cs.CV（2026-09-29）

📊 共 84 篇论文 | 🔗 18 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (31 🔗8) 支柱九：具身大模型 (Embodied Foundation Models) (26 🔗4) 支柱三：空间感知与语义 (Perception & Semantics) (11 🔗3) 支柱一：机器人控制 (Robot Control) (7 🔗1) 支柱四：生成式动作 (Generative Motion) (3) 支柱六：视频提取与匹配 (Video Extraction) (2 🔗1) 支柱七：动作重定向 (Motion Retargeting) (2 🔗1) 支柱八：物理动画 (Physics-based Animation) (1) 支柱五：交互与反应 (Interaction & Reaction) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (31 篇)

#题目一句话要点标签🔗⭐
1 World2Motion: Turning Video World Models into 3D Human Motion Generators 提出World2Motion以解决3D人类动作生成的环境限制问题 world model world models motion generation
2 SFE-VGGT: Source-Free VGGT Distillation for Event-Based Monocular Depth Estimation 提出SFE-VGGT以解决事件摄像头深度估计中的源数据依赖问题 teacher-student distillation depth estimation
3 VISTA: Internalizing Collective Visual Experience via On-Policy Distillation for Active Multimodal Agents 提出VISTA以解决多模态智能体的视觉经验利用不足问题 reinforcement learning distillation multimodal
4 PolyOCR-Venus: Unified OCR Foundation Models for Text-Centric Visual Intelligence 提出PolyOCR以解决复杂视觉环境中的OCR任务平衡问题 distillation foundation model instruction following
5 Do-JEPA: From Masking to Intervention in Latent World Models 提出Do-JEPA以解决潜在世界模型的干预问题 world model world models JEPA
6 ResComEmb: Effective and Efficient Multimodal Embedding via Residual Homogeneity Compression 提出ResComEmb以解决多模态嵌入的效率与表达力问题 representation learning large language model multimodal
7 LongLive-Plug: Once-for-All Distillation for Video Generation 提出LongLive-Plug以解决视频生成模型蒸馏效率问题 world model world models distillation
8 Honeycomb: Constant-Size Scene Memory Representation for Video World Models 提出Honeycomb以解决视频世界模型中的场景记忆问题 world model world models spatiotemporal ✅
9 HelixWorld: A Real-time Interactive Audio-Visual World Model 提出HelixWorld以解决音频视觉世界模型的交互性问题 world model world models distillation
10 Salt++: Context-Aligned Post-Training for Few-Step Streaming Multimodal Generation 提出Salt++以解决少步流媒体多模态生成中的上下文对齐问题 distillation multimodal ✅
11 Speed in the Blind Spot: An Interpretability Analysis of Dynamic Perception in VLMs for Autonomous Driving 分析动态感知在自动驾驶中的可解释性问题 MAE motion representation vision-language-action
12 EVO-WAM: Evolving World Action Models through Video-Action Verification 提出EVO-WAM框架以解决机器人学习任务适应性问题 world action model world action models ✅
13 V-JEPA Policy: Building Effective World-Action Models on Predictive Visual Latents 提出V-JEPA Policy以构建有效的世界-动作模型 flow matching JEPA vision-language-action ✅
14 Waypoint-1.5: A Real-Time Video World Model for Consumer Hardware 提出Waypoint 1.5以解决实时视频生成的交互性问题 world model world models
15 MeteoVerse: Unified Weather-Controllable Video World Model 提出MeteoVerse以解决视频世界模型中的天气控制问题 world model world models
16 Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE 提出SplitMoE以解决视频生成模型中的均匀性陷阱问题 world model world models spatiotemporal
17 Med-RADIO: Reducing All Medical Domains Into One via Multi-Teacher Distillation 提出Med-RADIO框架以整合多模态医学知识 distillation foundation model ✅
18 MG-Thinker: Bi-Axial Self-Reflection for Multi-Image Reasoning Grounding 提出MG-Thinker以解决多图像推理定位问题 reinforcement learning multimodal chain-of-thought
19 HaPRL: Human-Anchored Process Reinforcement Learning for Visual Search Agent 提出HaPRL框架以解决视觉搜索代理的训练效率问题 reinforcement learning foundation model
20 TaoFlowForge: Progressive Native Mesh Generation via Cascaded Flow Matching 提出TaoFlowForge以解决轻量化和可编辑的3D网格生成问题 flow matching foundation model
21 On-Policy Visual Evidence Distillation 提出ReVuE以解决视觉代理中的证据蒸馏问题 distillation multimodal ✅
22 Learning via Self-Consistency for Diffusion-based Video Reasoning 提出自一致性方法以提升扩散模型的视频推理能力 distillation large language model chain-of-thought
23 Seeing What Should Be Heard: Diagnosing and Repairing Cross-Modal Shortcuts in Omni-Modal LLMs 提出DMC-Repair以解决多模态大语言模型中的跨模态捷径问题 reinforcement learning large language model multimodal
24 RS-OPSD: Reliable Privileged On-Policy-Self-Distillation for Ultra-High-Resolution Remote Sensing VQA 提出RS-OPSD以解决超高分辨率遥感视觉问答问题 distillation privileged information
25 Spatial-OPSD: Self-Improving Spatial Reasoning via Label-Free Self-Distillation 提出Spatial-OPSD框架以解决空间推理中的标签依赖问题 distillation privileged information ✅
26 EGSD: Event-Grounded Self-Distillation for Streaming Video Understanding 提出EGSD以解决流媒体视频理解中的自蒸馏问题 distillation privileged information
27 SoL-Refiner: Speed-of-Light One-Step Refinement for High-Resolution Video 提出SoL-Refiner以解决高分辨率视频生成的效率问题 reinforcement learning distillation spatiotemporal
28 Rethinking Representations for World-Action Modeling 提出ReWAM以解决世界-动作建模中的表示问题 policy learning world model world models
29 LIFT: Layout-In-Future Video Generation under Large Viewpoint Change via On-Policy Self-Distillation 提出LIFT框架以解决大视角变化下的视频生成问题 distillation
30 Rollout-Marginal Distillation for Long-Horizon Autoregressive Video Generation 提出Rollout-Marginal Distillation以解决长时间序列视频生成中的预测误差问题 distillation ✅
31 Codebook-Guided Cross-Modal Knowledge Distillation for Structurally Heterogeneous Features 提出基于代码本的跨模态知识蒸馏以解决特征结构异构问题 distillation

🔬 支柱九：具身大模型 (Embodied Foundation Models) (26 篇)

#题目一句话要点标签🔗⭐
32 ReCAP: Retrieval-Guided Capability Reuse for Multimodal Continual Instruction Tuning 提出ReCAP框架以解决多模态持续指令调优中的知识重用问题 large language model multimodal
33 VIF-Bench: Evaluating Visual Instruction Following in Multi-Reference Image Generation 提出VIF-Bench以评估多参考图像生成中的视觉指令跟随能力 multimodal instruction following
34 EpiCon: Collective Agent Learning through Co-Evolving Multimodal Memory 提出EpiCon框架以解决多智能体经验共享问题 multimodal
35 TomoTransformer: Towards a Foundation Model for CT Reconstruction 提出TomoTransformer以解决CT重建中的视角变化问题 foundation model
36 FedSocket: Recipient-Executable Knowledge Exchange for Heterogeneous Multimodal Federated Learning 提出FedSocket以解决异构多模态联邦学习中的知识交换问题 multimodal
37 Evaluating the Evaluators: Diagnosing Large Multimodal Models for AI-Generated Image Assessment 提出SQUARE-Bench以解决AI生成图像评估的全面性问题 multimodal
38 HyperSAM: A Promptable Foundation Model for Hyperspectral Remote Sensing 提出HyperSAM以解决高光谱遥感模型构建难题 foundation model
39 SAM Meets VLM: Parameter-Decoupled Full-Parameter Training for Unified Medical Reasoning and Segmentation 提出参数解耦的全参数训练框架以解决医学推理与分割统一问题 large language model multimodal language conditioned
40 Task-Oriented Visual Feature Compression via Residual Vector Quantization for Device-Edge Multimodal Inference 提出Q-TOFC以解决设备边缘多模态推理中的特征压缩问题 multimodal
41 GleanVID: Complementary Token Selection for Efficient Video Large Language Models 提出GleanVID以解决视频大语言模型的冗余信息问题 large language model
42 VesselBench-800K: A Large-scale Perception Benchmark for Multimodal Vessel Detection, Counting, and Density Estimation 提出VesselBench-800K以解决多模态船舶感知问题 multimodal ✅
43 SYNCR: Diagnosing and Learning Cross-Video Reasoning from Simulation 提出SYNCR框架以解决跨视频推理问题 large language model multimodal
44 Why MLLMs Struggle to Count: Overcoming Individuation and Aggregation Bottlenecks with ConvStack 提出ConvStack以解决多模态大语言模型的计数瓶颈问题 large language model multimodal
45 Representation Dynamics Reveal Semantic Saliency and Similarity for Visual Token Pruning in MLLMs 提出MSDG-Prune以解决多模态大语言模型的视觉令牌冗余问题 large language model multimodal ✅
46 Decoding Affective Nuances: Enhancing MLLMs via Hierarchical Emotion Reasoning and Contrastive Discriminative Pruning 提出DAN框架以解决多模态大语言模型情感推理不足问题 large language model multimodal
47 GlassFormer: Learning Real-time Glass Segmentation using Radar-Depth Fusion 提出GlassFormer以解决透明表面分割问题 multimodal
48 VideoLoop: Looped Working Memory Against Semantic Thrashing in Long-Form Video Agents 提出VideoLoop以解决长视频代理中的语义混乱问题 multimodal
49 GA-EIRFS: A Geometry-Augmented Repeat-Factor Sampling Method for Long-Tailed LiDAR 3D Object Detection 提出GA-EIRFS以解决长尾LiDAR 3D目标检测问题 multimodal ✅
50 Selective Channel Restoration for Backdoored Vision-Language Models 提出选择性通道恢复方法以解决视觉语言模型后门攻击问题 multimodal
51 Are In-Context Images Worth 10 Dimensions? 提出共享判别几何以优化视觉语言模型的分类能力 large language model
52 Tracing the Evidence: Faithful Token Attribution Through Vision-Language Reasoning 提出VTrace以解决多模态证据追踪问题 multimodal ✅
53 TReVS: Integrating Textual Relevance and Visual Saliency for Efficient Vision-Language Model Token Pruning 提出TReVS以解决视觉语言模型的高效令牌修剪问题 large language model
54 Complementary Retrieval-Augmented Prompting for Consistent Long-Form Video Generation 提出互补检索增强提示以解决长视频生成一致性问题 foundation model
55 When to Retrieve, When to Stay: Uncertainty-Aware Temporal Evidence Allocation for Streaming Video-LLMs 提出WRWS框架以解决视频理解中的不确定性问题 large language model
56 VISTA-Bench: Benchmarking Multilingual Image Translation with Image-Specific Rubrics 提出VISTA-Bench以解决多语言图像翻译评估不足问题 multimodal
57 Scene Retargeting: Learning Object Placement with Analogical Transfer 提出场景重定向方法以解决3D场景布局生成问题 embodied AI

🔬 支柱三：空间感知与语义 (Perception & Semantics) (11 篇)

#题目一句话要点标签🔗⭐
58 Prior-Driven Enhancements in 3D Gaussian Splatting: Normals and Depths Regularization 提出基于几何先验的3D高斯点云优化方法以解决渲染精度问题 depth estimation 3D gaussian splatting 3DGS
59 DispFlow-GS: Displacement Flow Supervision with Motion Disentangling for Monocular Deformable 3D Gaussian Splatting 提出DispFlow-GS以解决动态场景重建中的运动建模问题 3D gaussian splatting 3DGS gaussian splatting
60 NRF-GS: Neural Residual Fields for Expressive and Compact Gaussian Splatting 提出NRF-GS以解决3D高斯点云表达能力不足问题 3D gaussian splatting 3DGS gaussian splatting
61 Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering 提出Imagine3D-LLM以解决多视角图像的3D理解问题 3D gaussian splatting gaussian splatting splatting
62 WINGS: Reference-Free Gaussian Splatting Inpainting with 3D-Native Generative Priors 提出无参考高斯点云修复方法以解决3D编辑问题 3D gaussian splatting gaussian splatting splatting
63 UniAfford: Token-Routed Multitask Learning for Generalizable 2D-3D Affordance Perception 提出UniAfford以解决2D-3D可操作性感知的碎片化问题 affordance multimodal ✅
64 EndoPrior-GS: Dynamic Endoscopic Reconstruction with a Joint Texture Prior 提出EndoPrior-GS以解决动态内窥镜重建中的几何和光照问题 3D gaussian splatting 3DGS gaussian splatting ✅
65 Collision-Aware and Observation-Aligned Object-Centric Scene Reconstruction from Point Cloud 提出COOL框架以解决物体中心场景重建中的碰撞与观察对齐问题 scene reconstruction
66 Does the VGGT Family Need All Its Layers? 提出VGGT模型层冗余分析以优化3D结构重建 VGGT ✅
67 LazySloth: Bounded LLM-based Lazy Tree Search for Fast Long Video Comprehension 提出LazySloth以解决长视频理解中的效率问题 scene understanding multimodal
68 The Camera Inside the Editor: Reading the Implicit Camera of Image Editors with Painted Calibration Patterns 提出隐式相机模型以解析图像编辑器的绘制过程 MoGe

🔬 支柱一：机器人控制 (Robot Control) (7 篇)

#题目一句话要点标签🔗⭐
69 InsightMap: Structured Spatial Modeling for Embodied Multimodal Reasoning 提出InsightMap以解决语言引导导航中的空间建模问题 Unitree multimodal
70 Exemplar2VQA: A Scalable Exemplar-Driven Visual Question Answering Generation Framework via Multi-Agent Coding 提出Exemplar2VQA以解决3D问答数据稀缺问题 sim-to-real embodied AI large language model ✅
71 RoXDrive: Closed-Loop Reinforcement Learning for End-to-End Autonomous Driving via Action-Faithful Rollouts 提出RoXDrive以解决闭环自主驾驶中的因果混淆问题 sim-to-real reinforcement learning imitation learning
72 Multi-Granularity Language-Guided Imitation Learning via Instruction Decomposition 提出多粒度语言指导模仿学习以解决任务描述模糊问题 manipulation policy learning imitation learning
73 Real2Gym: Building Gyms from Videos, Bringing Skills to Robots 提出Real2Gym以解决从视频到机器人技能转化的挑战 manipulation sim2real real2sim
74 ORMA: Optimization-based Monocular 4D Reconstruction of Articulated Animals 提出ORMA框架以解决单目视频中动物四维重建问题 quadruped
75 The Domain Is a Residue: Adapting Self-Supervised Features, Not Generators 提出Representation Feature Adapter以解决图像天气效果去除问题 sim-to-real

🔬 支柱四：生成式动作 (Generative Motion) (3 篇)

#题目一句话要点标签🔗⭐
76 Why Cross-Skeleton Retargeting Is Non-Identifiable: Structural Limits of Generative Motion Models 提出源实例保真度以解决跨骨架重定向的非可识别性问题 motion generation
77 BeatDance: Generating Beat-Consistent 3D Dance with Hierarchical Spatial-Temporal Modeling 提出BeatDance以解决3D舞蹈与音乐节拍同步问题 physically plausible human motion
78 RED: Reconstruction Evolution Dynamics for Generalizable AI-Generated Image Detection 提出RED框架以解决AI生成图像检测中的重建演化问题 VQ-VAE

🔬 支柱六：视频提取与匹配 (Video Extraction) (2 篇)

#题目一句话要点标签🔗⭐
79 APM-Bench: Benchmarking Cross-session Persistent Memory for Egocentric Streaming Video Assistants 提出APM-Bench以解决跨会话持久记忆问题 egocentric
80 Does Local Video Understanding Transfer Across Encounters? The EgoGears Benchmark 提出EgoGears基准以解决视频理解迁移问题 egocentric ✅

🔬 支柱七：动作重定向 (Motion Retargeting) (2 篇)

#题目一句话要点标签🔗⭐
81 AESOP: Asymmetric Human-Camera Generation with Translation-Intensity Control 提出AESOP框架以解决人类动作与摄像机生成的非对称依赖问题 human motion
82 OmniTaskonomy: When Does Visual Generation Improve Visual Understanding? 提出OmniTaskonomy以揭示视觉生成对理解的影响 spatial relationship ✅

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
83 OmniRoute: Mapping Temporal Semantic Evidence to Audio-Visual Token Budgets for Efficient Omnimodal Large Language Models 提出OmniRoute以解决音视频序列处理效率问题 spatiotemporal large language model multimodal

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
84 WeLike2Party! In-Context Motion Transfer for Multi-Human Image Animation 提出WeLike2Party以解决多人体图像动画中的运动转移问题 multi-person interaction motion representation

⬅️ 返回 cs.CV 首页 · 🏠 返回主页