FocusVTC improves visual text compression with adaptive resolution
FocusVTC: Efficient and High-Performance Visual Text Compression with Adaptive Resolution
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Long documents can be hard for language AI models to understand because they require a lot of computer power. This paper introduces FocusVTC, a method that makes reading long text easier by turning text into images at different resolutions. It shows the whole page in low detail but zooms in on important areas to keep key information clear. The authors created a big set of examples to teach the models where to look and when to zoom in, making the process faster and more accurate.
What this means in practice
- •For ai system developers: Build more efficient language models that handle long documents by combining low-detail global views with focused high-detail regions.
- •For document processing teams: Speed up and improve accuracy on tasks that involve understanding complex documents by using adaptive visual text compression.
- •For cloud service providers: Offer faster multimodal AI services with reduced memory and compute costs by integrating adaptive resolution visual text compression.$Commercial implications: Enables new AI cloud features that efficiently process long text inputs as images, reducing resource use and increasing throughput.
Authors
FangZhi Zhong, Xuerui Qiu, Yuqi Pan, Ya Liu, Shaowei Gu, Bo Xu, Guoqi Li
Abstract
Long-context reasoning in large language models incurs substantial computation and memory costs. Visual text compression (VTC) reduces input length by rendering text as images, but fixed-resolution rendering creates a compression-performance trade-off: low DPI saves tokens at the expense of legibility, whereas high DPI spends tokens on irrelevant content. We introduce FocusVTC, which breaks this trade-off through adaptive resolution while preserving general multimodal capabilities. It combines compressed low-DPI global views with selective region enhancement, integrating enhanced views into ongoing reasoning. We construct 29.4K high-quality Reasoning-Evidence Localization (REL) chain-of-thought examples (REL-CoT) that link reasoning traces to page indices and bounding boxes. Multi-resolution REL supervised fine-tuning (REL-SFT) teaches the model to localize relevant regions, and Group Relative Policy Optimization learns when to enhance resolution and how to use the resulting observations, without a separate continual-pretraining stage. At 72 DPI on RULER v1, FocusVTC scores 87.4 at $2.9\times$ input compression, including tool observations, versus 57.5 for Glyph at $3.0\times$ input compression. It surpasses its text-input backbone on LongBench (56.40 versus 55.86), improves the MRCR macro-average by 13.91 points, and achieves a 51.19 macro-average on VTCBench. The MRCR latency evaluation also shows a $2.79\times$ online end-to-end speedup over Text. General multimodal capabilities are preserved, with MMMU increasing from 65.12 to 66.73 and MME from 2424.02 to 2457.62.