AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

V2PE: Improving Multimodal Long-Context Capability of Vision-Language Models with Variable Visual Position Encoding

Junqi Ge, Ziyi Chen, Jintao Lin, Jinguo Zhu, Xihui Liu, Jifeng Dai, Xizhou Zhu

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 28 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune on augmented long-context data instead of short-context only (ablation)InternVL2-2B · Long-VQA / MM-NIAHaccuracyImproved
reduce positional increment of visual tokens to 1/16 or 1/256 (fixed) (ablation)InternVL2-FT-32K · Long-VQA / MM-NIAHaccuracyImproved
add V2PE (variable visual position increments) in training (ablation)InternVL2-2B · MM-NIAH image retrieval, 256K tokensaccuracy: 1.5 -> 56.9Improved
add second-stage 256K fine-tuning with V2PE (ablation)InternVL2-V2PE-32K · MM-NIAH image retrieval, 512K tokensaccuracy: 10.8 -> 95.1Improved
apply V2PE to short-context benchmarks with small increments (ablation)InternVL2-V2PE · DocVQA, InfoVQAaccuracyHurt
apply V2PE to visual tokens only versus textual tokens or both (ablation)InternVL2-V2PE-32K · MM-NIAH (six tasks)accuracyImproved
use V2PE instead of visual token compression (ablation)InternVL2-V2PE-32K · MM-NIAH image retrievalaccuracyImproved
use variable positional increments instead of fixed during training (ablation)InternVL2-FT-32K · MM-NIAH image retrievalaccuracyImproved
use V2PE instead of linear interpolation or NTK-aware RoPE extension (ablation)InternVL2-FT-32K · Long-VQA / MM-NIAHaccuracyImproved
use V2PE-32K versus InternVL2-2BInternVL2-2B · MM-NIAH Avgaccuracy: 21.0 -> 70.2Improved
use V2PE-32K versus InternVL2-2BInternVL2-2B · MileBench Avgaccuracy: 49.9 -> 55.9Improved
use V2PE-32K versus InternVL2-2B on short-contextInternVL2-2B · General benchmarks averageaccuracy: 72.4 -> 73.0No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy