Zirui Li's Project Portfolio

Explore AI and NLP projects covering multimodal retrieval, model interpretability, and creative applications built during my PhD journey.

DocMMIR: Cross-Domain Document Multimodal Retrieval Framework

DocMMIR: Cross-Domain Document Multimodal Retrieval Framework

arXiv badge ACL Anthology badge HF Dataset badge License: MIT badge

DocMMIR is a unified cross-domain multimodal document retrieval framework that combines text and visual information, building large-scale cross-domain benchmarks and exploring different fusion strategies and loss functions to improve retrieval performance.

MultimodalInformation RetrievalDeep LearningComputer VisionNatural Language Processing
2025-05-19