Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.11846 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915857213423616 |
|---|---|
| author | Gao, Yonghan Chen, Zehong Xu, Lijian Chen, Jingzhi Guan, Jingwei Zeng, Xingyu |
| author_facet | Gao, Yonghan Chen, Zehong Xu, Lijian Chen, Jingzhi Guan, Jingwei Zeng, Xingyu |
| contents | Recent visual-text compression (VTC) methods, typified by DeepSeek-OCR, report impressive high token compression ratios for long-context modeling tasks by leveraging text-to-image rendering. However, existing evaluation protocols heavily rely on downstream task performance. Such evaluation metrics fail to accurately measure text preservation due to the strong inherent linguistic priors of Multimodal Large Language Models (MLLMs). In this work, we introduce a new evaluation framework that decouples MLLMs' capabilities to faithfully assess VTC quality. Within this framework, we further introduce the ZeroSense Benchmark to ensure low semantic correlation of testing samples. By eliminating contextual dependencies, our benchmark guarantees that the evaluation results are purely reflective of VTC quality, unaffected by the semantic inference capabilities of downstream models. Extensive experiments across multiple datasets demonstrate that VTC quality and downstream task accuracy diverge significantly, highlighting the necessity of our decoupled evaluation framework. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_11846 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ZeroSense:How Vision matters in Long Context Compression Gao, Yonghan Chen, Zehong Xu, Lijian Chen, Jingzhi Guan, Jingwei Zeng, Xingyu Computer Vision and Pattern Recognition Recent visual-text compression (VTC) methods, typified by DeepSeek-OCR, report impressive high token compression ratios for long-context modeling tasks by leveraging text-to-image rendering. However, existing evaluation protocols heavily rely on downstream task performance. Such evaluation metrics fail to accurately measure text preservation due to the strong inherent linguistic priors of Multimodal Large Language Models (MLLMs). In this work, we introduce a new evaluation framework that decouples MLLMs' capabilities to faithfully assess VTC quality. Within this framework, we further introduce the ZeroSense Benchmark to ensure low semantic correlation of testing samples. By eliminating contextual dependencies, our benchmark guarantees that the evaluation results are purely reflective of VTC quality, unaffected by the semantic inference capabilities of downstream models. Extensive experiments across multiple datasets demonstrate that VTC quality and downstream task accuracy diverge significantly, highlighting the necessity of our decoupled evaluation framework. |
| title | ZeroSense:How Vision matters in Long Context Compression |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.11846 |