XPASS-Vis: A Dataset for Cross-Domain Personalized Image Aesthetic Assessment

2026-06-14 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

XPASS-Vis is introduced as the first dataset specifically designed for cross-domain Personalized Image Aesthetic Assessment (PIAA), addressing the limitation of existing datasets confined to single domains or lacking sufficient per-annotator samples. This new dataset comprises 6,526 stimuli across three visual domains—art, fashion, and landscape—and features ratings from 129 annotators, resulting in 87,836 user-stimulus interactions. Each interaction includes an overall aesthetic score and nine aesthetic-emotion ratings. Crucially, each annotator provided over 200 ratings per domain, facilitating robust personalization both within and across domains. The research also establishes baseline models for cross-domain PIAA using unsupervised domain adaptation (UDA), demonstrating that the top-performing UDA method recovers approximately 60% (Spearman's ρ = .28) of the supervised upper bound, indicating meaningful transferability of preferences while highlighting a need for PIAA-specific adaptation strategies.

Key takeaway

For AI Scientists and Computer Vision Engineers developing personalized image aesthetic assessment models, XPASS-Vis offers a critical resource for exploring cross-domain transferability. You should leverage this dataset to train and evaluate models that generalize aesthetic preferences across diverse visual domains like art, fashion, and landscape. This enables more robust PIAA systems, but be aware that significant gaps remain, necessitating the development of PIAA-specific adaptation strategies to achieve supervised performance levels.

Key insights

Personalized aesthetic preferences are meaningfully transferable across visual domains, despite individual subjectivity.

Principles

• Aesthetic preference is deeply personal yet partially consistent.
• Cross-domain PIAA requires sufficient per-annotator samples.
• Transferability across domains is possible but incomplete.

Method

Baseline models for cross-domain PIAA are established via unsupervised domain adaptation (UDA), transferring models from a labeled source to an unlabeled target domain.

In practice

• Utilize XPASS-Vis for cross-domain PIAA research.
• Develop PIAA-specific domain adaptation strategies.
• Evaluate UDA approaches for aesthetic preference transfer.

Topics

• Personalized Image Aesthetic Assessment
• Cross-Domain Generalization
• XPASS-Vis Dataset
• Unsupervised Domain Adaptation
• Computer Vision
• Aesthetic Preference

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• Texture-Shape Bias Balancing for Robust Synthetic-to-Real Semantic Segmentation in Automotive NIR Imagery — A generative augmentation framework reduces the synthetic-to-real domain gap for automotive NIR semantic segmentation by balancing texture-shape bias.
• Beyond Absolute Scores: Relative Edit-induced Difference for Generalizable Image Aesthetic Assessment — Human aesthetic perception is dynamic and relative, not absolute, driven by subconscious comparisons and visual factors.
• A Multi-Modal Framework with Cross-Subject Pseudo-Labeling and Semantic Alignment for Micro-Gesture Recognition — A multi-modal framework uses pseudo-labeling and semantic alignment to improve micro-gesture recognition across subjects.
• CV-Arena: An Open Benchmark for Instructional Computer Vision Problem Solving with Human-AI Collaborative Preferences — CV-Arena offers a new benchmark and evaluation protocol for complex, instruction-guided image editing.
• Colour Extraction Pipeline for Odonates using Computer Vision — A computer vision pipeline segments Odonates and extracts color data to enable large-scale ecological correlation analysis.
• AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction — An integrated DL-LLM system accurately predicts individual image aesthetics by combining low-level and high-level feature extraction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.