s2t2/keras-resources

Directory of tutorials and open-source code repositories for working with Keras, the Python deep learning library

Keras resources

This is a directory of tutorials and open-source code repositories for working with Keras, the Python deep learning library.

If you have a high-quality tutorial or project to add, please open a PR.

Official starter resources

• keras.io - Keras documentation
• Getting started with the Sequential model
• Getting started with the functional API
• Keras FAQ

Tutorials

• Quick start: the Iris dataset in Keras and scikit-learn
• Using pre-trained word embeddings in a Keras model
• Building powerful image classification models using very little data
• Building Autoencoders in Keras
• A complete guide to using Keras as part of a TensorFlow workflow
• Introduction to Keras, from University of Waterloo: video - slides
• Introduction to Deep Learning with Keras, from CERN: video - slides
• Installing Keras for deep learning
• Develop Your First Neural Network in Python With Keras Step-By-Step
• Practical Neural Networks with Keras: Classifying Yelp Reviews (Shows basic classification and how to set up a GPU instance on AWS)
• Understanding Stateful LSTM Recurrent Neural Networks in Python with Keras
• Time Series Prediction with LSTM Recurrent Neural Networks in Python with Keras
• Keras video tutorials from Dan Van Boxel
• Keras Deep Learning Tutorial for Kaggle 2nd Annual Data Science Bowl
• Collection of tutorials setting up DNNs with Keras
• Fast.AI - Practical Deep Learning For Coders, Part 1 (great information on deep learning in general, heavily uses Keras for the labs)
• Keras Tutorial: Content Based Image Retrieval Using a Convolutional Denoising Autoencoder
• A Bit of Deep Learning and Keras: a multipart video introduction to deep learning and keras
• Five simple examples of the Keras Functional API
• Keras video tutorials playlist from Deeplizard

Books based on Keras

• Deep Learning with Python
• Deep Learning with Keras
• Deep Learning and the Game of Go (MEAP)

Code examples

Working with text

• Reuters topic classification
• LSTM on the IMDB dataset (text sentiment classification)
• Bidirectional LSTM on the IMDB dataset
• 1D CNN on the IMDB dataset
• 1D CNN-LSTM on the IMDB dataset
• LSTM-based network on the bAbI dataset
• Memory network on the bAbI dataset (reading comprehension question answering)
• Sequence to sequence learning for performing additions of strings of digits
• LSTM text generation
• Using pre-trained word embeddings
• Monolingual and Multilingual Image Captioning
• FastText on the IMDB dataset
• Structurally constrained recurrent nets text generation
• Character-level convolutional neural nets for text classification
• LSTM to predict gender of a name
• Language/dialect identification with multiple character-level CNNs

Working with images

• Simple CNN on MNIST
• Simple CNN on CIFAR10 with data augmentation
• Inception v3
• VGG 16 (with pre-trained weights)
• VGG 19 (with pre-trained weights)
• ResNet 50 (with pre-trained weights): 1 - 2
• FractalNet
• AlexNet, VGG 16, VGG 19, and class heatmap visualization
• Visual-Semantic Embedding
• Variational Autoencoder: with deconvolutions - with upsampling
• Visual question answering
• Deep Networks with Stochastic Depth
• Smile detection with a CNN
• VGG-CAM
• t-SNE of image CNN fc7 activations
• VGG16 Deconvolution network
• Wide Residual Networks (with pre-trained weights): 1 - 2
• Ultrasound nerve segmentation: 1 - 2
• DeepMask object segmentation
• Densely Connected Convolutional Networks: 1 - 2
• Snapshot Ensembles: Train 1, Get M for Free
• Single Shot MultiBox Detector
• Popular Image Segmentation Models : FCN, Segnet, U-Net etc.

Creative visual applications

• Real-time style transfer
• Style transfer: 1 - 2
• Image analogies: Generate image analogies using neural matching and blending.
• Visualizing the filters learned by a CNN
• Deep dreams
• GAN / DCGAN: 1 - 2 - 3 - 4
• InfoGAN
• pix2pix
• DFI: Deep Feature Interpolation
• Colorful Image colorization: B&W to color

Reinforcement learning

• DQN
• FlappyBird DQN
• async-RL: Tensorflow + Keras + OpenAI Gym implementation of 1-step Q Learning from "Asynchronous Methods for Deep Reinforcement Learning"
• keras-rl: A library for state-of-the-art reinforcement learning. Integrates with OpenAI Gym and implements DQN, double DQN, Continuous DQN, and DDPG.

Miscallenous architecture blueprints

• Stateful LSTM
• Siamese network
• Pretraining on a different dataset
• Neural programmer-interpreter

Third-party libraries

• Elephas: Distributed Deep Learning with Keras & Spark
• Hyperas: Hyperparameter optimization
• Hera: in-browser metrics dashboard for Keras models
• Kerlym: reinforcement learning with Keras and OpenAI Gym
• Qlearning4K: reinforcement learning add-on for Keras
• seq2seq: Sequence to Sequence Learning with Keras
• Seya: Keras extras
• Keras Language Modeling: Language modeling tools for Keras
• Recurrent Shop: Framework for building complex recurrent neural networks with Keras
• Keras.js: Run trained Keras models in the browser, with GPU support
• keras-vis: Neural network visualization toolkit for keras.

Projects built with Keras

• RocAlphaGo: An independent, student-led replication of DeepMind's 2016 Nature publication, "Mastering the game of Go with deep neural networks and tree search"
• BetaGo: Deep Learning Go bots using Keras
• DeepJazz: Deep learning driven jazz generation using Keras
• dataset-sts: Semantic Text Similarity Dataset Hub
• snli-entailment: Independent implementation of attention model for textual entailment from the paper "Reasoning about Entailment with Neural Attention".
• Headline generator: independent implementation of Generating News Headlines with Recurrent Neural Networks
• LipNet: independent implementation of LipNet: End-to-End Sentence-level Lipreading.
• caption_generator: An implementation of image caption generation in natural language inspired from Show and Tell: A Neural Image Caption Generator.
• NMT-Keras: Neural Machine Translation using Keras.
• Conx - easy-to-use layer on top of Keras, with visualizations (eg, no knowledge of numpy needed)