9K Samples Rival R1, Most RL Gains Trace Back to SFT

Featured

• VGGT-Det: Mining VGGT Internal Priors for Sensor-Geometry-Free Multi-View Indoor 3D Object Detection score 13
  入选 HF Daily Papers; HF 热度: 29 upvotes (+4); 有代码实现; 关键词(1): deployment; 顶会接收: CVPR
• Spectral Attention Steering for Prompt Highlighting score 10
  入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 有代码实现; 关键词(1): latency; 顶会接收: ICLR
• LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model score 9
  入选 HF Daily Papers; HF 热度: 14 upvotes (+3); 有代码实现; 关键词(1): text-to-image
• CHIMERA: Compact Synthetic Data for Generalizable LLM Reasoning score 8
  入选 HF Daily Papers; HF 热度: 32 upvotes (+4); 关键词(5): fine-tuning, post-training, reasoning, open-source, synthetic data
• When Does RL Help Medical VLMs? Disentangling Vision, SFT, and RL Gains score 8
  入选 HF Daily Papers; HF 热度: 8 upvotes (+2); 有代码实现; 关键词(4): fine-tuning, post-training, reasoning, vision-language
• Learn Hard Problems During RL with Reference Guided Fine-tuning score 7
  入选 HF Daily Papers; HF 热度: 11 upvotes (+3); 关键词(2): fine-tuning, reasoning
• DIVA-GRPO: Enhancing Multimodal Reasoning through Difficulty-Adaptive Variant Advantage score 7
  机构: Huawei; 关键词(3): scaling, GRPO, reasoning; 顶会接收: ICLR
• Unified Vision-Language Modeling via Concept Space Alignment score 6
  入选 HF Daily Papers; HF 热度: 6 upvotes (+2); 关键词(4): instruction tuning, pre-training, vision-language, text-to-video
• ArtLLM: Generating Articulated Assets via 3D LLM score 5
  入选 HF Daily Papers; HF 热度: 3 upvotes (+1); 关键词(1): robotics

Also Worth Noting

• pySpatial: Generating 3D Visual Programs for Zero-Shot Spatial Reasoning score 4
  关键词(3): fine-tuning, code generation, reasoning; 顶会接收: ICLR
• ShiftLUT: Spatial Shift Enhanced Look-Up Tables for Efficient Image Restoration score 4
  关键词(4): compression, deployment, latency, edge; 顶会接收: CVPR
• Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards score 4
  关键词(2): post-training, text-to-image; 顶会接收: CVPR
• Dr.Occ: Depth- and Region-Guided 3D Occupancy from Surround-View Cameras for Autonomous Driving score 4
  关键词(1): MoE; 顶会接收: CVPR
• Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift score 4
  关键词(2): lightweight, deployment; 顶会接收: CVPR
• Beyond Global Similarity: Towards Fine-Grained, Multi-Condition Multimodal Retrieval score 4
  关键词(2): reasoning, vision-language; 顶会接收: CVPR
• CARD: Towards Conditional Design of Multi-agent Topological Structures score 4
  关键词(3): agentic, code generation, reasoning; 顶会接收: ICLR
• DeAR: Fine-Grained VLM Adaptation by Decomposing Attention Head Roles score 4
  关键词(1): vision-language; 顶会接收: CVPR
• Token-level Data Selection for Safe LLM Fine-tuning score 4
  关键词(1): fine-tuning; 顶会接收: ICLR
• RnG: A Unified Transformer for Complete 3D Modeling from Partial Observations score 4
  关键词(1): real-time; 顶会接收: CVPR
• Reasoning Boosts Opinion Alignment in LLMs score 4
  关键词(1): reasoning; 顶会接收: ICLR
• The Lattice Representation Hypothesis of Large Language Models score 4
  关键词(1): reasoning; 顶会接收: ICLR
• AgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models score 4
  关键词(2): pruning, vision-language; 顶会接收: ICLR
• FoSS: Modeling Long Range Dependencies and Multimodal Uncertainty in Trajectory Prediction via Fourier State Space Integration score 4
  关键词(2): state space, reasoning; 顶会接收: CVPR
• Attention Smoothing Is All You Need For Unlearning score 4
  关键词(1): distillation; 顶会接收: ICLR