aiqbench analysis
LensVLM introduces a novel inference framework for Vision Language Models that dynamically expands compressed visual tokens back to high resolution only when necessary, effectively bypassing tokenization limits while maintaining accuracy. This approach significantly improves performance on text QA benchmarks by leveraging learned tools to selectively restore detail rather than relying on unreliable low-resolution visual reading. The method demonstrates strong generalization across multimodal document and code understanding tasks, offering a practical pathway to handle longer contexts without standard sequence truncation.
LensVLM: Selective Context Expansion for Compressed Visual Representation of Text
arXiv:2605.07019v2 Announce Type: replace-cross Abstract: Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder's effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3$\times$ effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1$\times$ effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.
Abstract:Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder's effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3$\times$ effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1$\times$ effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2605.07019 [cs.CV]
(or arXiv:2605.07019v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2605.07019 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 305 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
Performance Profiling on AMD GPUs – Part 5: Profiling-Driven Kernel Optimization with an AI Code-Assist Tool
AMD · Jul 16, 2026
[AINews] not much happened today
Latent Space · Sep 18, 2026
CoVisco: Codec-Native Vision Encoder with Native Token Compression for Unified Image-Video Understanding
arXiv · Oct 1, 2026