cs.CV（2026-02-27）

📊 共 42 篇论文 | 🔗 10 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (16 🔗5) 支柱二：RL算法与架构 (RL & Architecture) (10 🔗2) 支柱三：空间感知与语义 (Perception & Semantics) (9 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (3) 支柱四：生成式动作 (Generative Motion) (1) 支柱一：机器人控制 (Robot Control) (1 🔗1) 支柱七：动作重定向 (Motion Retargeting) (1) 支柱八：物理动画 (Physics-based Animation) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (16 篇)

#题目一句话要点标签🔗⭐
1 Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation 提出自适应视觉增强AIR框架，缓解多模态大语言模型中的幻觉问题 large language model multimodal
2 GuardAlign: Test-time Safety Alignment in Multimodal Large Language Models GuardAlign：多模态大语言模型中基于测试时对齐的安全防御框架 large language model multimodal
3 Venus: Benchmarking and Empowering Multimodal Large Language Models for Aesthetic Guidance and Cropping Venus：提升多模态大语言模型的美学指导与裁剪能力 large language model multimodal ✅
4 Multimodal Optimal Transport for Unsupervised Temporal Segmentation in Surgical Robotics 提出TASOT，利用多模态最优传输实现手术机器人视频的无监督时序分割 multimodal zero-shot transfer ✅
5 A multimodal slice discovery framework for systematic failure detection and explanation in medical image classification 提出多模态切片发现框架，用于医学图像分类中系统性错误检测与解释 multimodal
6 PointCoT: A Multi-modal Benchmark for Explicit 3D Geometric Reasoning PointCoT：提出用于3D几何推理的多模态基准，解决MLLM在点云理解中的几何幻觉问题。 large language model multimodal chain-of-thought
7 Breaking the Data Barrier: Robust Few-Shot 3D Vessel Segmentation using Foundation Models 利用预训练模型，提出一种鲁棒的小样本3D血管分割方法，有效应对数据匮乏和领域迁移问题。 foundation model
8 Any Model, Any Place, Any Time: Get Remote Sensing Foundation Model Embeddings On Demand rs-embed：遥感基础模型嵌入按需获取，解决异构性难题 foundation model ✅
9 Suppressing Prior-Comparison Hallucinations in Radiology Report Generation via Semantically Decoupled Latent Steering 提出SDLS方法，通过语义解耦潜在空间引导，抑制放射报告生成中的历史对比幻觉。 large language model foundation model zero-shot transfer
10 Can Unified Generation and Understanding Models Maintain Semantic Equivalence Across Different Output Modalities? VGUBench揭示了统一多模态大模型在跨模态语义对齐上的不足 large language model multimodal
11 HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit HiDrop：通过分层视觉Token缩减提升多模态大语言模型的效率。 large language model multimodal ✅
12 Steering and Rectifying Latent Representation Manifolds in Frozen Multi-modal LLMs for Video Anomaly Detection 提出SteerVAD，通过引导和修正冻结多模态LLM中的潜在表征流形，解决视频异常检测问题。 large language model
13 Interpretable Debiasing of Vision-Language Models for Social Fairness 提出DeBiasLens，通过可解释方式消除视觉-语言模型中的社会偏见 multimodal
14 Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks 提出Ref-Adv基准，揭示MLLM在指代表达理解中视觉推理的局限性 multimodal
15 UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking UTPTrack：提出一种简单统一的Token剪枝框架，用于提升视觉跟踪效率。 multimodal ✅
16 DLEBench: Evaluating Small-scale Object Editing Ability for Instruction-based Image Editing Model 提出DLEBench，用于评估指令驱动图像编辑模型在小目标编辑上的能力。 instruction following

🔬 支柱二：RL算法与架构 (RL & Architecture) (10 篇)

#题目一句话要点标签🔗⭐
17 Thinking with Images as Continuous Actions: Numerical Visual Chain-of-Thought 提出数值视觉思维链（NV-CoT），实现多模态大语言模型中基于连续坐标的图像推理。 reinforcement learning large language model multimodal ✅
18 Annotation-Free Visual Reasoning for High-Resolution Large Multimodal Models via Reinforcement Learning 提出HART框架，通过强化学习实现高分辨率大模型无标注视觉推理 reinforcement learning multimodal
19 Pseudo Contrastive Learning for Diagram Comprehension in Multimodal Models 提出伪对比学习方法，提升多模态模型在图表理解中的性能 contrastive learning multimodal
20 VideoPulse: Neonatal heart rate and peripheral capillary oxygen saturation (SpO2) estimation from contact free video VideoPulse：通过面部视频非接触式估计新生儿心率和血氧饱和度 MAE PULSE
21 DACESR: Degradation-Aware Conditional Embedding for Real-World Image Super-Resolution 提出DACESR，利用退化感知条件嵌入增强真实世界图像超分辨率效果 Mamba contrastive learning multimodal ✅
22 A Mixed Diet Makes DINO An Omnivorous Vision Encoder 提出Omnivorous Vision Encoder，解决DINOv2跨模态特征对齐问题 distillation foundation model
23 Fixed Anchors Are Not Enough: Dynamic Retrieval and Persistent Homology for Dataset Distillation 提出RETA框架，通过动态检索和拓扑对齐提升数据集蒸馏的泛化能力。 distillation
24 MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation MSVBench：面向人类水平的多镜头视频生成评估基准 world model multimodal
25 See, Act, Adapt: Active Perception for Unsupervised Cross-Domain Visual Adaptation via Personalized VLM-Guided Agent 提出Sea$^2$，通过个性化VLM引导的智能体实现无监督跨域视觉自适应 reinforcement learning visual grounding
26 AHAP: Reconstructing Arbitrary Humans from Arbitrary Perspectives with Geometric Priors AHAP：提出一种无需相机标定的任意视角人体三维重建框架 contrastive learning SMPL

🔬 支柱三：空间感知与语义 (Perception & Semantics) (9 篇)

#题目一句话要点标签🔗⭐
27 Prune Wisely, Reconstruct Sharply: Compact 3D Gaussian Splatting via Adaptive Pruning and Difference-of-Gaussian Primitives 提出自适应剪枝与差分高斯基元的紧凑3D高斯溅射方法，提升渲染质量并显著降低模型尺寸。 3D gaussian splatting 3DGS gaussian splatting
28 SR3R: Rethinking Super-Resolution 3D Reconstruction With Feed-Forward Gaussian Splatting 提出SR3R，通过前馈高斯溅射实现3D超分辨率重建，提升泛化性和实时性。 3D gaussian splatting 3DGS gaussian splatting
29 Open-Vocabulary Semantic Segmentation in Remote Sensing via Hierarchical Attention Masking and Model Composition 提出ReSeg-CLIP以解决遥感数据的开放词汇语义分割问题 open-vocabulary open vocabulary
30 Hyperdimensional Cross-Modal Alignment of Frozen Language and Image Models for Efficient Image Captioning 提出HDFLIM，通过超维计算对齐冻结的语言和图像模型，实现高效图像描述生成。 semantic mapping semantic map foundation model ✅
31 Evidential Neural Radiance Fields 提出Evidential NeRF，在NeRF中实现高质量场景重建和不确定性量化。 NeRF neural radiance field scene reconstruction
32 DiffusionHarmonizer: Bridging Neural Reconstruction and Photorealistic Simulation with Online Diffusion Enhancer DiffusionHarmonizer：利用在线扩散增强器桥接神经重建与照片级真实感仿真 3D gaussian splatting gaussian splatting splatting
33 No Calibration, No Depth, No Problem: Cross-Sensor View Synthesis with 3D Consistency 提出一种无需标定和深度信息的跨模态视图合成方法，实现3D一致性。 3D gaussian splatting 3DGS gaussian splatting
34 Altitude-Aware Visual Place Recognition in Top-Down View 提出一种高度适应的视觉地点识别方法以解决高空变化问题 depth estimation metric depth
35 Micro-expression Recognition Based on Dual-branch Feature Extraction and Fusion 提出双分支特征提取融合网络，提升微表情识别精度 optical flow

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

#题目一句话要点标签🔗⭐
36 AoE: Always-on Egocentric Human Video Collection for Embodied AI 提出AoE系统，利用智能手机和颈戴支架，低成本高效收集第一人称视角人类交互视频数据。 egocentric embodied AI foundation model
37 EgoGraph: Temporal Knowledge Graph for Egocentric Video Understanding 提出EgoGraph，用于理解以自我为中心的超长时序视频，解决现有方法在长期依赖建模上的不足。 egocentric
38 Egocentric Visibility-Aware Human Pose Estimation 提出EvaPose，解决第一人称视角人体姿态估计中关键点不可见性问题 egocentric

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
39 U-Mind: A Unified Framework for Real-Time Multimodal Interaction with Audiovisual Generation U-Mind：用于实时多模态交互的统一框架，支持视听生成 motion generation multimodal instruction following

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
40 Action-Geometry Prediction with 3D Geometric Prior for Bimanual Manipulation 提出基于3D几何先验的动作-几何预测方法，用于双臂操作任务 manipulation bi-manual bimanual manipulation ✅

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

#题目一句话要点标签🔗⭐
41 FocusTrack: One-Stage Focus-and-Suppress Framework for 3D Point Cloud Object Tracking 提出FocusTrack，通过单阶段聚焦抑制框架实现高性能3D点云目标跟踪 motion estimation

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

#题目一句话要点标签🔗⭐
42 SpikeTrack: A Spike-driven Framework for Efficient Visual Tracking SpikeTrack：一种用于高效视觉跟踪的脉冲驱动框架 spatiotemporal ✅

⬅️ 返回 cs.CV 首页 · 🏠 返回主页