Revisiting Multimodal Positional Encoding in Vision–Language Models

Jie Huang, Xuejing Liu, Sibo Song, RuiBing Hou, Hong Chang, Junyang Lin, Shuai Bai

Keywords: Vision-Language Models, Multimodal Position Encoding

TL;DR: We analyze multimodal RoPE, distill three guidelines, and introduce MHRoPE and MRoPE‑I—plug‑and‑play variants that consistently outperform prior methods.

Abstract: Multimodal position encoding is essential for vision-language models, yet there has been little systematic investigation into multimodal position encoding. We conduct a comprehensive analysis of multimodal Rotary Positional Embedding (RoPE) by examining its two core components: position design and frequency allocation. Through extensive experiments, we identify three key guidelines: positional coherence, full frequency utilization, and preservation of textual priors—ensuring unambiguous layout, rich representation, and faithful transfer from the pre-trained LLM. Based on these insights, we propose Multi-Head RoPE (MHRoPE) and MRoPE-Interleave (MRoPE-I), two simple and plug-and-play variants that require no architectural changes. Our methods consistently outperform existing approaches across diverse benchmarks, with significant improvements in both general and fine-grained multimodal understanding. Code is avaliable at https://github.com/JJJYmmm/Multimodal-RoPEs.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 10980