Understanding Deep Image Representations by Inverting Them

TL;DR

This paper introduces a framework to invert various image representations, including CNNs, revealing how much visual information they retain and how invariant they are to geometric and photometric changes.

Contribution

It presents a novel, general method for inverting image representations, enabling analysis of the information preserved in CNNs and other features.

Findings

CNN layers retain photographically accurate information

Different CNN layers show varying degrees of invariance

The method outperforms previous approaches in inverting HOG and SIFT

Abstract

Image representations, from SIFT and Bag of Visual Words to Convolutional Neural Networks (CNNs), are a crucial component of almost any image understanding system. Nevertheless, our understanding of them remains limited. In this paper we conduct a direct analysis of the visual information contained in representations by asking the following question: given an encoding of an image, to which extent is it possible to reconstruct the image itself? To answer this question we contribute a general framework to invert representations. We show that this method can invert representations such as HOG and SIFT more accurately than recent alternatives while being applicable to CNNs too. We then use this technique to study the inverse of recent state-of-the-art CNN image representations for the first time. Among our findings, we show that several layers in CNNs retain photographically accurate information about the image, with different degrees of geometric and photometric invariance.