Towards Distortion-Predictable Embedding of Neural Networks.

This repository contains the work for the aforementioned master project in the
labCV (Computer Vision Lab) at EPFL (Switzerland) by Axel Angel in 2015. This
work is released under the GPLv3 license, see LICENSE.

Read the thesis.pdf for the entire text.

The repository is published to:
 (1) Provide tools for datasets generation and visualisation with the Caffe deep
     learning framework and the Python programming language.
 (2) Demonstrate a new state-of-the-art method to control embedding of neural
     networks without imposing hard constraint (unlike regression).
 (3) Provide models, results and all tools to reproduce the work of this master
     project with minor efforts.

The directories are structured as follows.

The work is separated by datasets into three folders containing model definition
prototxt, the model snapshots, dataset in npz for Python and Python scripts.

 * lenet_mnist: t-SNE applied on MNIST with LeNet (for comparison).
 * siamese_mnist: our method applied to MNIST.
 * siamese_norb: our method applied to NORB.

Other folders are:

 * caffe: submodule for Caffe (dependency of this project).
 * deps_caffe: Debian dependency for Caffe.
 * links: useful links for Caffe and Deep Learning.
 * papers: references papers.
 * pres: contains the final presentation slides.
 * report: contains the final report.
 * src: scripts for handling datasets, computing measures, etc.
 * todo: personal notes

 * caffenet_imagenet: unused directory.
 * housenumbers: unused directory.