A TensorFlow implementation of the paper AE2-Nets:Autoencoder in Autoencoder Networks, CVPR'2019

• TensorFlow 1.12.0
• Python 3.6.7
• sklearn
• numpy
• scipy
• h5py

Learning on data represented with multiple views (e.g.,multiple types of descriptors or modalities) is a rapidlygrowing direction in machine learning and computer vi-sion. Although effectiveness achieved, most existing algo-rithms usually focus on classification or clustering tasks.Differently, in this paper, we focus on unsupervised repre-sentation learning and propose a novel framework termedAutoencoder in Autoencoder Networks (AE2-Nets), whichintegrates information from heterogeneous sources into anintact representation by the nested autoencoder framework.The proposed method has the following merits: (1) ourmodel jointly performs view-specific representation learn-ing (with the inner autoencoder networks) and multi-viewinformation encoding (with the outer autoencoder network-s) in a unified framework; (2) due to the degradation pro-cess from the latent representation to each single view, ourmodel flexibly balances the complementarity and consis-tence among multiple views. The proposed model is effi-ciently solved by the alternating direction method (ADM),and demonstrates the effectiveness compared with state-of-the-art algorithms.

This repository contains a subset of the experiments mentioned in the paper.

python test_hand.py

If you find AE2-Nets helps your research, please cite our paper:

@InProceedings{Zhang_2019_CVPR,
author = {Zhang, Changqing and Liu, Yeqing and Fu, Huazhu},
title = {AE2-Nets: Autoencoder in Autoencoder Networks},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

For any additional questions, feel free to email [email protected].