What Do Vision-Language Models Encode for Personalized Image Aesthetics Assessment?

TL;DR

This paper investigates how vision-language models encode aesthetic attributes and demonstrates their potential for personalized image aesthetics assessment without fine-tuning.

Contribution

It reveals that VLMs encode diverse aesthetic attributes useful for personalization and shows that simple linear models can effectively perform PIAA.

Findings

VLMs encode diverse aesthetic attributes across layers.

Simple linear models can perform PIAA effectively.

Aesthetic information transfer varies across architectures and domains.

Abstract

Personalized image aesthetics assessment (PIAA) is an important research problem with practical real-world applications. While methods based on vision-language models (VLMs) are promising candidates for PIAA, it remains unclear whether they internally encode rich, multi-level aesthetic attributes required for effective personalization. In this paper, we first analyze the internal representations of VLMs to examine the presence and distribution of such aesthetic attributes, and then leverage them for lightweight, individual-level personalization without model fine-tuning. Our analysis reveals that VLMs encode diverse aesthetic attributes that propagate into the language decoder layers. Building on these representations, we demonstrate that simple linear models can perform PIAA effectively. We further analyze how aesthetic information is transferred across layers in different VLM architectures and across image domains. Our findings provide insights into how VLMs can be utilized for modeling subjective, individual aesthetic preferences. Our code is available at https://github.com/ynklab/vlm-latent-piaa.