Scaling Up Personalized Aesthetic Assessment via Task Vector Customization
Jooyeol Yun, Jaegul Choo
2024-07-13
Summary
This paper discusses a new method for personalizing how we assess the aesthetics of images, which means figuring out how visually appealing they are based on individual preferences. The authors propose a way to use existing databases to create more adaptable and effective models for predicting aesthetic scores.
What's the problem?
Current methods for assessing image aesthetics rely heavily on expensive and carefully curated datasets, which limits their ability to scale and adapt to different user preferences. This makes it challenging to create personalized models that can accurately reflect individual tastes in various contexts.
What's the solution?
To tackle this issue, the authors introduce a method that utilizes readily available databases for general image aesthetic assessment. They treat each database as a separate task with different levels of personalization potential. By finding the best combinations of task vectors—representing specific traits of each database—they can build personalized models that work well for individuals. This approach allows them to use a lot of data effectively, leading to better generalization across different scenarios.
Why it matters?
This research is significant because it provides a scalable solution for personalized image aesthetic assessment, making it easier to tailor aesthetic evaluations to individual preferences. By improving how we assess image aesthetics, this method can enhance applications in areas like photography, social media, and design, ultimately helping people find and create images that align with their unique tastes.
Abstract
The task of personalized image aesthetic assessment seeks to tailor aesthetic score prediction models to match individual preferences with just a few user-provided inputs. However, the scalability and generalization capabilities of current approaches are considerably restricted by their reliance on an expensive curated database. To overcome this long-standing scalability challenge, we present a unique approach that leverages readily available databases for general image aesthetic assessment and image quality assessment. Specifically, we view each database as a distinct image score regression task that exhibits varying degrees of personalization potential. By determining optimal combinations of task vectors, known to represent specific traits of each database, we successfully create personalized models for individuals. This approach of integrating multiple models allows us to harness a substantial amount of data. Our extensive experiments demonstrate the effectiveness of our approach in generalizing to previously unseen domains-a challenge previous approaches have struggled to achieve-making it highly applicable to real-world scenarios. Our novel approach significantly advances the field by offering scalable solutions for personalized aesthetic assessment and establishing high standards for future research. https://yeolj00.github.io/personal-projects/personalized-aesthetics/