FashionReGen: LLM-Empowered Fashion Report Generation

Yujuan Ding, Yunshan Ma, Wenqi Fan, Yige Yao, Tat-Seng Chua, Qing Li

Abstract

Fashion analysis refers to the process of examining and evaluating trends, styles, and elements within the fashion industry to understand and interpret its current state, generating fashion reports. It is traditionally performed by fashion professionals based on their expertise and experience, which requires high labour cost and may also produce biased results for relying heavily on a small group of people. In this paper, to tackle the Fashion Report Generation (FashionReGen) task, we propose an intelligent Fashion Analyzing and Reporting system based the advanced Large Language Models (LLMs), debbed as GPT-FAR. Specifically, it tries to deliver FashionReGen based on effective catwalk analysis, which is equipped with several key procedures, namely, catwalk understanding, collective organization and analysis, and report generation. By posing and exploring such an open-ended, complex and domain-specific task of FashionReGen, it is able to test the general capability of LLMs in fashion domain. It also inspires the explorations of more high-level tasks with industrial significance in other domains.

Introduction

The fashion industry moves at breakneck speed, making trend forecasting essential for brands and consumers. Traditionally, fashion reports are produced manually by experts who review thousands of runway looks to identify shifts in styles, silhouettes, and fabrics. This manual process is expensive, slow, and prone to subjective bias. Automating this task—Fashion Report Generation (FashionReGen)—requires overcoming challenges in fine-grained visual recognition, domain-specific statistical analysis, and multimodal synthesis.

Methodology: The GPT-FAR System

The authors propose GPT-FAR (Fashion Analyzing and Reporting), a first-of-its-kind automated pipeline that transforms raw catwalk images into professional-grade analysis reports. The system operates in three distinct stages: Catwalk Understanding, Collective Organization and Analysis, and Report Generation.

The three-stage GPT-FAR pipeline: Catwalk Understanding, Collective Organization and Analysis, and Fashion Report Generation.

1. Catwalk Understanding and Tag Cleaning

In the first stage, the system uses GPT-4V to perform fine-grained garment tagging, identifying categories and attributes (e.g., necklines, fabrics) based on professional standards like WGSN. A critical challenge is that LLMs often produce inconsistent labels (e.g., "draped" vs. "draping"). To resolve this, the authors developed a two-stage tag cleaning process.

The two-stage tag cleaning process involving manual format unification and an iterative automatic synonym emergence step to create a unified tag dictionary.

This iterative process uses GPT-4 to group synonymous tags into a unified dictionary, converting qualitative "vibes" into structured variables suitable for statistical math.

2. Collective Organization and Analysis

The system aggregates individual garment tags into collection-level data. It calculates professional metrics such as Mix (the proportion of a specific style), Year-on-Year (YoY) index (growth or decline of a trend), and Evolving Trend lists.

3. Fashion Report Generation

The final stage synthesizes these insights into a professional layout. The system automatically plots statistical charts and retrieves representative images. GPT-4V then generates professional prose by "reading" these charts through in-context learning, ensuring the tone matches industry standards while preventing hallucinations.

Key Insights and Results

The researchers evaluated GPT-FAR using a dataset of catwalk images from 2019–2023, focusing on major brands like Chanel, Dior, and Louis Vuitton.

  • Professional Prose: The system successfully generated multi-page reports with "reasonable, fluent, and professional" text, mimicking industry leaders. For instance, it correctly identified that "Dresses and Skirts marked their territory as frontrunners... accounting for over a third of the market."
  • Efficiency: The pipeline significantly reduces the labor and intellectual cost required to synthesize thousands of runway looks into a coherent narrative.
  • Grounding through Data: A major insight was that providing the LLM with specific data points from charts (in-context learning) is essential to ground the model and ensure accurate trend interpretation.

Conclusion and Future Work

GPT-FAR demonstrates the potential of LLMs to automate high-level, domain-specific tasks in the fashion industry. Future work will focus on expanding data diversity—incorporating social media and sales data—and developing objective technical metrics to evaluate the "intelligence" and accuracy of automated fashion insights.

Summary updated

Abstract

Fashion analysis refers to the process of examining and evaluating trends, styles, and elements within the fashion industry to understand and interpret its current state, generating fashion reports. It is traditionally performed by fashion professionals based on their expertise and experience, which requires high labour cost and may also produce biased results for relying heavily on a small group of people. In this paper, to tackle the Fashion Report Generation (FashionReGen) task, we propose an intelligent Fashion Analyzing and Reporting system based the advanced Large Language Models (LLMs), debbed as GPT-FAR. Specifically, it tries to deliver FashionReGen based on effective catwalk analysis, which is equipped with several key procedures, namely, catwalk understanding, collective organization and analysis, and report generation. By posing and exploring such an open-ended, complex and domain-specific task of FashionReGen, it is able to test the general capability of LLMs in fashion domain. It also inspires the explorations of more high-level tasks with industrial significance in other domains. Video illustration and more materials of GPT-FAR can be found in https://github.com/CompFashion/FashionReGen.
BibTeX
@inproceedings{ding2024fashionregen,
  author = {Yujuan Ding and Yunshan Ma and Wenqi Fan and Yige Yao and Tat-Seng Chua and Qing Li},
  title = {{FashionReGen: LLM-Empowered Fashion Report Generation}},
  booktitle = {The Web Conference},
  year = {2024},
  doi = {10.1145/3589335.3651232},
}
Figure