arXiv：cs.AI· Xiaoya Tang, Shilong Liu, Daniel Smolyak, Xiwen Li, Quynh Nguyen, Tolga Tasdizen·· 16 小时前
ReConV：用重建引导的视觉 token 压缩框架
Bottlenecked semantic compression by representation reconstruction
AI 导读
针对 Vision Transformer 处理长视觉 token 序列计算开销高的问题，研究者提出重建引导的视觉 token 压缩框架 ReConV，用紧凑的 Summary Bank（SMB）汇总完整视觉序列，并保留重建残差显示未被 SMB 充分捕获的 token。
正文
Abstract:Vision Transformers incur high computational costs when processing long visual token sequences. Existing reduction methods typically rank tokens independently and discard those deemed unimportant or redundant. We propose a reconstruction-guided framework for visual token condensation (ReConV), which summarizes the complete visual sequence with a compact Summary Bank (SMB) and retains tokens whose information is not adequately captured by the SMB, as indicated by their reconstruction residuals. At each compression layer, a lightweight position-conditioned decoder reconstructs the current tokens, leaving shared information compressed in the SMB while explicitly preserving poorly reconstructed details. Experiments on image classification and multimodal understanding demonstrate favorable performance-efficiency trade-offs: ReConV surpasses the uncompressed baseline with 48\% fewer GMACs on ImageNet and enables reusable image-level representations for efficient multi-turn inference. Code is available at TokenDrop.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.17412 [cs.CV] |
| (or arXiv:2610.17412v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17412 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xiaoya Tang [view email]
[v1]
Mon, 12 Oct 2026 21:24:38 UTC (35,199 KB)
来源：arXiv：cs.AI · arxiv.org