Diverse-Intent Multi-Turn Fashion Image Retrieval

2026-07-22

AI Summary

This work addresses the limitations of existing multi-turn fashion image retrieval methods, which assume user intents follow fixed attribute-editing patterns and rely on textual intermediaries that discard fine-grained visual details. To overcome these challenges, the authors introduce DIM-Fashion, a novel benchmark dataset comprising 26K multi-turn conversations that, for the first time, supports complex intent shifts and backtracking behaviors. They further propose FashionAM, a framework that leverages multimodal large language models and vision-language pretraining to directly align dialogue queries with fashion image embeddings in a shared space, enabling end-to-end multimodal alignment without textual mediation. Experiments demonstrate that FashionAM significantly outperforms current state-of-the-art methods across multiple tasks, substantially enhancing robustness and adaptability to heterogeneous user intents.

Abstract

Real-world fashion search involves interactive retrieval across multiple turns. However, existing multi-turn retrieval methods are built on a restrictive assumption that every interaction follows the same attribute-editing paradigm, leaving heterogeneous intent transitions unexplored. Moreover, existing approaches often rely on textification to bridge multimodal queries and visual retrieval, which may lose fine-grained visual cues. To address these gaps, we introduce DIM-Fashion, a benchmark of 26K multi-turn sessions constructed from 13 fashion retrieval datasets across 7 tasks, featuring diverse intent transitions and rollback behaviors. We further propose FashionAM, an MLLM-VLP framework that directly aligns multimodal conversational queries with a fashion-oriented gallery embedding space, avoiding intermediate textification. Extensive experiments demonstrate the effectiveness of FashionAM over existing approaches. The dataset and code will be made publicly available upon acceptance.

Problem

• multi-turn retrieval
• diverse intent
• fashion image retrieval
• multimodal queries
• visual cues

Innovation

• multi-turn retrieval
• diverse intent
• fashion image retrieval
• multimodal alignment
• textification-free