How Much Position Information Do Convolutional Neural Networks Encode?

TL;DR

This paper investigates how much absolute positional information deep CNNs encode, revealing they surprisingly retain significant position data, which has implications for understanding their internal representations.

Contribution

It provides empirical evidence that CNNs encode substantial absolute position information and explores how and where this information is represented within the networks.

Findings

CNNs encode significant absolute positional information

Position information is distributed across multiple layers

Implications for interpretability and design of CNNs

Abstract

In contrast to fully connected networks, Convolutional Neural Networks (CNNs) achieve efficiency by learning weights associated with local filters with a finite spatial extent. An implication of this is that a filter may know what it is looking at, but not where it is positioned in the image. Information concerning absolute position is inherently useful, and it is reasonable to assume that deep CNNs may implicitly learn to encode this information if there is a means to do so. In this paper, we test this hypothesis revealing the surprising degree of absolute position information that is encoded in commonly used neural networks. A comprehensive set of experiments show the validity of this hypothesis and shed light on how and where this information is represented while offering clues to where positional information is derived from in deep CNNs.