MXNet Examples

This page contains a curated list of awesome MXNet examples, tutorials and blogs. It is inspired by awesome-php and awesome-machine-learning. See also Awesome-MXNet for a similar list.

• Contributing
• List of examples
  • Languages Binding Examples
  • Deep Learning Examples in the MXNet Project Repository
  • Other Deep Learning Examples with MXNet
  • IPython Notebooks
  • Mobile App Examples
  • Web Predictive Services
• List of tutorials
  • GPU Technology Conference 2016 Hands-on session
  • Deep learning for hackers with MXnet
  • MXnet setup on AWS
  • Kaggle tutorials
  • Learning Note
• Machine Learning Challenge Winning Solutions
• Tools with MXnet

Contributing

If you want to contribute to this list and the examples, please open a new pull request.

Examples

Example applications or scripts should be submitted in this example folder.

Tutorials

If you have a tutorial idea for the website, download the Jupyter notebook tutorial template.

Tutorial location

Notebook tutorials should be submitted in the docs/tutorials folder, so that they maybe rendered in the web site's tutorial section.

Do not forget to udpdate the docs/tutorials/index.md for your tutorial to show up on the website.

Tutorial formatting

The site expects the format to be markdown, so export your notebook as a .md via the Jupyter web interface menu (File > Download As > Markdown). Then, to enable the download notebook button in the web site's UI (example), add the following as the last line of the file (example):

<!-- INSERT SOURCE DOWNLOAD BUTTONS -->

If you want some lines to show-up in the markdown but not in the generated notebooks, add this comment <!--notebook-skip-line--> after your ![png](img_url). Like this:

![png](img_url.png)<!--notebook-skip-line-->

Typically when you have a plt.imshow() you want the image tag [png](img.png) in the .md but not in the downloaded notebook as the user will re-generate the plot at run-time.

Tutorial tests

As part of making sure all our tutorials are running correctly with the latest version of MXNet, each tutorial is run automatically through a python2 and python3 jupyter notebook kernel in the CI, in a GPU environment, checking for errors and warnings.

Add your own test here tests/tutorials/test_tutorials.py. (If you forget, don't worry your PR will not pass the sanity check).

If your tutorial depends on specific packages, simply add them to this provisionning script: ci/docker/install/ubuntu_tutorials.sh

List of examples

Languages Binding Examples

• MXNet C++ API
  • C++ examples - Example code for using C++ interface, including NDArray, symbolic layer and models.
• MXNet Python API
• MXNet Java API
• MXNet Scala API
• MXNet R API
• MXNet Julia API
• MXNet Perl API
• go-mxnet-predictor - Go binding for inference
• MXNet JNI - JNI(Android) library
• MXNet Amalgamation - Amalgamation (entire library in a single file)
• MXNet Javascript - MXNetJS: Javascript Package for Deep Learning in Browser (without server)

Deep Learning Examples in the MXNet Project Repository

• Autoencoder - unsupervised feature learning
• Bayesian Methods - various examples related to Bayesian Methods
• Bidirectional LSTM Sorting - use a bidirectional LSTM to sort an array
• Caffe - how to call Caffe operators from MXNet
• CNN for Chinese Text Classification - a MXnet example for Chinese text classification
• CNN for Text Classification - a MXnet example for text classification
• CTC with MXNet - a modification of warpctc
• Deep Embedded Clustering - unsupervised deep embedding for clustering analysis
• Dense-Sparse-Dense Training - Dense-Sparse-Dense Training for deep neural networks
• Fully Convolutional Networks - fully convolutional networks for semantic segmentation
• Generative Adversarial Networks with R - GAN examples in R
• Gluon Examples - several examples using the Gluon API
  • Style Transfer - a style transfer example using gluon
  • Word Language Model - an example that trains a multi-layer RNN on the Penn Treebank language modeling benchmark
  • SN-GAN - an example that utilizes spectral normalization to train GAN(Generative adversarial network) using Gluon API
• Image Classification with R - image classification on MNIST,CIFAR,ImageNet-1k,ImageNet-Full, with multiple GPU and distributed training.
• Kaggle 1st national data science bowl - a MXnet example for Kaggle Nation Data Science Bowl 1
• Kaggle 2nd national data science bowl - a tutorial for Kaggle Second Nation Data Science Bowl
• Memory Cost - a script to show the memory cost of different allocation strategies
• Model Parallelism - various model parallelism examples
  • Model Parallelism with LSTM - an example showing how to do model parallelism with a LSTM
  • Model Parallelism with Matrix Factorization - a matrix factorization algorithm for recommendations
• Module API - examples with the Python Module API
• Multi-task Learning - how to use MXNet for multi-task learning
• MXNet Adversarial Variational Autoencoder - combines a variational autoencoder with a generative adversarial network
• Noise-contrastive estimation loss - used to speedup multi-class classification
• Neural Style - use deep learning for style transfer in images
• Numpy Operator Customization - Examplea on quick customize new ops with Numpy
• Profiling - generate profiling results in json files
• Quantization and Calibration Examples - examples of quantizing a FP32 model to INT8 and performing low-precision inference with Intel MKL-DNN on CPU or cuDNN on GPU
• R-CNN - R-CNN with distributed implementation and data parallelization
• Recommender Systems - examples of how to build various kinds of recommender systems
• Reinforcement Learning - a variety of reinforcement learning examples
  • A3C
  • DDPG - example of training DDPG for CartPole
  • DQN - examples of training DQN and Double DQN to play Atari Games
  • Parallel Advantage-Actor Critic
• Restricted Boltzmann Machine - an example of the binary restricted Boltzmann machine learning MNIST
• RNN Time Major - RNN implementation with Time-major layout
• Recurrent Neural Net - creating recurrent neural networks models using high level mxnet.rnn interface
• Sparse - a variety of sparse examples
  • Factorization Machine
  • Linear Classification
  • Matrix Factorization
  • Wide Deep
• Single Shot MultiBox Detector - SSD object recognition example
• Stochastic Depth - implementation of the stochastic depth algorithm
• Support Vector Machine - an SVM example using MNIST
• Variational Auto Encoder - implements the Variational Auto Encoder in MXNet using MNIST

Other Deep Learning Examples with MXNet

• Face Recognition with ArcFace - ONNX model for face recognition with notebooks for training, validating and running inference in MXNet by abhinavs95
• Chinese plate recognition - Recognize Chinese vehicle plate, by imistyrain
• Fast R-CNN by Jian Guo
• “End2End Captcha Recognition (OCR)” by xlvector github link Blog in Chinese
• “Prediction step of xlvector's lstm ocr” by melody-rain github link Blog in Chinese
• “Solving classification + regression with MXnet in Multi Input + Multi Obj” by xlvector github link Blog in Chinese
• “Learn to sort by LSTM” by xlvector github link Blog in Chinese
• Neural Art using extremely lightweight (<500K) neural network Lightweight version of mxnet neural art implementation by Pavel Gonchar
• Neural Art with generative networks by zhaw
• Faster R-CNN in MXNet with distributed implementation and data parallelization
• Asynchronous Methods for Deep Reinforcement Learning in MXNet by zmonoid
• Deep Q-learning in MXNet by zmonoid
• Face Detection with End-to-End Integration of a ConvNet and a 3D Model (ECCV16) by tfwu, source code for paper Yunzhu Li, Benyuan Sun, Tianfu Wu and Yizhou Wang, “Face Detection with End-to-End Integration of a ConvNet and a 3D Model”, ECCV 2016 https://arxiv.org/abs/1606.00850
• End-to-End Chinese plate recognition base on MXNet by szad670401
• Reproduce ResNet-v2 (Identity Mappings in Deep Residual Networks) using MXNet by tornadomeet
• Learning similarity among images in MXNet by xlvector in Chinese. Github link
• Matrix decomposition (SVD) with MXNet by xlvector in Chinese. Github link
• MultiGPU enabled image generative models (GAN and DCGAN) by Tianqi Chen
• Deep reinforcement learning for playing flappybird by mxnet by LIHaoran
• Neural Style in Markov Random Field (MRF) and Perceptual Losses Realtime transfer by zhaw
• MTCNN Face keypoints detection and alignment by yuanyang, source code for paper “Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks”, Kaipeng Zhang, Zhanpeng Zhang, Zhifeng Li and Yu Qiao, IEEE Signal Processing Letters, 23(10), 2016
• SSD: Single Shot MultiBox Object Detector by zhreshold
• Fast Neural Style in Scala by Ldpe2G
• LSTM Human Activity Recognition by Ldpe2G
• Visual Question Answering by liuzhi136
• Deformable ConvNets (github) by MSRACVer
• OCR with bi-LSTM and CTC Loss in Gluon by ThomasDelteil
• Visual Search with Gluon and HNSWlib, by ThomasDelteil, online demo here
• MXNet-face - Using MXNet for a face-related algorithm by tornadomeet where the single model gets 97.13%+-0.88% accuracy on LFW, and with only 20MB size

IPython Notebooks

• Predict with Pre-trained model - Notebook on how to predict with pretrained model.
• composite symbol - A demo of how to composite a symbolic Inception-BatchNorm Network
• cifar-10 recipe - A step by step demo of how to use MXNet
• cifar-100 - A demo of how to train a 75.68% accuracy CIFAR-100 model
• simple bind - A demo of low level training API.
• Multi task tutorial - A demo of how to train and predict multi-task network on both MNIST and your own dataset.
• class active maps - A demo of how to localize the discriminative regions in an image using global average pooling (GAP) in CNNs.
• DMLC MXNet Notebooks DMLC's repo for various notebooks ranging from basic usages of MXNet to state-of-the-art deep learning applications.
• AWS Seoul Summit 2017 Demos The demo codes and ipython notebooks in AWS Seoul Summit 2017.
• Character-level CNN for text classification Performing category classification on Amazon reviews using Gluon and character-level Convolutional Neural Networks. Online demo here

Mobile App Examples

• MXNet Android Classification App - Image classification on Android with MXNet.
• MXNet iOS Classification App - Image classification on iOS with MXNet.
• Compile MXnet on Xcode (in Chinese) - a step-by-step tutorial of compiling MXnet on Xcode for iOS app

Web Predictive Services

• MXNet Shinny - Source code for quickly creating a Shiny R app to host online image classification.
• Machine Eye - Web service for local image file/image URL classification without uploading.

List of tutorials

GPU Technology Conference 2016 Hands-on session

• Video on GTC 2016 site
• Video backup in Mainland China
• iPython Notebook

Deep learning for hackers with MXNet

• Deep learning for hackers with MXNet (1) GPU installation and MNIST English Chinese - a tutorial of installing MXnet with GPU and introduction to deep learning by MNIST example.
• Deep learning for hackers with MXNet (2): Neural art English Chinese - a tutorial of generating Van Gogh style cat paintings.

MXNet on the cloud

• Setup Amazon AWS GPU instance with MXnet - AWS GPU instance setup with GPU (CUDA with latest cuDNN and S3 support)
• Intro Guide to AWS (MXNet with Julia) - A step-by-step guide of using spot instances with Amazon Web Services (AWS) to help you save money when training DSB models on MXNet by Mike Kim
• Building Deep Neural Networks in the Cloud with Azure GPU VMs, MXNet and Microsoft R Server by Cortana Intelligence and ML Blog Team at Microsoft
• Applying Deep Learning at Cloud Scale, with Microsoft R Server & Azure Data Lake by Cortana Intelligence and ML Blog Team at Microsoft
• Training Deep Neural Neural Networks on ImageNet Using Microsoft R Server and Azure GPU VMs by Cortana Intelligence and ML Blog Team at Microsoft
• Cloud-Scale Text Classification with Convolutional Neural Networks on Microsoft Azure by Cortana Intelligence and ML Blog Team at Microsoft
• Distributed Deep Learning Made Easy at AWS/Amazon for deploying deep learning clusters using MXNet

Kaggle tutorials

• Kaggle 2nd Annual Data Science Bowl End-to-End Deep Learning Tutorial (Python) - an end-to-end python tutorial for Kaggle heart disease diagnose competition (public leaderboard score 0.0392)
• Kaggle 2nd Annual Data Science Bowl End-to-End Deep Learning Tutorial (R) - an end-to-end R tutorial for Kaggle heart disease diagnose competition
• Dogs vs. Cats classification with mxnet and R - end-to-end (not winning) tutorial with an example of fine-tuning in R

Learning Note

• Learning Note in Chinese - MXNet learning note in Chinese.
• Getting Started with MXNet by indico.io (Chinese Translation MXNet实践 by 侠天 )
• {mxnet} R package from MXnet, an intuitive Deep Learning framework including CNN & RNN by TJO
• MXnet with R: combined power of deep learning in Chinese by Tong He
• Understand MXNet dependency engine in Chinese by Yu Yang

Machine Learning Challenge Winning Solutions

• Dmitrii Tsybulevskii, 1st place of the Yelp Restaurant Photo Classification. Link to the Kaggle interview.

Tools with MXnet

• TensorFuse - Common interface for Theano, CGT, TensorFlow, and mxnet (experimental) by dementrock
• MXnet-face - Using mxnet for face-related algorithm by tornadomeet where the single model get 97.13%+-0.88% accuracy on LFW, and with only 20MB size.
• MinPy - Pure numpy practice with third party operator Integration and MXnet as backend for GPU computing
• MXNet Model Server - a flexible and easy to use tool for serving Deep Learning models