FashionStylist: An Expert Knowledge-enhanced Multimodal Dataset for Fashion Understanding

Kaidong Feng; Zhuoxuan Huang; Huizhong Guo; Yuting Jin; Xinyu Chen; Yue Liang; Yifei Gai; Li Zhou

cs.CVApr 10, 2026

FashionStylist: An Expert Knowledge-enhanced Multimodal Dataset for Fashion Understanding

Authors: Kaidong Feng, Zhuoxuan Huang, Huizhong Guo, Yuting Jin, Xinyu Chen, Yue Liang, Yifei Gai, Li Zhou,

Organizations: Yanshan University Qinhuangdao, China · Central South University Changsha, China · Zhejiang University Hangzhou, China · Southwest University Chongqing, China ·

Abstract

Fashion understanding requires both visual perception and expert-level reasoning about style, occasion, compatibility, and outfit rationale. However, existing fashion datasets remain fragmented and task-specific, often focusing on item attributes, outfit co-occurrence, or weak textual supervision, and thus provide limited support for holistic outfit understanding. In this paper, we introduce FashionStylist, an expert-annotated benchmark for holistic and expert-level fashion understanding. Constructed through a dedicated fashion-expert annotation pipeline, FashionStylist provides professionally grounded annotations at both the item and outfit levels. It supports three representative tasks: outfit-to-item grounding, outfit completion, and outfit evaluation. These tasks cover realistic item recovery from complex outfits with layering and accessories, compatibility-aware composition beyond co-occurrence matching, and expert-level assessment of style, season, occasion, and overall coherence. Experimental results show that FashionStylist serves not only as a unified benchmark for multiple fashion tasks, but also as an effective training resource for improving grounding, completion, and outfit-level semantic evaluation in MLLM-based fashion systems.

Figures & tables

Figure 1. Overview of our proposed FashionStylist, where the green part presents the pipeline of dataset construction, the purple part provides toy data examples in FashionStylist, and the blue part introduces the proposed benchmark tasks.

Figure 2. ( Left ) Number of unique attribute values in FashionStylist across item- and outfit-level annotations. ( Right ) Normalized co-occurrence frequency between the top-5 most common colors and outfit styles.

Model	Mode	R@10 ( $\uparrow$ )	N@10 ( $\uparrow$ )	PSNR ( $\uparrow$ )	SSIM ( $\uparrow$ )	FID ( $\downarrow$ )	KID ( $\downarrow$ )
Flux.1- Kontext	Default	0.0765	0.0456	11.574	0.7255	77.879	0.0120
	Sft	0.0890	0.0575	12.175	0.7351	71.709	0.0126
	Improv.	+16.3%	+26.1%	+5.2%	+1.3%	+7.9%	-5.0%
Qwen- ImageEdit	Default	0.1812	0.1138	10.544	0.6720	63.598	0.0104
	Sft	0.2254	0.1406	11.677	0.6782	45.505	0.0042
	Improv.	+24.4%	+23.6%	+10.7%	+0.9%	+28.4%	+59.6%

Table 2. Performance comparison on Task #1, where

\uparrow

(

\downarrow

) denotes higher (lower) is better, Improv. denotes the relative improvement of Sft over Default , bold indicates the best within each model, and indicates the best across all models.

Figure 3. Performance comparison across two representative item categories. We report 100/FID so that higher values consistently indicate better performance across all metrics.

Type	Model	Mode	R@10 ( $\uparrow$ )	R@20 ( $\uparrow$ )	N@10 ( $\uparrow$ )	N@20 ( $\uparrow$ )
Retrieval	POG	Default	0.0022	0.0065	0.0007	0.0018
		Eke.	0.0054	0.0131	0.0028	0.0047
		Improv.	+145.5%	+101.5%	+300.0%	+161.1%
	CIRP	Default	0.0239	0.0423	0.0131	0.0175
		Eke.	0.0293	0.0456	0.0173	0.0213
		Improv.	+22.6%	+7.8%	+32.1%	+21.7%

Table 3. Performance comparison on Task #2, where Eke. denotes “Expert knowledge-enhanced”, Improv. denotes the relative improvement of Eke. over Default .

Model	Mode	Style ( $\uparrow$ )	Season ( $\uparrow$ )	Occasion ( $\uparrow$ )	Mod. ( $\uparrow$ )
Gemma3-4b	Instruct	0.1325	0.1950	0.2725	0.0100
	Think	0.1425	0.2575	0.2425	0.1200
	Sft	0.2500	0.4425	0.2875	0.0850
	Improv.	+88.7%	+126.9%	+5.5%	+750.0%
Qwen3VL-4b	Instruct	0.1975	0.3250	0.2525	0.1150
	Think	0.2125	0.3625	0.2250	0.1000

Table 4. Performance comparison on Task #3, where Improv. indicates the relative gain of Sft over Instruct .

Multimodal Understanding

Multimodal understanding studies how AI systems interpret and reason across text, images, video, audio, and other signals.

See all 479 papers

OmniFashion: Towards Generalist Fashion Intelligence via Multi-Task Vision-Language Learning
Mar 3, 2026Zhengwei Yang, Andi Long, Hao Li +3Vision-Language Models Large Vision-Language Models
National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, School of Computer Science, Wuhan University · Harbin Institute of Technology
Grounding Free-Form Instructions for Fashion Complementary Image Generation
Aug 24, 2026Matteo Attimonelli, Claudio Pomo, Alessandro De Bellis +3Image Generation Conditional Image Generation
Politecnico di Bari, Italy · Sapienza University of Rome, Italy · University of Klagenfurt, Austria
Loom: Hybrid Retrieval-Scoring Outfit Recommendation with Semantic Material Compatibility and Occasion-Aware Embedding Priors
May 11, 2026Anushree BerliaRecommender Systems Hybrid Retrieval

Mar 3, 2026J/K move · Enter open · S save

OmniFashion: Towards Generalist Fashion Intelligence via Multi-Task Vision-Language Learning

Zhengwei Yang, Andi Long, Hao Li, Zechao Hu+2

National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, School of Computer Science, Wuhan University · Harbin Institute of Technology

Vision-Language Models Large Vision-Language Models Image Retrieval Recommender Systems

Open paper AskPDF arXiv

Fashion intelligence spans multiple tasks, i.e., retrieval, recommendation, recognition, and dialogue, yet remains hindered by fragmented supervision and incomplete fashion annotations. These limitations jointly restrict the formation of consistent visual-semantic structures, preventing recent vision-language models (VLMs) from serving as a generalist fashion brain that unifies understanding and reasoning across tasks. Therefore, we construct FashionX, a million-scale dataset that exhaustively annotates visible fashion items within an outfit and organizes attributes from global to part-level. Built upon this foundation, we propose OmniFashion, a unified vision-language framework that bridges diverse fashion tasks under a unified fashion dialogue paradigm, enabling both multi-task reasoning and interactive dialogue. Experiments on multi-subtasks and retrieval benchmarks show that OmniFashion achieves strong task-level accuracy and cross-task generalization, highlighting its offering of a scalable path toward universal, dialogue-oriented fashion intelligence.

Abstract

Multimodal Understanding

Explore similar work

OmniFashion: Towards Generalist Fashion Intelligence via Multi-Task Vision-Language Learning

Grounding Free-Form Instructions for Fashion Complementary Image Generation

Loom: Hybrid Retrieval-Scoring Outfit Recommendation with Semantic Material Compatibility and Occasion-Aware Embedding Priors