论文来源 | 32B硬件代码开源进第一梯队，Agent难题仅23%

重点关注

CoME-VL: Scaling Complementary Multi-Encoder Vision-Language Learning score 6
入选 HF Daily Papers；有代码实现；关键词(3): scaling, pretraining, vision-language
Agentic-MME: What Agentic Capability Really Brings to Multimodal Intelligence? score 5
入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(1): agentic

也值得关注

Overcoming the "Impracticality" of RAG: Proposing a Real-World Benchmark and Multi-Dimensional Diagnostic Framework score 4
关键词(4): deployment, retrieval-augmented, RAG, reasoning；顶会接收: AAAI
Drift-Resilient Temporal Priors for Visual Tracking score 4
关键词(1): lightweight；顶会接收: CVPR
MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applications score 4
关键词(2): pre-training, pretraining；顶会接收: CVPR
DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object Detection score 4
关键词(2): deployment, reasoning；顶会接收: ICLR
A Unified Perspective on Adversarial Membership Manipulation in Vision Models score 4
关键词(1): serving；顶会接收: CVPR
Deformation-based In-Context Learning for Point Cloud Understanding score 4
关键词(1): reasoning；顶会接收: CVPR
Toward an Operational GNN-Based Multimesh Surrogate for Fast Flood Forecasting score 4
机构: NVIDIA；关键词(1): production
AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents score 4
入选 HF Daily Papers；关键词(1): tool use
InCoder-32B-Thinking: Industrial Code World Model for Thinking score 4
入选 HF Daily Papers；关键词(2): reasoning, open-source
Rascene: High-Fidelity 3D Scene Imaging with mmWave Communication Signals score 3
顶会接收: CVPR
THOM: Generating Physically Plausible Hand-Object Meshes From Text score 3
顶会接收: CVPR
High-resolution probabilistic estimation of three-dimensional regional ocean dynamics from sparse surface observations score 3
机构: Google
Unlocking Positive Transfer in Incrementally Learning Surgical Instruments: A Self-reflection Hierarchical Prompt Framework score 3
顶会接收: CVPR
UniSpector: Towards Universal Open-set Defect Recognition via Spectral-Contrastive Visual Prompting score 3
顶会接收: CVPR