扩散OCR解码快3.2倍，单流架构2秒出片

重点关注

• MinerU-Diffusion: Rethinking Document OCR as Inverse Rendering via Diffusion Decoding score 10
  入选 HF Daily Papers；HF 热度: 112 upvotes (+4)；有代码实现；关键词(2): latency, vision-language
• Speed by Simplicity: A Single-Stream Architecture for Fast Audio-Video Generative Foundation Model score 10
  入选 HF Daily Papers；HF 热度: 101 upvotes (+4)；有代码实现；关键词(2): distillation, open-source
• From Static Templates to Dynamic Runtime Graphs: A Survey of Workflow Optimization for LLM Agents score 10
  入选 HF Daily Papers；HF 热度: 43 upvotes (+4)；有代码实现；关键词(3): deployment, agentic, tool use
• SpatialBoost: Enhancing Visual Representation through Language-Guided Reasoning score 10
  入选 HF Daily Papers；HF 热度: 40 upvotes (+4)；有代码实现；关键词(1): reasoning
• Repurposing Geometric Foundation Models for Multi-view Diffusion score 10
  入选 HF Daily Papers；HF 热度: 39 upvotes (+4)；有代码实现；关键词(2): pretraining, text-to-image
• mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT score 10
  入选 HF Daily Papers；HF 热度: 35 upvotes (+4)；有代码实现；关键词(1): fine-tuning
• Manifold-Aware Exploration for Reinforcement Learning in Video Generation score 10
  入选 HF Daily Papers；HF 热度: 32 upvotes (+4)；有代码实现；关键词(2): GRPO, post-training
• On the Direction of RLVR Updates for LLM Reasoning: Identification and Exploitation score 10
  入选 HF Daily Papers；HF 热度: 21 upvotes (+4)；有代码实现；关键词(1): reasoning
• Uncertainty-guided Compositional Alignment with Part-to-Whole Semantic Representativeness in Hyperbolic Vision-Language Models score 10
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(1): vision-language；顶会接收: CVPR
• Omni-WorldBench: Towards a Comprehensive Interaction-Centric Evaluation for World Models score 9
  入选 HF Daily Papers；HF 热度: 116 upvotes (+4)；有代码实现

也值得关注

• AdditiveLLM2: A Multi-modal Large Language Model for Additive Manufacturing score 5
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；关键词(2): instruction tuning, pretraining
• Which Concepts to Forget and How to Refuse? Decomposing Concepts for Continual Unlearning in Large Vision-Language Models score 4
  关键词(1): vision-language；顶会接收: CVPR
• Back to Point: Exploring Point-Language Models for Zero-Shot 3D Anomaly Detection score 4
  关键词(1): vision-language；顶会接收: CVPR
• PEARL: Geometry Aligns Semantics for Training-Free Open-Vocabulary Semantic Segmentation score 4
  关键词(1): latency；顶会接收: CVPR
• Stabilizing Iterative Self-Training with Verified Reasoning via Symbolic Recursive Self-Alignment score 4
  关键词(2): DPO, reasoning；顶会接收: ICLR
• OmniFM: Toward Modality-Robust and Task-Agnostic Federated Learning for Heterogeneous Medical Imaging score 4
  关键词(2): deployment, fine-tuning；顶会接收: CVPR
• Let's Think with Images Efficiently! An Interleaved-Modal Chain-of-Thought Reasoning Framework with Dynamic and Precise Visual Thoughts score 4
  关键词(1): reasoning；顶会接收: AAAI
• The Golden Subspace: Where Efficiency Meets Generalization in Continual Test-Time Adaptation score 4
  关键词(2): scaling, lightweight；顶会接收: CVPR
• Unified Spatiotemporal Token Compression for Video-LLMs at Ultra-Low Retention score 4
  关键词(2): compression, latency；顶会接收: CVPR
• GeoFusion-CAD: Structure-Aware Diffusion with Geometric State Space for Parametric 3D Design score 4
  关键词(3): lightweight, state space, mamba；顶会接收: CVPR
• Tuning Real-World Image Restoration at Inference: A Test-Time Scaling Paradigm for Flow Matching Models score 4
  机构: Zhejiang University；关键词(2): scaling, text-to-image
• FreeArtGS: Articulated Gaussian Splatting Under Free-moving Scenario score 4
  关键词(1): robotics；顶会接收: CVPR
• PAM: A Pose-Appearance-Motion Engine for Sim-to-Real HOI Video Generation score 4
  关键词(3): deployment, embodied, synthetic data；顶会接收: CVPR
• Drop-In Perceptual Optimization for 3D Gaussian Splatting score 4
  机构: Apple；关键词(1): compression
• TiCo: Time-Controllable Training for Spoken Dialogue Models score 4
  机构: MIT；关键词(2): post-training, open-source