New DIM-Fashion benchmark and FashionAM model advance multi-turn fashion image retrieval

Researchers have introduced DIM-Fashion, a new benchmark designed to improve multi-turn fashion image retrieval by addressing the limitations of existing methods. Current approaches often assume a uniform intent transition, failing to account for diverse user behaviors like rollbacks. DIM-Fashion, comprising 26,000 multi-turn sessions from various fashion retrieval datasets, aims to capture these complex interactions. Additionally, the team developed FashionAM, a multimodal large language model and vision-language pre-training framework that directly processes conversational queries against visual embeddings, bypassing intermediate textification to preserve finer visual details. AI

IMPACT Enhances multimodal understanding for complex, interactive search scenarios, potentially improving e-commerce and personalized recommendation systems.

RANK_REASON Research paper introducing a new benchmark and model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• DagsHub
• DIM-Fashion
• FashionAM
• Gotit.pub
• Hugging Face
• Litmaps
• ScienceCast
• scite Smart Citations

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →