High-Level Perceptual Similarity is Enabled by Learning Diverse Tasks

TL;DR

This paper demonstrates that human perceptual image similarity can be effectively predicted by combining representations learned from diverse visual tasks, without additional training, achieving state-of-the-art results on the TLL benchmark.

Contribution

It shows that perceptual similarity is a byproduct of learning multiple visual tasks and that simple feature concatenation from these tasks outperforms recent methods.

Findings

Significantly surpasses recent baselines on TLL benchmark

Highlights the emergent nature of perceptual similarity from diverse visual representations

Suggests that perceptual similarity can be predicted without explicit training for that task

Abstract

Predicting human perceptual similarity is a challenging subject of ongoing research. The visual process underlying this aspect of human vision is thought to employ multiple different levels of visual analysis (shapes, objects, texture, layout, color, etc). In this paper, we postulate that the perception of image similarity is not an explicitly learned capability, but rather one that is a byproduct of learning others. This claim is supported by leveraging representations learned from a diverse set of visual tasks and using them jointly to predict perceptual similarity. This is done via simple feature concatenation, without any further learning. Nevertheless, experiments performed on the challenging Totally-Looks-Like (TLL) benchmark significantly surpass recent baselines, closing much of the reported gap towards prediction of human perceptual similarity. We provide an analysis of these results and discuss them in a broader context of emergent visual capabilities and their implications on the course of machine-vision research.