arXiv 2610.23987 · cs.CV · 2026-10-14
What Generation Knows: Unlocking Fine-Grained Understanding in Unified Multimodal Models
Tengju Ru, Weitong Lian, Zecong Tang, Zhiheng Liu, Lingyi Meng, Haoran Li, Kaixuan Wang, Bohao Li 等 9 人
机构待核实
在最新榜 #279 本站收录于 2026-10-24
一句话要点
这篇论文系统分析统一多模态模型里“生成侧”表示对细粒度视觉理解的作用，发现它们对局部属性和空间关系更敏感，并直接利用原生生成侧表示来补强理解，而不是只把生成当作辅助监督。
This paper systematically analyzes what generation-side representations in unified multimodal models offer for fine-grained visual understanding, finding them more sensitive to local attributes and spatial relations, and uses native generation-side features directly to strengthen understanding rather than only as auxiliary supervision.
阅读 arXiv 原文 PDF Hugging Face 讨论 代码仓库
主题与学科
同主题：多模态
- MJev-Spatial: System One Spatial Intelligence via Native Attention Matching ▲556
- LoopVL: Recurrent Visual Intelligence ▲417
- VideoJev: System One Model for Video Understanding and Interaction ▲375
- S-Space: Exploring Spatial Workspace in Multimodal Language Models ▲270
- UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement ▲266
同主题：视觉理解与 3D
- Spectral-Spatial Interpretation 2026-10-15 · ▲753
- GroundingPI: A Grounding Foundation Model towards Physical Intelligence with Visual Primitives 2026-09-30 · ▲310
- VisionHOPE: Visual Backbones as Self-Modifying Learning Systems 2026-09-27 · ▲296
- WeVisOne-VL: Unified Visual Perception from Objects to Pixels 2026-10-22 · ▲247
- EgoXperience: Learning World-Grounded Egocentric Representations from Human Experience 2026-10-19 · ▲189
数据说明
元数据来自 arXiv API（已核实）；点赞来自 Hugging Face 每日论文；引用来自 Semantic Scholar。最近刷新 2026-10-26。
要点由本站根据摘要自写，不转载摘要原文，请阅读原文。