cs.CV（2025-10-04）

📊 共 14 篇论文 | 🔗 5 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (7 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (4 🔗2) 支柱三：空间感知与语义 (Perception & Semantics) (2 🔗1) 支柱四：生成式动作 (Generative Motion) (1)

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

#题目一句话要点标签🔗⭐
1 LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorization 提出LIBERO-PRO以解决现有VLA模型评估不公正问题 vision-language-action VLA ✅
2 Bridge Thinking and Acting: Unleashing Physical Potential of VLM with Generalizable Action Expert 提出基于可泛化动作专家的框架，提升VLM在物理世界的动作执行能力 vision-language-action VLA
3 Exploring Instruction Data Quality for Explainable Image Quality Assessment 针对可解释图像质量评估，提出基于聚类的数据选择方法IQA-Select，提升数据效率。 large language model multimodal
4 Harnessing Synthetic Preference Data for Enhancing Temporal Understanding of Video-LLMs TimeWarp：利用合成偏好数据增强视频大语言模型的时间理解能力 large language model ✅
5 Talking Tennis: Language Feedback from 3D Biomechanical Action Recognition 提出基于3D生物力学动作识别的网球动作语言反馈框架，提升训练效果。 large language model
6 DHQA-4D: Perceptual Quality Assessment of Dynamic 4D Digital Human 提出DHQA-4D数据集与DynaMesh-Rater模型，用于动态4D数字人感知质量评估 multimodal
7 The Overlooked Value of Test-time Reference Sets in Visual Place Recognition 提出参考集微调方法，提升视觉定位在跨域场景下的泛化性能 foundation model

🔬 支柱二：RL算法与架构 (RL & Architecture) (4 篇)

#题目一句话要点标签🔗⭐
8 UGround: Towards Unified Visual Grounding with Unrolled Transformers UGround：提出基于解缠Transformer的统一视觉定位框架，解决误差累积和缺乏空间信息问题。 reinforcement learning visual grounding ✅
9 Contrastive-SDE: Guiding Stochastic Differential Equations with Contrastive Learning for Unpaired Image-to-Image Translation 提出Contrastive-SDE，利用对比学习引导随机微分方程，解决非配对图像转换问题。 contrastive learning
10 Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training 提出增强型自蒸馏框架，用于高效脉冲神经网络训练 distillation ✅
11 Unsupervised Transformer Pre-Training for Images: Self-Distillation, Mean Teachers, and Random Crops DINOv2深度解读：非监督Transformer预训练，自蒸馏与均值教师方法 distillation

🔬 支柱三：空间感知与语义 (Perception & Semantics) (2 篇)

#题目一句话要点标签🔗⭐
12 Cross-View Open-Vocabulary Object Detection in Aerial Imagery 提出跨视角开放词汇目标检测框架，解决航拍图像目标识别难题 open-vocabulary open vocabulary
13 Optimized Minimal 4D Gaussian Splatting OMG4：优化最小4D高斯溅射，显著降低动态场景表示的存储开销。 gaussian splatting splatting ✅

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
14 Generating Human Motion Videos using a Cascaded Text-to-Video Framework 提出CAMEO级联框架，用于从文本生成逼真的人体运动视频 text-to-motion embodied AI

⬅️ 返回 cs.CV 首页 · 🏠 返回主页