ZeroSense:How Vision matters in Long Context Compression
This work addresses the evaluation challenge for researchers and practitioners in multimodal AI, offering a more reliable benchmark for assessing visual-text compression methods, though it is incremental as it focuses on improving evaluation rather than proposing a new compression method.
The paper tackles the problem of evaluating visual-text compression (VTC) methods, which rely on downstream task performance that can be skewed by multimodal large language models' linguistic priors, and introduces a new evaluation framework that decouples these factors to assess VTC quality more accurately, showing through experiments that VTC quality and downstream task accuracy often diverge significantly.
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.