cs.CV（2026-03-17）

📊 共 54 篇论文 | 🔗 17 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (19 🔗8) 支柱二：RL算法与架构 (RL & Architecture) (14 🔗3) 支柱三：空间感知与语义 (Perception & Semantics) (10 🔗3) 支柱一：机器人控制 (Robot Control) (5 🔗1) 支柱四：生成式动作 (Generative Motion) (2) 支柱八：物理动画 (Physics-based Animation) (2 🔗1) 支柱七：动作重定向 (Motion Retargeting) (1 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (19 篇)

#题目一句话要点标签🔗⭐
1 GAP-MLLM: Geometry-Aligned Pre-training for Activating 3D Spatial Perception in Multimodal Large Language Models 提出GAP-MLLM，通过几何对齐预训练提升多模态大语言模型3D空间感知能力 large language model multimodal visual grounding
2 VisBrowse-Bench: Benchmarking Visual-Native Search for Multimodal Browsing Agents VisBrowse-Bench：用于多模态浏览代理的视觉原生搜索基准 large language model multimodal ✅
3 When Thinking Hurts: Mitigating Visual Forgetting in Video Reasoning via Frame Repetition 提出FrameRepeat框架，通过帧重复缓解视频推理中视觉信息遗忘问题 large language model multimodal chain-of-thought
4 KidsNanny: A Two-Stage Multimodal Content Moderation Pipeline Integrating Visual Classification, Object Detection, OCR, and Contextual Reasoning for Child Safety KidsNanny：用于儿童安全的双阶段多模态内容审核流水线 multimodal
5 Fast-WAM: Do World Action Models Need Test-time Future Imagination? 提出Fast-WAM，无需测试时未来想象，加速具身控制任务。 vision-language-action VLA ✅
6 Kestrel: Grounding Self-Refinement for LVLM Hallucination Mitigation Kestrel：提出基于视觉 grounding 和自精炼的 LVLM 幻觉缓解框架 multimodal visual grounding
7 MLLM-based Textual Explanations for Face Comparison 分析MLLM在人脸比对中生成解释的可靠性，揭示其幻觉问题 large language model multimodal ✅
8 InViC: Intent-aware Visual Cues for Medical Visual Question Answering 提出InViC框架，通过意图感知视觉线索提升医学VQA中图像理解能力。 large language model multimodal
9 360° Image Perception with MLLMs: A Comprehensive Benchmark and a Training-Free Method 提出Free360，一种无需训练的360°图像VQA框架，提升MLLM在全景图像理解能力。 large language model multimodal
10 What DINO saw: ALiBi positional encoding reduces positional bias in Vision Transformers 使用ALiBi位置编码减少Vision Transformer中的位置偏差，提升零样本迁移能力 foundation model
11 Retrieving Counterfactuals Improves Visual In-Context Learning 提出CIRCLES框架，通过检索反事实样本提升视觉上下文学习能力 multimodal ✅
12 World Reconstruction From Inconsistent Views 提出一种非刚性对齐方法，从不一致的视频帧中重建3D世界。 foundation model
13 BUSSARD: Normalizing Flows for Bijective Universal Scene-Specific Anomalous Relationship Detection 提出BUSSARD，用标准化流进行双射通用场景特定异常关系检测 multimodal ✅
14 VIEW2SPACE: Studying Multi-View Visual Reasoning from Sparse Observations 提出VIEW2SPACE基准，研究稀疏视角下的多视角视觉推理，并提出Grounded Chain-of-Thought方法。 chain-of-thought
15 Cross-modal learning for plankton recognition 提出基于跨模态自监督学习的浮游生物识别方法，利用图像和光学测量数据提升识别精度。 multimodal ✅
16 Persistent Story World Simulation with Continuous Character Customization EverTale：提出持续角色定制的故事世界模拟器，解决角色一致性与场景融合问题。 chain-of-thought
17 Grounding the Score: Explicit Visual Premise Verification for Reliable Vision-Language Process Reward Models 提出EVPV，通过显式视觉前提验证提升视觉-语言过程奖励模型的可靠性。 multimodal ✅
18 Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training 提出IOMM：通过图像掩码建模实现高效的UMM视觉生成预训练 multimodal ✅
19 Boosting Quantitive and Spatial Awareness for Zero-Shot Object Counting 提出QICA框架，提升零样本物体计数中的数量感知和空间感知能力 zero-shot transfer

🔬 支柱二：RL算法与架构 (RL & Architecture) (14 篇)

#题目一句话要点标签🔗⭐
20 HGP-Mamba: Integrating Histology and Generated Protein Features for Mamba-based Multimodal Survival Risk Prediction HGP-Mamba：融合组织学和生成蛋白特征，用于基于Mamba的多模态生存风险预测 Mamba foundation model multimodal ✅
21 Iris: Bringing Real-World Priors into Diffusion Model for Monocular Depth Estimation Iris：将真实世界先验知识融入单目深度估计扩散模型 distillation depth estimation monocular depth
22 Learning Human-Object Interaction for 3D Human Pose Estimation from LiDAR Point Clouds 提出HOIL框架，利用人-物交互学习提升LiDAR点云3D人体姿态估计精度 contrastive learning contact-aware human-object interaction
23 Evo-Retriever: LLM-Guided Curriculum Evolution with Viewpoint-Pathway Collaboration for Multimodal Document Retrieval Evo-Retriever：基于LLM引导的课程演化和视角-路径协作的多模态文档检索 contrastive learning multimodal
24 ViT-AdaLA: Adapting Vision Transformers with Linear Attention ViT-AdaLA：通过线性注意力自适应视觉Transformer，解决长序列扩展性问题。 linear attention large language model foundation model
25 Deep Reinforcement Learning-driven Edge Offloading for Latency-constrained XR pipelines 提出基于深度强化学习的边缘卸载框架，优化时延约束XR应用 reinforcement learning deep reinforcement learning
26 SF-Mamba: Rethinking State Space Model for Vision SF-Mamba：面向视觉任务，通过辅助patch交换和批量折叠，提升Mamba模型的效率和性能。 Mamba state space model
27 Reliable Reasoning in SVG-LLMs via Multi-Task Multi-Reward Reinforcement Learning 提出CTRL-S框架，通过多任务多奖励强化学习提升SVG-LLMs的推理可靠性。 reinforcement learning chain-of-thought
28 Fast-HaMeR: Boosting Hand Mesh Reconstruction using Knowledge Distillation Fast-HaMeR：利用知识蒸馏加速手部网格重建，适用于资源受限设备。 distillation hand reconstruction ✅
29 Rationale Matters: Learning Transferable Rubrics via Proxy-Guided Critique for VLMReward Models 提出Proxy-GRM，通过代理引导的评价标准学习提升视觉-语言模型奖励模型的性能。 reinforcement learning RLHF multimodal ✅
30 RASLF: Representation-Aware State Space Model for Light Field Super-Resolution RASLF：提出表征感知状态空间模型，用于光场超分辨率重建 SSM state space model
31 Micro-AU CLIP: Fine-Grained Contrastive Learning from Local Independence to Global Dependency for Micro-Expression Action Unit Detection 提出Micro-AU CLIP框架，解决微表情动作单元检测中局部独立性和全局依赖性建模问题。 contrastive learning
32 VIGOR: VIdeo Geometry-Oriented Reward for Temporal Generative Alignment VIGOR：面向视频几何一致性的时序生成对齐方法，提升视频扩散模型质量 reinforcement learning foundation model
33 EFF-Grasp: Energy-Field Flow Matching for Physics-Aware Dexterous Grasp Generation EFF-Grasp：基于能量场流匹配的物理感知灵巧抓取生成 flow matching

🔬 支柱三：空间感知与语义 (Perception & Semantics) (10 篇)

#题目一句话要点标签🔗⭐
34 M^3: Dense Matching Meets Multi-View Foundation Models for Monocular Gaussian Splatting SLAM M^3：融合多视角基础模型与稠密匹配的单目高斯溅射SLAM gaussian splatting splatting scene reconstruction
35 Leveling3D: Leveling Up 3D Reconstruction with Feed-Forward 3D Gaussian Splatting and Geometry-Aware Generation Leveling3D：结合前馈3D高斯溅射与几何感知生成，提升3D重建质量 depth estimation 3D gaussian splatting 3DGS
36 Rethinking Pose Refinement in 3D Gaussian Splatting under Pose Prior and Geometric Uncertainty 提出基于蒙特卡洛采样的3DGS位姿优化方法，提升位姿先验和几何不确定性下的鲁棒性 3D gaussian splatting 3DGS gaussian splatting
37 MessyKitchens: Contact-rich object-level 3D scene reconstruction 提出MessyKitchens数据集，并设计MOD网络用于接触丰富的物体级3D场景重建 depth estimation scene reconstruction sam 3D ✅
38 WildDepth: A Multimodal Dataset for 3D Wildlife Perception and Depth Estimation WildDepth：用于3D野生动物感知和深度估计的多模态数据集 depth estimation multimodal
39 $D^3$-RSMDE: 40$\times$ Faster and High-Fidelity Remote Sensing Monocular Depth Estimation 提出D³-RSMDE框架，加速40倍并提升遥感单目深度估计质量 depth estimation monocular depth
40 PureCLIP-Depth: Prompt-Free and Decoder-Free Monocular Depth Estimation within CLIP Embedding Space PureCLIP-Depth：提出一种完全无提示、无解码器的CLIP嵌入空间单目深度估计模型 depth estimation monocular depth ✅
41 $x^2$-Fusion: Cross-Modality and Cross-Dimension Flow Estimation in Event Edge Space 提出$x^2$-Fusion，通过事件边缘空间统一多模态特征，提升动态场景光流和场景流估计精度。 scene understanding optical flow scene flow
42 VideoMatGen: PBR Materials through Joint Generative Modeling VideoMatGen：提出基于视频扩散Transformer的PBR材质联合生成方法 height map physically plausible
43 NanoGS: Training-Free Gaussian Splat Simplification NanoGS：一种免训练的高斯溅射简化框架，降低存储和传输成本。 3DGS ✅

🔬 支柱一：机器人控制 (Robot Control) (5 篇)

#题目一句话要点标签🔗⭐
44 ECHO: Edge-Cloud Humanoid Orchestration for Language-to-Motion Control ECHO：面向语言驱动的人形机器人全身控制的边缘-云协同框架 humanoid humanoid robot whole-body control
45 Ground Reaction Inertial Poser: Physics-based Human Motion Capture from Sparse IMUs and Insole Pressure Sensors GRIP：结合稀疏IMU与压力传感器的物理可信人体运动捕捉 humanoid physically plausible human motion
46 Segmentation-Based Attention Entropy: Detecting and Mitigating Object Hallucinations in Large Vision-Language Models 提出基于分割注意力熵（SAE）的方法，用于检测和缓解大型视觉语言模型中的对象幻觉问题。 quadruped multimodal visual grounding
47 S-VAM: Shortcut Video-Action Model by Self-Distilling Geometric and Semantic Foresight 提出S-VAM以解决视频动作模型实时推理与高保真预见问题 manipulation distillation foundation model ✅
48 Demystifing Video Reasoning 揭示视频生成模型中的推理机制：扩散去噪过程而非帧序列是关键 manipulation

🔬 支柱四：生成式动作 (Generative Motion) (2 篇)

#题目一句话要点标签🔗⭐
49 Interact3D: Compositional 3D Generation of Interactive Objects Interact3D：用于交互对象组合式3D生成，解决遮挡和空间关系保持问题 physically plausible spatial relationship
50 V-Co: A Closer Look at Visual Representation Alignment via Co-Denoising V-Co：通过协同去噪更深入地研究视觉表征对齐，提升像素空间扩散模型性能。 classifier-free guidance

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

#题目一句话要点标签🔗⭐
51 Point-to-Mask: From Arbitrary Point Annotations to Mask-Level Infrared Small Target Detection 提出Point-to-Mask框架，以低成本点标注实现红外小目标mask级检测。 spatiotemporal ✅
52 3D tomography of exchange phase in a Si/SiGe quantum dot device 提出3D相位体积提取方法以解决量子点设备中的交换相互作用问题 PULSE

🔬 支柱七：动作重定向 (Motion Retargeting) (1 篇)

#题目一句话要点标签🔗⭐
53 OneWorld: Taming Scene Generation with 3D Unified Representation Autoencoder OneWorld：提出3D统一表示自编码器，提升三维场景生成跨视角一致性。 geometric consistency foundation model ✅

🔬 支柱六：视频提取与匹配 (Video Extraction) (1 篇)

#题目一句话要点标签🔗⭐
54 SOMA: Unifying Parametric Human Body Models SOMA：统一参数化人体模型，实现跨模型数据融合与应用 SMPL SMPL-X

⬅️ 返回 cs.CV 首页 · 🏠 返回主页