Revisiting Multimodal Positional Encoding in Vision-Language Models
This work addresses a key bottleneck in vision-language models for researchers and practitioners, offering incremental improvements through plug-and-play variants.
The paper tackled the problem of multimodal positional encoding in vision-language models by conducting a systematic analysis of Rotary Positional Embedding (RoPE) and proposing two variants, MHRoPE and MRoPE-I, which consistently outperformed existing approaches across diverse benchmarks with significant improvements in general and fine-grained multimodal understanding.
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 will be avaliable at https://github.com/JJJYmmm/Multimodal-RoPEs.