Spatial understanding is a necessary step in moving multimodal large language models (MLLMs) toward reasoning about the 3D physical world. Their visual pretraining data, however, is almost entirely 2D, so the visual tokens processed by MLLMs often lack inherent 3D structural information. A growing number of methods inject such structure by transferring representations from external vision or geometry models, and collectively establish that such transfer helps. Behind these methods lies a largely underexplored question: what makes a visual representation good for spatial understanding? To answer it, we first separate what is transferred from how it is transferred, and then trace the resulting differences back to properties of the representations themselves. What we transfer comes from 3D geometry models, self-supervised image and video models, vision-language models, and generative models, in the form of latent features or explicit depth and point maps. We transfer it either by distillation, using it as a supervision target, or by fusion, feeding it as an additional input. We find that a representation's 3D structure correlates strongly with how much it helps, measured by cross-view consistency for latent features and by depth and reconstruction accuracy for explicit geometry. We further show that the form of a representation dictates the mechanism it needs. Latent features gain as much from fusion as from distillation; explicit geometry gains little from fusion, and most of its distillation gain depends on a geometry decoding head. Guided by these findings, we distill both latent features and explicit geometry into a single MLLM. Taking only RGB video as input, it consistently improves over its base model on both static 3D and dynamic 4D spatial reasoning benchmarks.
Tell Model Where to Look: Mitigating Hallucinations in MLLMs by Vision-Guided Attention
School of Computer Science and Technology, Beijing Institute of Technology（北京理工大学计算机科学与技术学院）；Zhongguancun Academy（中关村学院）；Southeast Academy of Information Technology, Beijing Institute of Technology（北京理工大学信息科技东南学院）；Zhongguancun Laboratory（中关村实验室）
Training-free Uncertainty Guidance for Complex Visual Tasks with MLLMs
Technical University of Munich（慕尼黑技术大学）；Munich Center for Machine Learning（慕尼黑机器学习中心）；Helmholtz Munich（海德堡-慕尼黑亥姆霍兹研究中心）；Google（谷歌）；LTCI, Télécom Paris, Institut Polytechnique de Paris（巴黎理工大学 LTCI 实验室）
Open Eyes, Then Reason: Fine-grained Visual Mathematical Understanding in MLLMs
Australian Institute for Machine Learning, University of Adelaide（阿德莱德大学澳大利亚机器学习研究所）；Georgia Institute of Technology（佐治亚理工学院）；Nanjing University of Science and Technology（南京理工大学）；University of Oxford（牛津大学）；The Ohio State University（俄亥俄州立大学）
Mitigating Visual Degradation in MLLMs via Spatial-Spectral Visual Anchor Learning
China University of Petroleum (East China)（中国石油大学（华东））；Shanghai Jiao Tong University（上海交通大学）；Wuhan University（武汉大学）；Great Wall Motor（长城汽车）；Nanyang Technological University（南洋理工大学）；The University of Hong Kong（香港大学）
ERA: Entropy-Guided Visual Token Pruning with Rectified Attention for Efficient MLLMs
School of Future Technology, Dalian University of Technology（大连理工大学未来技术学院）；S-Lab, Nanyang Technological University（南洋理工大学S-Lab）；OPPO Research Institute（OPPO研究院）；Department of Computing, The Hong Kong Polytechnic University（香港理工大学计算机系）
Unveiling the Ignorance of MLLMs: Seeing Clearly, Answering Incorrectly
Hong Kong University of Science and Technology（香港科技大学）；School of AI, Shanghai Jiao Tong University（上海交通大学人工智能学院）；Beijing University of Posts and Telecommunications（北京邮电大学）；Peking University（北京大学）；Everlyn AI；Youtu Lab, Tencent（腾讯优图实验室）；Beijing Academy of Artificial Intelligence（北京人工智能研究院）；University of Central Florida（中佛罗里达大学）
DisCo: Towards Distinct and Coherent Visual Encapsulation in Video MLLMs
University of Chinese Academy of Sciences（中国科学院大学）；The University of Hong Kong（香港大学）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Shanghai Innovation Institute（上海创新研究院）；Fudan University（复旦大学）
VisualQuest: A Benchmark for Abstract Visual Reasoning in MLLMs
School of Computer Science and Technology, Dalian University of Technology, Dalian, China（大连理工大学计算机科学与技术学院）；School of Foreign Languages, Dalian University of Technology, Dalian, China（大连理工大学外语学院）；School of Computer Science and Technology, Xinjiang Normal University, Urumqi, China（新疆师范大学计算机科学与技术学院）
Where MLLMs Attend and What They Rely On: Explaining Autoregressive Token Generation
Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Department of Computer Science, Hong Kong Baptist University（香港 Baptist 大学计算机科学系）；College of Computing and Data Science, NTU（NTU 计算与数据科学学院）；Huawei（华为）；School of Flexible Electronics, SYSU（SYSU 灵活电子学院）；School of Cyber Science and Technology, Shenzhen Campus of Sun Yat-sen University（Sun Yat-sen 大学深圳校区计算机科学与技术学院）
From What to Which: Decoding Modifier Grounding in Frozen MLLMs
AI for Good (AIGO)（人工智能促进美好（AIGO））；Istituto Italiano di Tecnologia（意大利技术研究院）；University of Siena（锡耶纳大学）；University of Genoa（热那亚大学）；University of Verona（维罗纳大学）
Customizing Visual Emotion Evaluation for MLLMs: An Open-vocabulary, Multifaceted, and Scalable Approach
Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）；VCIP & TMCC & DISSec, College of Computer Science, Nankai University（南开大学计算机学院）；Department of Psychological and Cognitive Sciences, Tsinghua University（清华大学心理学与认知科学系）；University of Chinese Academy of Sciences（中国科学院大学）
MLLM-4D: Towards Visual-based Spatial-Temporal Intelligence
University of Macau（澳门大学）；Xi'an Jiaotong University（西安交通大学）；The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））；GVC Lab, Great Bay University（大湾大学GVC实验室）
Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMs
Nanjing University（南京大学）；The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））；China Academy of Launch Vehicle Technology（中国运载火箭技术研究院）；Beihang University（北京航空航天大学）
RedundancyLens: Revealing and Exploiting Visual Token Processing Redundancy for Efficient Decoder-Only MLLMs
South China University of Technology（华南理工大学）；Intsig Information Co., Ltd.（Intsig信息有限公司）；INTSIG-SCUT Joint Lab on Document Analysis and Recognition（INTSIG-SCUT文档分析与识别联合实验室）
