Identifying and Mitigating Flaws of Deep Perceptual Similarity Metrics

TL;DR

This paper critically examines Deep Perceptual Similarity metrics used in image comparison, revealing their flaws and proposing improvements through in-depth analysis and specially designed challenging images.

Contribution

It provides a detailed analysis of DPS metrics' strengths and weaknesses, offering new insights and potential improvements to enhance their reliability.

Findings

DPS metrics have notable flaws when comparing certain challenging images.

A comprehensive analysis reveals specific weaknesses in spatially-aware deep feature comparisons.

Proposed improvements aim to address identified flaws in DPS metrics.

Abstract

Measuring the similarity of images is a fundamental problem to computer vision for which no universal solution exists. While simple metrics such as the pixel-wise L2-norm have been shown to have significant flaws, they remain popular. One group of recent state-of-the-art metrics that mitigates some of those flaws are Deep Perceptual Similarity (DPS) metrics, where the similarity is evaluated as the distance in the deep features of neural networks. However, DPS metrics themselves have been less thoroughly examined for their benefits and, especially, their flaws. This work investigates the most common DPS metric, where deep features are compared by spatial position, along with metrics comparing the averaged and sorted deep features. The metrics are analyzed in-depth to understand the strengths and weaknesses of the metrics by using images designed specifically to challenge them. This work contributes with new insights into the flaws of DPS, and further suggests improvements to the metrics. An implementation of this work is available online: https://github.com/guspih/deep_perceptual_similarity_analysis/