3D仅需0.1%token，视频微调反伤空间理解

重点关注

• Complementary Reinforcement Learning score 10
  入选 HF Daily Papers；HF 热度: 31 upvotes (+4)；有代码实现；关键词(1): agentic
• GigaWorld-Policy: An Efficient Action-Centered World--Action Model score 10
  入选 HF Daily Papers；HF 热度: 22 upvotes (+4)；有代码实现；关键词(2): deployment, reasoning
• LoST: Level of Semantics Tokenization for 3D Shapes score 10
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)；有代码实现；关键词(1): compression
• Prompt-Free Universal Region Proposal Network score 9
  入选 HF Daily Papers；有代码实现；关键词(1): fine-tuning；顶会接收: CVPR
• Stereo World Model: Camera-Guided Stereo Video Generation score 9
  入选 HF Daily Papers；HF 热度: 10 upvotes (+3)；有代码实现；关键词(2): distillation, embodied
• Unified Spatio-Temporal Token Scoring for Efficient Video VLMs score 8
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现；关键词(4): scaling, lightweight, pruning, vision-language
• MOSS-TTS Technical Report score 8
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；有代码实现；关键词(2): deployment, pretraining
• Temporal Gains, Spatial Costs: Revisiting Video Fine-Tuning in Multimodal Large Language Models score 7
  入选 HF Daily Papers；HF 热度: 18 upvotes (+3)；关键词(2): serving, fine-tuning
• FINER: MLLMs Hallucinate under Fine-grained Negative Queries score 7
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现；关键词(2): finetuning, DPO
• VideoAtlas: Navigating Long-Form Video in Logarithmic Compute score 7
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现；关键词(1): scaling

也值得关注

• AdaZoom-GUI: Adaptive Zoom-based GUI Grounding with Instruction Refinement score 4
  机构: Tsinghua；关键词(3): deployment, GRPO, vision-language
• VirPro: Visual-referred Probabilistic Prompt Learning for Weakly-Supervised Monocular 3D Detection score 4
  关键词(2): pretraining, vision-language；顶会接收: CVPR
• EI: Early Intervention for Multimodal Imaging based Disease Recognition score 4
  关键词(2): edge, fine-tuning；顶会接收: CVPR
• PanoVGGT: Feed-Forward 3D Reconstruction from Panoramic Imagery score 4
  关键词(1): reasoning；顶会接收: CVPR
• Edit-As-Act: Goal-Regressive Planning for Open-Vocabulary 3D Indoor Scene Editing score 4
  关键词(1): reasoning；顶会接收: CVPR
• ReLaGS: Relational Language Gaussian Splatting score 4
  关键词(2): pruning, reasoning；顶会接收: CVPR
• STEP: Detecting Audio Backdoor Attacks via Stability-based Trigger Exposure Profiling score 4
  机构: Zhejiang University；关键词(1): deployment
• CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution score 4
  机构: Zhejiang University；关键词(1): reasoning
• Fine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients score 4
  关键词(4): quantization, deployment, latency, post-training；顶会接收: CVPR
• TINA: Text-Free Inversion Attack for Unlearned Text-to-Image Diffusion Models score 4
  关键词(2): deployment, text-to-image；顶会接收: CVPR
• CodeScout: An Effective Recipe for Reinforcement Learning of Code Search Agents score 4
  机构: Carnegie Mellon；关键词(2): agentic, coding
• Video Understanding: From Geometry and Semantics to Unified Models score 4
  机构: Cambridge；关键词(1): reasoning
• CARE: Covariance-Aware and Rank-Enhanced Decomposition for Enabling Multi-Head Latent Attention score 4
  关键词(1): fine-tune；顶会接收: ICLR
• AdaRadar: Rate Adaptive Spectral Compression for Radar-based Perception score 4
  关键词(3): compression, quantization, pruning；顶会接收: CVPR
• Sharpness-Aware Minimization in Logit Space Efficiently Enhances Direct Preference Optimization score 4
  关键词(1): DPO；顶会接收: ICLR