tensorflow

Collection of examples for learning TensorFlow

Introduction to TensorFlow

The following notebooks give an introduction to TensorFlow and are partly covered in the introductory presentation

• linear_regression/01_LineFit.ipynb A simple line fit
• simple_ops/Mandelbrot.ipynb Drawing the Madelbrot set using TensorFlow
• Control_Flow/Mandelbrot.ipynb Explains tensorflow loops and applies them to draw the Madelbrot set

Advanced TensorFlow

The following notebooks deal with more deep learning relate aspects of TF and are partly convered in this presentation

Building networks, accessing them

• Building_Nice_Networks/Scoping.ipynb Use variable scopes to build networks from building blocks.

• linear_regression/03_checkpointing.ipynb demonstrates how to checkpoint and reload a model

• stored_models/Using_Trained_Nets.ipynb Load a trained network, invastigate the network to find relevant entry / exit points. Use these to feed data through the network (for classification of novel images). The notebook further shows how to get variables like the kernels from a CNN from the network.

• stored_models/Loading_Frozen_Graph.ipynb Artstyle transfer: loading a frozen model, which does art style transfer

• stored_models/Finetuning.ipynb Setting up a network with TF-Slim and do transfer learning

Debugging networks

• linear_regression/02_Inspecting_the_graph.ipynb How to use summaries and visualize them in tensorboard

• debugging/print Explanation of the tf.Print() function

• debugging/debug_with_python.ipynb Debugging by embedding python code

Deep Learning: The path to fully connected networks

• path_to_fc_nets These notebooks cover the path to fully connected neural networks starting from logistic regression

Deep Learning: Recurent Neural Networks

• RNN Some scripts on recurrent neural networks covered in the talk

Deep Learning: Convolutional Neural Networks

• CAS Course Will be ported in this directory soon

Miscelaneous

• vae contains examples of Variational Autoencoder VAE

• faces shows how to transfer a network for face recognition from caffee to tensorflow

• inception_cifar10 Shows how to extract features from a network trained on ImageNet and applies them on CIFAR_10.