weight decay ( L2Norm only ) | needs set weight_regularizer_constant & type |
 152  | learnig_rate_decay | learning rate decay | need to set step |
 153
 154 ### APIs
 155 Currently, we provide [C APIs](https://github.com/nnstreamer/nntrainer/blob/master/api/capi/include/nntrainer.h) for Tizen. [C++ APIs](https://github.com/nnstreamer/nntrainer/blob/master/api/ccapi/include) are also provided for other platform. Java & C# APIs will be provided soon.
 156
 157
 158 ### Examples for NNTrainer
 159
 160 #### [Custom Shortcut Application](https://github.com/nnstreamer/nntrainer/tree/main/Applications/Tizen_native/CustomShortcut)
 161
 162
 163 A demo application which enable user defined custom shortcut on galaxy watch.
 164
 165 #### [MNIST Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/MNIST)
 166
 167 An example to train mnist dataset. It consists two convolution 2d layer, 2 pooling 2d layer, flatten layer and fully connected layer.
 168
 169 #### [Reinforcement Learning Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/ReinforcementLearning/DeepQ)
 170
 171 A reinforcement learning example with cartpole game. It is using DeepQ algorithm.
 172
 173 #### [Transfer Learning Examples](https://github.com/nnstreamer/nntrainer/tree/main/Applications/TransferLearning)
 174
 175 Transfer learning examples with for image classification using the Cifar 10 dataset and for OCR. TFlite is used for feature extractor and modify last layer (fully connected layer) of network.
 176
 177 #### [ResNet Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/Resnet)
 178
 179 An example to train resnet18 network.
 180
 181 #### [VGG Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/VGG)
 182
 183 An example to train vgg16 network.
 184
 185 #### [ProductRating Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/ProductRatings)
 186
 187 This application contains a simple embedding-based model that predicts ratings given a user and a product.
 188
 189 #### [SimpleShot Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/SimpleShot)
 190
 191 An example to demonstrate few-shot learning : SimpleShot
 192
 193 #### [Custom Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/Custom)
 194
 195 An example to demonstrate how to create custom layers, optimizers or other supported objects.
 196
 197 <!-- #### Tizen CAPI Example -->
 198
 199 <!-- An example to demonstrate c api for Tizen. It is same transfer learing but written with tizen c api.~ -->
 200 <!-- Deleted instead moved to a [test](https://github.com/nnstreamer/nntrainer/blob/master/test/tizen_capi/unittest_tizen_capi.cpp) -->
 201
 202 #### [KNN Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/KNN)
 203
 204 A transfer learning example with for image classification using the Cifar 10 dataset. TFlite is used for feature extractor and compared with KNN.
 205
 206 #### [Logistic Regression Example](https://github.com/nnstreamer/nntrainer/tree/main/Applications/LogisticRegression)
 207
 208 A logistic regression example using NNTrainer.
 209
 210 ## [Getting Started](https://github.com/nnstreamer/nntrainer/blob/main/docs/getting-started.md)
 211
 212 Instructions for installing NNTrainer.
 213
 214 ### [Running Examples](https://github.com/nnstreamer/nntrainer/blob/main/docs/how-to-run-examples.md)
 215
 216 Instructions for preparing NNTrainer for execution
 217
 218 ## Open Source License
 219
 220 The nntrainer is an open source project released under the terms of the Apache License version 2.0.
 221
 222 ## Contributing
 223
 224 Contributions are welcome! Please see our [Contributing](https://github.com/nnstreamer/nntrainer/blob/main/docs/contributing.md) Guide for more details.
 225
 226 [![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/0)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/0)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/1)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/1)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/2)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/2)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/3)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/3)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/4)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/4)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/5)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/5)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/6)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/6)[![](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/images/7)](https://sourcerer.io/fame/dongju-chae/nnstreamer/nntrainer/links/7)