cs.CV（2026-01-07）

📊 共 30 篇论文 | 🔗 3 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (11) 支柱九：具身大模型 (Embodied Foundation Models) (7 🔗1) 支柱三：空间感知与语义 (Perception & Semantics) (5) 支柱一：机器人控制 (Robot Control) (3 🔗2) 支柱六：视频提取与匹配 (Video Extraction) (1) 支柱七：动作重定向 (Motion Retargeting) (1) 支柱四：生成式动作 (Generative Motion) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (11 篇)

#题目一句话要点标签🔗⭐
1 Pixel-Wise Multimodal Contrastive Learning for Remote Sensing Images 提出像素级多模态对比学习PIMC，有效提升遥感图像时间序列分析性能 contrastive learning multimodal
2 Semantic Belief-State World Model for 3D Human Motion Prediction 提出语义信念状态世界模型(SBWM)用于解决3D人体运动预测中的长时漂移问题。 reinforcement learning world model latent dynamics
3 Staged Voxel-Level Deep Reinforcement Learning for 3D Medical Image Segmentation with Noisy Annotations 提出SVL-DRL框架，解决医学图像分割中带噪声标注的问题。 reinforcement learning deep reinforcement learning DRL
4 MVP: Enhancing Video Large Language Models via Self-supervised Masked Video Prediction 提出MVP：通过自监督掩码视频预测增强视频大语言模型 reinforcement learning large language model
5 From Brute Force to Semantic Insight: Performance-Guided Data Transformation Design with LLMs 提出NNGPT，利用LLM和性能反馈自动设计最优数据增强策略。 reinforcement learning large language model chain-of-thought
6 REFA: Real-time Egocentric Facial Animations for Virtual Reality 提出基于VR头显内红外相机的实时面部动画系统，无需校准。 distillation egocentric
7 Thinking with Frames: Generative Video Distortion Evaluation via Frame Reward Model 提出REACT：基于帧奖励模型的生成视频结构扭曲评估框架 reinforcement learning chain-of-thought
8 CrackSegFlow: Controllable Flow-Matching Synthesis for Generalizable Crack Segmentation with the CSF-50K Benchmark 提出CrackSegFlow，结合CSF-50K基准，提升裂缝分割的泛化性和可控性 flow matching
9 ToTMNet: FFT-Accelerated Toeplitz Temporal Mixing Network for Lightweight Remote Photoplethysmography 提出ToTMNet，利用FFT加速的Toeplitz时序混合网络实现轻量级远程光电容积脉搏波估计。 MAE PULSE
10 Diffusion-DRF: Differentiable Reward Flow for Video Diffusion Fine-Tuning 提出Diffusion-DRF以解决视频扩散模型微调中的奖励信号问题 DPO direct preference optimization
11 Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion Models 提出LocalDPO，通过局部细节偏好优化提升视频扩散模型生成质量 preference learning DPO

🔬 支柱九：具身大模型 (Embodied Foundation Models) (7 篇)

#题目一句话要点标签🔗⭐
12 Analyzing Reasoning Consistency in Large Multimodal Models under Cross-Modal Conflicts 提出主动视觉上下文精炼方法，提升大模型在跨模态冲突下的推理一致性 multimodal chain-of-thought
13 Scanner-Induced Domain Shifts Undermine the Robustness of Pathology Foundation Models 揭示病理学预训练模型对扫描仪差异的脆弱性，强调校准和嵌入稳定性的重要性 foundation model
14 EvalBlocks: A Modular Pipeline for Rapidly Evaluating Foundation Models in Medical Imaging EvalBlocks：用于医学影像领域基础模型快速评估的模块化流水线 foundation model ✅
15 MGPC: Multimodal Network for Generalizable Point Cloud Completion With Modality Dropout and Progressive Decoding MGPC：多模态网络，通过模态Dropout和渐进式解码实现通用点云补全 multimodal
16 I2E: From Image Pixels to Actionable Interactive Environments for Text-Guided Image Editing I2E：提出一种基于可交互环境的文本引导图像编辑框架，解决复杂组合编辑任务。 vision-language-action chain-of-thought
17 Klear: Unified Multi-Task Audio-Video Joint Generation Klear：统一的多任务音视频联合生成模型，解决对齐、泛化和数据稀缺问题。 instruction following
18 SDCD: Structure-Disrupted Contrastive Decoding for Mitigating Hallucinations in Large Vision-Language Models 提出SDCD：一种结构扰乱对比解码算法，用于缓解大型视觉语言模型中的对象幻觉问题 multimodal

🔬 支柱三：空间感知与语义 (Perception & Semantics) (5 篇)

#题目一句话要点标签🔗⭐
19 IDESplat: Iterative Depth Probability Estimation for Generalizable 3D Gaussian Splatting IDESplat：迭代深度概率估计，提升通用3D高斯溅射重建效果 depth estimation 3D gaussian splatting gaussian splatting
20 Bayesian Monocular Depth Refinement via Neural Radiance Fields 提出MDENeRF，利用神经辐射场迭代优化单目深度估计，提升几何细节。 depth estimation monocular depth NeRF
21 G2P: Gaussian-to-Point Attribute Alignment for Boundary-Aware 3D Semantic Segmentation 提出G2P，通过高斯到点的属性对齐实现边界感知的3D语义分割 3D gaussian splatting gaussian splatting splatting
22 Systematic Evaluation of Depth Backbones and Semantic Cues for Monocular Pseudo-LiDAR 3D Detection 系统评估深度骨干网络与语义线索以提升单目伪LiDAR 3D检测精度 metric depth Depth Anything
23 Gen3R: 3D Scene Generation Meets Feed-Forward Reconstruction Gen3R：融合重建先验与视频扩散模型，实现场景级3D生成 VGGT

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
24 CSMCIR: CoT-Enhanced Symmetric Alignment with Memory Bank for Composed Image Retrieval 提出CSMCIR，通过对称对齐和记忆库增强组合图像检索性能。 manipulation large language model multimodal
25 Can LLMs See Without Pixels? Benchmarking Spatial Intelligence from Textual Descriptions 提出SiT-Bench基准，评估大型语言模型在无像素输入下的空间智能 manipulation world model egocentric ✅
26 Choreographing a World of Dynamic Objects CHORD：通过蒸馏视频信息，生成动态物体和场景的通用框架 manipulation distillation ✅

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

#题目一句话要点标签🔗⭐
27 TRec: Egocentric Action Recognition using 2D Point Tracks TRec：利用2D点轨迹进行第一人称视角动作识别，提升识别精度。 egocentric

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

#题目一句话要点标签🔗⭐
28 SpatiaLoc: Leveraging Multi-Level Spatial Enhanced Descriptors for Cross-Modal Localization SpatiaLoc：利用多层次空间增强描述符实现跨模态定位 spatial relationship

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
29 FUSION: Full-Body Unified Motion Prior for Body and Hands via Diffusion FUSION：提出基于扩散模型的全身统一运动先验，用于生成身体和手部动作 motion synthesis

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

#题目一句话要点标签🔗⭐
30 MFC-RFNet: A Multi-scale Guided Rectified Flow Network for Radar Sequence Prediction 提出MFC-RFNet，融合多尺度特征与校正流，用于提升雷达回波序列预测精度。 spatiotemporal

⬅️ 返回 cs.CV 首页 · 🏠 返回主页