cs.CV（2026-04-14）

📊 共 43 篇论文 | 🔗 12 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (23 🔗8) 支柱三：空间感知与语义 (Perception & Semantics) (6 🔗2) 支柱一：机器人控制 (Robot Control) (5) 支柱二：RL算法与架构 (RL & Architecture) (5 🔗2) 支柱六：视频提取与匹配 (Video Extraction) (2) 支柱四：生成式动作 (Generative Motion) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (23 篇)

#题目一句话要点标签🔗⭐
1 CLASP: Class-Adaptive Layer Fusion and Dual-Stage Pruning for Multimodal Large Language Models CLASP：面向多模态大语言模型的类自适应层融合与双阶段剪枝 large language model multimodal ✅
2 All in One: A Unified Synthetic Data Pipeline for Multimodal Video Understanding 提出统一合成数据流水线，解决多模态视频理解中数据匮乏问题 large language model multimodal visual grounding
3 Chain-of-Models Pre-Training: Rethinking Training Acceleration of Vision Foundation Models 提出模型链预训练（CoM-PT），加速视觉基础模型训练且无性能损失。 large language model foundation model
4 Probabilistic Feature Imputation and Uncertainty-Aware Multimodal Federated Aggregation 提出P-FIN，解决多模态联邦学习中特征缺失和不确定性问题，提升医疗诊断安全性。 multimodal
5 Towards Long-horizon Agentic Multimodal Search 提出LMM-Searcher以解决长时段多模态搜索问题 multimodal ✅
6 Challenging Vision-Language Models with Physically Deployable Multimodal Semantic Lighting Attacks 提出多模态语义光照攻击MSLA，挑战视觉-语言模型在物理世界的安全性。 multimodal
7 AffectAgent: Collaborative Multi-Agent Reasoning for Retrieval-Augmented Multimodal Emotion Recognition 提出AffectAgent以解决多模态情感识别中的模态歧义问题 multimodal ✅
8 Brain-DiT: A Universal Multi-state fMRI Foundation Model with Metadata-Conditioned Pretraining Brain-DiT：基于元数据条件扩散预训练的通用多状态fMRI基础模型 foundation model ✅
9 GeoAlign: Geometric Feature Realignment for MLLM Spatial Reasoning GeoAlign通过几何特征重对齐提升MLLM的空间推理能力 large language model foundation model multimodal
10 Relaxing Anchor-Frame Dominance for Mitigating Hallucinations in Video Large Language Models 提出Decoder-side Temporal Rebalancing (DTR)以缓解视频大语言模型中的幻觉问题 large language model
11 MODIX: A Training-Free Multimodal Information-Driven Positional Index Scaling for Vision-Language Models MODIX：一种免训练的多模态信息驱动的位置索引缩放方法，提升视觉-语言模型性能 multimodal
12 Boosting Visual Instruction Tuning with Self-Supervised Guidance 提出V-GIFT，通过自监督指导提升视觉指令微调，增强MLLM的视觉推理能力 large language model multimodal ✅
13 Distorted or Fabricated? A Survey on Hallucination in Video LLMs 对视频大语言模型幻觉现象的全面综述，提出系统分类与缓解策略。 large language model visual grounding ✅
14 DPC-VQA: Decoupling Quality Perception and Residual Calibration for Video Quality Assessment 提出DPC-VQA，解耦质量感知与残差校准，高效评估视频质量 large language model multimodal
15 NTIRE 2026 The 3rd Restore Any Image Model (RAIM) Challenge: Professional Image Quality Assessment (Track 1) NTIRE 2026 RAIM挑战赛：探索MLLM在专业图像质量评估中的应用 large language model multimodal ✅
16 Why and When Visual Token Pruning Fails? A Study on Relevant Visual Information Shift in MLLMs Decoding 提出DSTP框架，解决MLLM解码过程中视觉token剪枝在复杂推理任务中性能下降问题 large language model multimodal
17 Agentic Discovery with Active Hypothesis Exploration for Visual Recognition HypoExplore：基于主动假设探索的Agentic视觉识别架构发现框架 large language model
18 Don't Show Pixels, Show Cues: Unlocking Visual Tool Reasoning in Language Models via Perception Programs 提出感知程序P²，通过语言原生线索提升多模态大语言模型视觉工具推理能力 multimodal
19 OmniFood8K: Single-Image Nutrition Estimation via Hierarchical Frequency-Aligned Fusion 提出OmniFood8K数据集和单图营养估计框架，解决中餐营养估计难题。 multimodal ✅
20 Combating Pattern and Content Bias: Adversarial Feature Learning for Generalized AI-Generated Image Detection 提出多维对抗特征学习框架，提升AI生成图像检测的泛化能力。 multimodal
21 Towards Realistic and Consistent Orbital Video Generation via 3D Foundation Priors 利用3D基础先验，实现逼真且一致的物体轨道视频生成 foundation model
22 Boosting Robust AIGI Detection with LoRA-based Pairwise Training 提出基于LoRA的Pairwise训练方法LPT，提升AIGI图像在复杂失真下的鲁棒检测性能。 foundation model
23 Redefining Quality Criteria and Distance-Aware Score Modeling for Image Editing Assessment 提出DS-IEQA框架，解决图像编辑质量评估中度量标准僵化和距离无关评分建模问题。 multimodal

🔬 支柱三：空间感知与语义 (Perception & Semantics) (6 篇)

#题目一句话要点标签🔗⭐
24 PDF-GS: Progressive Distractor Filtering for Robust 3D Gaussian Splatting 提出PDF-GS，通过渐进式干扰物过滤实现鲁棒的3D高斯溅射 3D gaussian splatting 3DGS gaussian splatting ✅
25 ArtifactWorld: Scaling 3D Gaussian Splatting Artifact Restoration via Video Generation Models ArtifactWorld：通过视频生成模型扩展3D高斯溅射伪影修复 3D gaussian splatting 3DGS gaussian splatting
26 ELoG-GS: Dual-Branch Gaussian Splatting with Luminance-Guided Enhancement for Extreme Low-light 3D Reconstruction ELoG-GS：基于亮度引导增强的双分支高斯溅射，用于极端弱光3D重建 gaussian splatting splatting geometric consistency ✅
27 Pi-HOC: Pairwise 3D Human-Object Contact Estimation 提出Pi-HOC，用于解决多人-多物体交互场景下的3D人体-物体接触估计问题 sam 3D SAM 3D human-object interaction
28 Grasp in Gaussians: Fast Monocular Reconstruction of Dynamic Hand-Object Interactions GraG：基于高斯模型的快速单目动态手-物交互三维重建 sam 3D SAM 3D
29 Cross-Attentive Multiview Fusion of Vision-Language Embeddings 提出CAMFusion，通过交叉注意力多视角融合提升3D场景语义分割性能。 open-vocabulary open vocabulary

🔬 支柱一：机器人控制 (Robot Control) (5 篇)

#题目一句话要点标签🔗⭐
30 PianoFlow: Music-Aware Streaming Piano Motion Generation with Bimanual Coordination 提出PianoFlow以解决双手协调钢琴动作生成问题 bi-manual flow matching motion synthesis
31 From Attenuation to Attention: Variational Information Flow Manipulation for Fine-Grained Visual Perception 提出VIF框架，解决MLLM在细粒度视觉感知中的信息衰减问题 manipulation large language model multimodal
32 Detecting Precise Hand Touch Moments in Egocentric Video 提出HiCE模块，用于精准检测第一视角视频中手与物体接触的时刻 manipulation egocentric spatiotemporal
33 Bridging the Micro--Macro Gap: Frequency-Aware Semantic Alignment for Image Manipulation Localization 提出FASA框架以解决图像操控定位中的微观与宏观差距问题 manipulation
34 Direct Discrepancy Replay: Distribution-Discrepancy Condensation and Manifold-Consistent Replay for Continual Face Forgery Detection 提出直接差异重放方法，解决持续人脸伪造检测中的灾难性遗忘问题 manipulation

🔬 支柱二：RL算法与架构 (RL & Architecture) (5 篇)

#题目一句话要点标签🔗⭐
35 RSGMamba: Reliability-Aware Self-Gated State Space Model for Multimodal Semantic Segmentation 提出RSGMamba，解决多模态语义分割中模态可靠性差异导致的特征退化问题。 Mamba state space model scene understanding
36 SEATrack: Simple, Efficient, and Adaptive Multimodal Tracker SEATrack：一种简单、高效、自适应的多模态目标跟踪器，提升性能效率。 representation learning multimodal ✅
37 Hypergraph-State Collaborative Reasoning for Multi-Object Tracking 提出HyperSSM框架，通过超图状态协同推理解决多目标跟踪中的运动估计难题。 SSM state space model motion estimation
38 Visual Preference Optimization with Rubric Rewards 提出rDPO框架以优化视觉偏好评估 DPO direct preference optimization multimodal
39 Cross-Modal Knowledge Distillation for PET-Free Amyloid-Beta Detection from MRI 提出PET引导的跨模态知识蒸馏，实现仅MRI的阿尔茨海默病Aβ检测 contrastive learning distillation ✅

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

#题目一句话要点标签🔗⭐
40 A Dataset and Evaluation for Complex 4D Markerless Human Motion Capture 提出复杂场景4D无标记人体运动捕捉数据集与评估基准，解决真实交互场景下的性能瓶颈。 SMPL SMPL-X human motion
41 EgoEsportsQA: An Egocentric Video Benchmark for Perception and Reasoning in Esports EgoEsportsQA：提出电子竞技第一视角视频问答基准，用于评估感知与推理能力。 egocentric large language model

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
42 Conflated Inverse Modeling to Generate Diverse and Temperature-Change Inducing Urban Vegetation Patterns 提出Conflated逆建模框架，生成多样且可控的城市植被模式以调控温度 physically plausible

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

#题目一句话要点标签🔗⭐
43 VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization 提出VideoFlexTok，一种灵活长度的由粗到精视频Token化方法，提升视频生成效率。 spatiotemporal

⬅️ 返回 cs.CV 首页 · 🏠 返回主页