Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
50 篇论文 · 按点赞排序
Weijia Mao, Hao Chen, Zhenheng Yang +1 authors
An adversarial reward learning framework in reinforcement learning for image generation improves image quality and aesthetics by using dense visual signals from vision foundation models.
Jun Cen, Siteng Huang, Yuqian Yuan +10 authors
A unified Vision-Language-Action (VLA) and world model, RynnVLA-002, jointly learns environmental dynamics and action planning, outperforming individual models in both simulation and real-world tasks.
Piaohong Wang, Motong Tian, Jiaxian Li +8 authors
O-Mem, an active user profiling memory framework, enhances contextual consistency and dynamic personalization in LLM-powered agents, improving performance on benchmarks and response efficiency.
Jie He, Richard He Bai, Sinead Williamson +3 authors
CLaRa enhances retrieval-augmented generation by introducing unified embedding-based compression and joint optimization, achieving state-of-the-art performance in QA benchmarks.
Kumar Krishna Agrawal, Longchao Liu, Long Lian +11 authors
Pillar-0, a radiology foundation model pretrained on diverse imaging datasets, outperforms existing models across various tasks and extends to new applications using RATE for label extraction.
Kateryna Chumachenko, Amala Sanjay Deshmukh, Jarno Seppanen +30 authors
Nemotron-Parse-1.1 is a lightweight OCR and document parsing model with improved capabilities in general OCR, markdown formatting, structured table parsing, and text extraction from images, using an encoder-decoder architecture.
Dilin Wang, Hyunyoung Jung, Tom Monnier +22 authors
WorldGen converts text prompts into interactive 3D environments using LLM-driven reasoning, procedural generation, diffusion-based 3D generation, and object-aware decomposition, enabling creators to build coherent, navigable worlds efficiently.
Teng Hu, Zhentao Yu, Guozhen Zhang +6 authors
Harmony addresses audio-visual synchronization in generative AI by introducing a Cross-Task Synergy training paradigm, Global-Local Decoupled Interaction Module, and Synchronization-Enhanced CFG to improve alignment and fidelity.
Hunyuan Vision Team, Pengyuan Lyu, Xingyu Wan +23 authors
HunyuanOCR, a lightweight Vision-Language Model, achieves state-of-the-art performance in OCR tasks through a unified end-to-end architecture combining Vision Transformer and lightweight LLM, supported by data-driven and RL strategies.
Ahmed Awadallah, Yash Lara, Raghav Magazine +9 authors
FaraGen creates synthetic datasets for computer use agents, enabling the training of efficient and high-performing models like Fara-7B on diverse web tasks, outperforming larger models on benchmarks.
Yolo Yunlong Tang, Daiki Shimada, Hang Hua +4 authors
Video-R4, a video reasoning LMM, uses iterative visual rumination to improve text-rich video QA by selecting, zooming, and re-encoding frames, achieving state-of-the-art results on various QA tasks.
Zhaolong Su, Wang Lu, Hao Chen +2 authors
UniGame, a self-adversarial post-training framework, improves consistency, understanding, generation, and robustness in unified multimodal models by introducing a lightweight perturber at the shared token interface.
Markus Pobitzer, Chang Liu, Chenyi Zhuang +3 authors
A unified framework using diffusion models with semantic control and cross-medium style augmentation generates consistent and high-fidelity multi-media painting processes, supported by a comprehensive dataset and evaluation metrics.
Qixun Wang, Yang Shi, Yifei Wang +5 authors
Monet, a training framework, enables MLLMs to reason in latent visual space using continuous embeddings, addressing challenges like computational cost and supervision, and outperforms on visual reasoning tasks.
Yusuf Çelebi, Mahmoud El Hussieni, Özay Ezerceli
PARROT evaluates the robustness of large language models against social pressure and sycophancy, revealing significant variability in model behavior and confidence shifts across different domains and authority templates.
Haoyang Hong, Jiajun Yin, Yuan Wang +14 authors
M-GRPO, an extension of Group Relative Policy Optimization for hierarchical multi-agent systems, improves stability and efficiency in tool-augmented reasoning tasks by aligning heterogeneous trajectories and decoupling agent training.
Min Zhao, Hongzhou Zhu, Yingze Wang +6 authors
UltraViCo addresses video length extrapolation by suppressing attention dispersion, improving quality and reducing repetition beyond training length.
Guangyuan Li, Siming Zheng, Shuolin Xu +5 authors
MagicWorld, an interactive video world model, integrates 3D geometry and historical retrieval to improve scene stability and continuity under user instructions.
Lun Huang, You Xie, Hongyi Xu +7 authors
Plan-X integrates a Semantic Planner and diffusion models to reduce visual hallucinations and improve instruction-aligned video generation by using multimodal semantic tokens.
Yang Zhou, Mingyu Zhao, Zhenting Wang +6 authors
M^3-Bench evaluates multimodal tool use with a focus on visual grounding, textual reasoning, and tool dependencies using a novel similarity-driven alignment method and interpretable metrics.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号