Research Theme

Fashion AI

We use AI to structure how a person's appearance comes across, supporting fashion analysis and suggestions for flattering styles. Our goal is to back up analysis that tends to rely on subjective judgment with explanations and evidence.

Fashion AILLMClassification support

Overview

What we work on

Use AI to structure features such as facial features, body frame, and the impression of lines.
Develop with attention to how results should be presented so that people find them convincing.
Use AI as a second opinion rather than a replacement for human judgment.

Papers

Applied AI / Fashion AI

Supporting fashion analysis with AI

Holistic Fashion Classification with Structured LLM Semantics and a Two-Dimensional YIN/YANG Theory

This study uses LLMs to structure a person's facial features, body frame, and the impression of lines, making fashion classification based on YIN/YANG theory easier to reproduce.

Fashion AILLMYIN/YANG theoryClassification support

View paper (PDF) View poster

Show details Hide details

Problem

Fashion analysis depends heavily on expert experience, and different analysts may reach different results for the same person. The challenge is how to support such subjective judgment.

Approach

We reorganized the conventional one-dimensional YIN/YANG classification into two axes: quality of line and maturity. Instead of asking the LLM to classify appearance directly, we have it produce structured descriptions of face shape, body frame, curvature, impression of maturity, and so on.

Key results

Evaluated on data for 100 celebrities, a configuration using GPT-5-mini and kNN achieved 55.00% Top-1 accuracy and a weighted F1 of 56.44. The results point toward using the system as a second opinion rather than a replacement for human judgment.

Ongoing Research

Research topics the lab is working on as of the 2026 academic year.

Fashion AI that suggests clothes that suit you

We are building an AI that extends YIN/YANG theory to two dimensions and classifies appearance into eight types, and integrating it with a virtual try-on AI, aiming for a system that goes seamlessly from taking a photo to analysis and trying on outfits.

Background

Fashion analysis judges which styles suit a person from their facial features, body frame, and overall impression, but in full-body analysis, results often differ between analysts, making objective classification difficult.

Fashion AIYIN/YANG theoryVirtual try-on