cs.CV（2026-09-28）

📊 共 88 篇论文 | 🔗 20 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (33 🔗5) 支柱九：具身大模型 (Embodied Foundation Models) (18 🔗3) 支柱三：空间感知与语义 (Perception & Semantics) (14 🔗6) 支柱一：机器人控制 (Robot Control) (9 🔗2) 支柱四：生成式动作 (Generative Motion) (5 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (3) 支柱七：动作重定向 (Motion Retargeting) (3 🔗1) 支柱八：物理动画 (Physics-based Animation) (2 🔗2) 支柱五：交互与反应 (Interaction & Reaction) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (33 篇)

#题目一句话要点标签🔗⭐
1 DiMoP: Diffusion-Driven Motion Representation Learning With Frame-Level Pseudo-Classification for Skeleton-Based Action Recognition 提出DiMoP以解决骨架动作识别中的运动表示学习问题 representation learning motion representation
2 RefineDrive: Reliable Failure-Guided Learning for Vision-Language-Action Driving 提出RefineDrive以解决自主驾驶中的失败学习问题 reinforcement learning vision-language-action VLA
3 Unified Trajectory Matching Policy Optimization: Diverse T2I Generation and VLA Generalization 提出统一轨迹匹配策略优化以解决T2I生成与VLA泛化问题 reinforcement learning vision-language-action VLA
4 Evidence-Aligned Multimodal On-Policy Self-Distillation for Fine-Grained Visual Understanding 提出证据对齐的多模态在线自蒸馏以解决细粒度视觉理解问题 distillation multimodal
5 CAR-VLA: Complexity-Aware and Risk-Adaptive Reasoning for Autonomous Driving 提出CAR-VLA以解决自动驾驶中的复杂性与风险适应性问题 reinforcement learning vision-language-action VLA ✅
6 Revisit to Segment: Working Memory Distillation for Reasoning Segmentation 提出SWiM框架以增强推理分割能力 reinforcement learning distillation large language model
7 WorldAttention: An Efficient Attention Architecture for Interactive Video World Models 提出WorldAttention以解决长视频生成中的计算复杂性问题 world model world models embodied AI
8 Counterfactual Attention Policy Distillation for Temporal Video Grounding 提出反事实注意力策略蒸馏以解决长视频时间定位问题 teacher-student distillation large language model
9 FlowTool: Controlling Tool Parameter in Image Retouching via Flow Matching 提出FlowTool以解决工具参数控制问题 flow matching large language model multimodal
10 WorldPlay2: Extending Real-Time Interactive World Models in Control and Horizon 提出WorldPlay2以解决实时互动世界模型的控制与一致性问题 world model world models distillation
11 RoGSW4RLD: Feed-Forward 4D Gaussian Lifting for Robot World Model Rollouts 提出RoGSW4RLD以解决多视角视频世界模型一致性问题 world model world models
12 What Makes World Action Models Generalize? An Empirical Study of Test-Time Future Modeling 提出Simple-WAM以提升世界动作模型的泛化能力 world action model world action models ✅
13 OPIS: An Input-Grounded Benchmark for Multi-Object Memory in Video World Models 提出OPIS基准以评估视频世界模型中的多对象记忆能力 world model world models
14 WM-VLM: Probing Internal World Models for Interleaved Visual-Textual Reasoning 提出WM-VLM以解决视觉-文本推理中的空间问题 world model world models
15 Distilling Visual Reasoning into Text Space 提出V2T框架以解决视觉推理中的中间表示问题 reinforcement learning distillation multimodal
16 CoDrive: Cross-Vehicle World-Consistent Video Generation with Precise Trajectory Control for Cooperative Driving 提出CoDrive以解决多车辆协作驾驶中的一致性问题 world model world models spatial relationship
17 Learning Native Reflection in Unified Models with Interleaved Reinforcement Learning 提出UMM-Reflection以解决多模态模型自我修复问题 reinforcement learning multimodal
18 DrawingsDreamer: A Unified Multi-View Engineering Drawings Generation Model 提出DrawingsDreamer以解决工程图纸生成中的几何精度问题 dreamer large language model
19 From UNI2-h to ConvNeXt-T: Lightweight Nuclei Instance Segmentation via Knowledge Distillation 提出轻量级蒸馏方法以解决数字病理中的细胞核实例分割问题 distillation foundation model
20 Role-Guided MOE for Encoder-Level Pathology Representation Learning in WSI Classification 提出角色引导的混合专家网络以提升WSI分类的病理表示学习 representation learning foundation model
21 PDMD: Projected Distribution Matching Distillation for Video Diffusion Models 提出PDMD以解决视频扩散模型中的批评者错误问题 distillation spatiotemporal ✅
22 AHMAD: Adaptive Hybrid Multi-task Vision Learning with Assisted Distillation for Keypoint Detection 提出AHMAD框架以解决多任务视觉学习的效率问题 distillation depth estimation
23 From Static to Dynamic: On-Policy Distillation from Image to Video Diffusion Models 提出MILD框架以解决视频生成中的动态一致性问题 distillation optical flow
24 PIVOT: Pivot-Aware On Policy Self Distillation for Multi-Turn VLM Agents 提出PIVOT以解决多轮VLM代理训练中的奖励稀疏问题 reinforcement learning distillation
25 G$^3$-LoRA: Organizing Reward-Weighted Video Data with Gradient-Guided Grouped LoRA 提出G$^3$-LoRA以解决异构奖励加权视频数据的训练问题 flow matching distillation multimodal
26 On-Policy Self-Distillation for Multi-Turn Image Editing 提出MT-OPSD框架以解决多轮图像编辑中的性能下降问题 distillation
27 Proxy2World: Learning to Generate Worlds From Lightweight Proxies without Seeing Them 提出Proxy2World以解决轻量级代理生成世界的问题 flow matching world model world models
28 ReSight-SMC: Two-Stage Power Sampling via Island SMC with Visual Scouts 提出ReSight-SMC以解决大规模视觉语言模型推理中的多模态采样问题 reinforcement learning multimodal
29 FILIGREE3D: Scaling Sparse Latent Flow Matching for Ultra-High-Resolution Image-to-3D Generation 提出Filigree3D以解决超高分辨率图像到3D生成的计算挑战 flow matching
30 CapField-OPD: Learning Continuous Capability Fields via Joint-Anchored Multi-Teacher On-Policy Distillation for Flow Models 提出CapField-OPD以解决流模型能力调用的脆弱性问题 distillation
31 Reinforcement Learning from Intermediate Renders for Image-to-Code Generation 提出IR4RL框架以优化图像到代码生成中的反馈问题 reinforcement learning
32 DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time 提出DynaTokens以解决动态场景下视频生成问题 world model world models ✅
33 Precise Editing and Flexible Referencing for Interactable Worlds 提出EditWorld以解决交互世界中精确编辑与灵活引用问题 world model world models ✅

🔬 支柱九：具身大模型 (Embodied Foundation Models) (18 篇)

#题目一句话要点标签🔗⭐
34 Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference 提出文本-视觉协同令牌缓存以解决VLA模型高计算成本问题 vision-language-action VLA visual grounding
35 SPIDER: Multi-Layer Semantic Token Pruning and Adaptive Sub-Layer Skipping in Multimodal Large Language Models 提出SPIDER以解决多模态大语言模型的效率问题 large language model multimodal
36 Resolution as a First-Class Decision: Task-Conditioned Routing for Efficient Multimodal Large Language Models 提出任务条件的分辨率路由以提高多模态大语言模型的推理效率 large language model multimodal
37 How Far Are We from Removing the Visual Encoder? Scaling Laws for Encoder-Free Multimodal Pretraining 提出无编码器多模态预训练方法以提升模型性能 large language model multimodal
38 Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models 提出δ-Vision以解决多模态语言模型中的视觉状态重建问题 large language model multimodal
39 Reduce, Then Encode: Multiscale Volumetric Reduction for 2D Foundation Models in Brain MRI 提出多尺度体积压缩方法以解决2D基础模型在脑MRI中的应用问题 foundation model
40 Rethinking Visual Token Compression for Video Large Language Models: A Simple Yet Strong Baseline 提出SimpleCluster以提升视频大语言模型的视觉令牌压缩效率 large language model ✅
41 ORAV: Benchmarking Audio-Video Generation from Multimodal Contexts 提出ORAV基准以解决多模态音视频生成挑战 multimodal
42 Recent Advances in Agentic Agri-Robotic Phenotyping: A Perspective Review from Fragmented Multimodal Sensing to Unified PhenoAgent Intelligence 提出PhenoAgent框架以解决农业机器人表型测量的碎片化问题 multimodal
43 Sprout: Building Dynamic Memory While Reasoning for Agentic Video Understanding 提出Sprout框架以解决长视频理解中的动态记忆问题 large language model multimodal
44 Automated Species Identification in Camera Trap Images for Wildlife Conservation 提出自注意力机制以解决低对比度相机捕获图像中的物种识别问题 large language model multimodal
45 Token-Disentangled Latent Test-Time Scaling for Vision-Language Reasoning 提出Token-Disentangled潜在测试时间缩放以优化视觉语言推理 large language model multimodal ✅
46 Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence 提出ReaLVR以解决潜在视觉推理中的证据缺口问题 large language model multimodal
47 Beyond Selection: Token Parameterization for Extreme Visual Token Compression 提出Braco以解决极端视觉令牌压缩问题 visual grounding
48 SubRot: Signed Gradient Subspace Calibration for VLM Rotation Quantization 提出SubRot以解决VLM旋转量化中的多模态能力保持问题 multimodal
49 P4Q: Co-designing Token Pruning and Quantization for Vision-Language Model Acceleration 提出P4Q框架以联合优化视觉标记剪枝与量化问题 multimodal
50 Backdoor as Probe: Test-Time Adversarial Defense for CLIP 提出Backdoor as Probe以解决CLIP模型的对抗性防御问题 foundation model ✅
51 PlaylistEval: Can Video-Language Judges Be Trusted at Day Scale and Beyond? 提出PlaylistEval以解决长视频理解中的评估信任问题 multimodal

🔬 支柱三：空间感知与语义 (Perception & Semantics) (14 篇)

#题目一句话要点标签🔗⭐
52 EviSplat: Preserving Multi-View Evidence in 3D Gaussian Splatting for Open-Vocabulary Segmentation 提出EviSplat以解决开放词汇分割中的多视角证据保留问题 3D gaussian splatting gaussian splatting splatting
53 Remote Sensing Sparse-View 3D Gaussian Splatting via Depth Image-Based Rendering 提出DIBR-GS以解决稀疏观测下的遥感新视角合成问题 monocular depth 3D gaussian splatting 3DGS ✅
54 AGILE-GS: Anchor-Guided Fast Next-Best-View Selection for Active 3D Gaussian Splatting 提出AGILE-GS以解决3D高斯点云下一最佳视角选择问题 3D gaussian splatting 3DGS gaussian splatting
55 Multi-Scale Semantic Mapping in Urban Environments via Observation Calibration and Policy Dependence Regularization 提出多尺度语义映射以解决城市环境中的观察校准问题 semantic mapping semantic map
56 Preference-Guided Adaptation for Open-Vocabulary Semantic Segmentation via Prompt Disagreement 提出偏好引导适应框架以解决开放词汇语义分割问题 open-vocabulary open vocabulary ✅
57 Rate-Distortion Adaptive Primitive Selection for Omnidirectional Gaussian Splatting 提出OIC-GS以解决全景高效编码与重建质量问题 gaussian splatting splatting
58 Less Is More: Genetic Frame Selection for Efficient Novel View Synthesis 提出基因框架选择方法以提高新视角合成效率 3D gaussian splatting gaussian splatting splatting
59 One Sensor, Whole Body - 3D Body Pose from a Single Consumer Earbud IMU 利用单个耳塞IMU实现3D全身姿态估计 sam 3D SAM 3D multimodal ✅
60 Many Eyes, One World: Feed-Forward 3D Reconstruction from Mixed Cameras 提出MEOW以解决混合相机的三维重建问题 3D reconstruction
61 VideoPhysEdit: Physical Counterfactual Video Editing via Rigid-Body Physical Scene Reconstruction 提出VideoPhysEdit以解决物理反事实视频编辑问题 scene reconstruction ✅
62 GenNVS: Geometry-enhanced Novel View Synthesis via Disentangled 3D Prior 提出GenNVS以解决单图像新视角合成中的几何结构模糊问题 3D gaussian splatting gaussian splatting splatting
63 LEGAU: Learning Semantic Gaussian Priors for Scalable Category-level Pose Estimation 提出LEGAU框架以解决类别级6D姿态估计问题 6D pose estimation multimodal
64 Reliability-Gated Fusion of Consumer Head and Foot IMUs for Lower-Body 3D Pose 提出可靠性门控融合方法以解决消费者IMU的低可靠性问题 sam 3D SAM 3D ✅
65 ReSS: Residual-Restoring Sparse Attention for 3D Vision Transformers 提出ReSS以解决3D视觉变换器稀疏注意力计算效率问题 VGGT ✅

🔬 支柱一：机器人控制 (Robot Control) (9 篇)

#题目一句话要点标签🔗⭐
66 D$^2$-VLA: Dual-Memory Dual-Frequency Vision-Language-Action Model For Long Dynamic Manipulation 提出D$^2$-VLA以解决长时间动态操控中的记忆与响应问题 manipulation vision-language-action VLA
67 ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning 提出ActionUNet以解决VLA模型在复杂环境中的鲁棒性问题 manipulation vision-language-action VLA ✅
68 ECHO: Event-Augmented Context with Hindsight and Outlook for Wrist-Only Manipulation 提出ECHO以解决腕部单摄像头操作中的空间时间限制问题 manipulation world action model world action models ✅
69 Dexterous Tactile World Model 提出灵巧触觉世界模型以提升操控预测精度 manipulation world model world models
70 ESTHER: Egocentric Stereo Hand Estimation and Reconstruction in the Wild 提出ESTHER以解决野外环境下的立体手部估计与重建问题 manipulation metric depth egocentric
71 FlowAct-R2: Beyond Talking Avatar via Streaming Multimodal References and Proactive Agent Planning 提出FlowAct-R2以实现实时互动类人视频生成 humanoid multimodal
72 DBCF: Dual-Branch Complementary Fusion of Foundation Models for Generalized Deepfake Detection 提出双分支互补融合模型以解决深伪检测问题 manipulation foundation model
73 PanoVLN: Towards Effective Panoramic Vision-and-Language Navigation 提出PanoVLN以解决全景视觉导航中的决策不足问题 quadruped spatial relationship VLN
74 Look Before You Judge: Training-Free Region Mining for Grounded and Explainable Deepfake Detection 提出无训练区域挖掘框架以解决可解释深伪检测问题 manipulation large language model multimodal

🔬 支柱四：生成式动作 (Generative Motion) (5 篇)

#题目一句话要点标签🔗⭐
75 Triangular Resampling for Long-Horizon Motion Generation 提出三角重采样方法以解决长时间运动生成中的误差积累问题 motion diffusion model motion diffusion motion generation
76 HUMAN-TCI: Hierarchical Multi-Stream Motion-Aware Network with Torso-Centered Interaction for Text-to-Motion Retrieval 提出HUMAN-TCI以解决文本引导的人体动作检索问题 text-to-motion motion retrieval human motion
77 BiMoGen: Bidirectional Motion-Text Generation via Unified Masked Discrete Diffusion 提出BiMoGen以解决双向运动-文本生成问题 text-to-motion motion generation human motion ✅
78 EvolvingAvatar: Interactive 3D Head Generation That Adapts as Conversations Unfold 提出EvolvingAvatar以解决互动3D头像生成问题 motion generation
79 SAGE: Subspace Alignment for Classifier-Free Guidance in Mixture-of-Experts Diffusion Models 提出SAGE以解决Mixture-of-Experts扩散模型中的分类器无关引导问题 classifier-free guidance

🔬 支柱六：视频提取与匹配 (Video Extraction) (3 篇)

#题目一句话要点标签🔗⭐
80 InfiniHand: Streaming World-Space Hand Motion Estimation from Egocentric Video 提出InfiniHand以解决世界空间手部运动估计问题 MANO egocentric motion estimation
81 Long Time No See: Benchmarking VLMs for Out-of-Sight Spatiotemporal Reasoning in Egocentric Videos 提出Beyond3D基准以解决动态视频中的物体出视推理问题 egocentric spatiotemporal visual grounding
82 Ego-Forge: Text and Geometric-Attention Free Exo-to-Egocentric Video Generation 提出Ego-Forge以解决无文本和几何偏差的视角转换视频生成问题 egocentric

🔬 支柱七：动作重定向 (Motion Retargeting) (3 篇)

#题目一句话要点标签🔗⭐
83 Do Emotion Concepts Generalize Across Sources, Modalities, and Architectures in Vision-Language Models? 提出CMES以解决情感概念在多模态模型中的泛化问题 motion representation large language model
84 GeoVerse: World-Consistent Novel View Synthesis in Geometric Latent Space 提出GeoVerse以解决稀疏图像下的新视图合成问题 geometric consistency foundation model
85 Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling 提出关系不一致建模以解决3D异常检测问题 geometric consistency ✅

🔬 支柱八：物理动画 (Physics-based Animation) (2 篇)

#题目一句话要点标签🔗⭐
86 ReVA: A Scene-Centric Dataset Beyond Repetition for Remote Sensing Video Question Answering 提出ReVA数据集以解决遥感视频问答中的重复性和静态性问题 spatiotemporal large language model multimodal ✅
87 Advancing Video-Text Pretraining with Multi-View Captions 提出多视角字幕生成框架以提升视频-文本预训练质量 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🔗⭐
88 HyperDAM: Hyperspectral Distractor-Aware Memory with Amodal Expansion for SAM 3 Tracking 提出HyperDAM以解决光谱视频目标跟踪中的干扰问题 HSI spatiotemporal foundation model

⬅️ 返回 cs.CV 首页 · 🏠 返回主页