CORE：以紧凑目标中心表征作为 LVLM 词元合并的新范式
CORE: Compact Object-centric REpresentations as a New Paradigm for Token Merging in LVLMs
用轻量分割解码器产出的物体掩码作为高层语义先验来指导视觉词元合并，并配合质心引导排序恢复合并后词元的空间顺序——这套「语义先验引导压缩 + 位置重排」思路可直接迁移到任何需要削减视觉词元数量的 VLM 推理管线中。
原文依据 CORE leverages an efficient segmentation decoder to generate object masks, which serve as a high-level semantic prior to guide the merging of visual tokens
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论大型视觉语言模型（LVLM）常因视觉词元随图像分辨率呈二次增长而承受高昂的计算与内存开销。现有词元压缩方法虽形式多样，但往往缺乏高层语义理解，导致次优的合并、信息冗余或上下文丢失。为克服这些局限，我们提出 CORE（紧凑目标中心表征），一种视觉词元压缩的新范式。CORE 利用高效的分割解码器生成目标掩码，将其作为高层语义先验，引导视觉词元合并为一组紧凑的目标中心表征。此外，一种新颖的质心引导排序机制为合并后的词元恢复连贯的空间顺序，从而保留关键的位置信息。大量实验表明，CORE 不仅在固定压缩率下的六个权威基准上取得新的最优性能，还在自适应压缩率设置下实现显著的效率提升。即使在极端压缩下，仅激进保留全部视觉词元的 2.2%，CORE 仍能维持基线性能的 97.4%。本工作证明了目标中心表征在高效且有效的 LVLM 处理中的优越性。代码已发布于 https://github.com/jingyulei/CORE。
英文原文摘要 1372 字符
Large Vision-Language Models (LVLMs) usually suffer from prohibitive computational and memory costs due to the quadratic growth of visual tokens with image resolution. Existing token compression methods, while varied, often lack a high-level semantic understanding, leading to suboptimal merges, information redundancy, or context loss. To address these limitations, we introduce CORE (Compact Object-centric REpresentations), a new paradigm for visual token compression. CORE leverages an efficient segmentation decoder to generate object masks, which serve as a high-level semantic prior to guide the merging of visual tokens into a compact set of object-centric representations. Furthermore, a novel centroid-guided sorting mechanism restores a coherent spatial order to the merged tokens, preserving vital positional information. Extensive experiments show that CORE not only establishes a new state-of-the-art on six authoritative benchmarks for fixed-rate compression, but also achieves dramatic efficiency gains in adaptive-rate settings. Even under extreme compression, after aggressively retaining with only 2.2% of all visual tokens, CORE still maintains 97.4% of baseline performance. Our work demonstrates the superiority of object-centric representations for efficient and effective LVLM processing. The code is available at https://github.com/jingyulei/CORE.