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。
Adam Polyak, Amit Zohar, Andrew Brown +85 authors
Movie Gen, a suite of foundation models, generates high-quality videos with synchronized audio, excelling in various tasks through architectural, training, and technical innovations.
50 篇论文 · 按点赞排序
Yadong Li, Haoze Sun, Mingan Lin +24 authors
Baichuan-Omni, a 7B open-source Multimodal Large Language Model, excels in processing image, video, audio, and text, showcasing competitive performance across multimodal benchmarks.
Jinjie Ni, Yifan Song, Deepanway Ghosal +10 authors
MixEval-X is a benchmark designed to optimize and standardize multi-modal evaluations, addressing issues of inconsistency and bias in real-world applications.
Shuai Tan, Biao Gong, Xiang Wang +6 authors
Animate-X, based on latent diffusion models, generates high-quality videos for various character types by enhancing motion representation through the Pose Indicator and a new benchmark, demonstrating superior performance.
Junyan Ye, Baichuan Zhou, Zilong Huang +12 authors
LOKI is a benchmark for evaluating the synthetic data detection capabilities of large multimodal models across various modalities.
Jinbin Bai, Tian Ye, Wei Chow +6 authors
Meissonic, a non-autoregressive masked image model, achieves high-quality image generation by incorporating advanced techniques and outperforms existing models like SDXL.
Peng Xia, Siwei Han, Shi Qiu +9 authors
MMIE is a large-scale benchmark for evaluating interleaved multimodal comprehension and generation in Large Vision-Language Models using diverse data and reliable automated evaluation metrics.
Ziyue Li, Tianyi Zhou
Mixture-of-Experts (MoE) LLMs can serve as effective embedding models without finetuning, with their routing weights and hidden states complementing each other to enhance performance.
Sijie Cheng, Kechen Fang, Yangyang Yu +6 authors
VidEgoThink, a comprehensive benchmark for evaluating egocentric video understanding capabilities, highlights the need for advancements in Multi-modal Large Language Models for effective application in Embodied AI.
Guanting Dong, Xiaoshuai Song, Yutao Zhu +3 authors
VIF-RAG-QA and FollowRAG Benchmark enhance Instruction-Following alignment in RAG systems through automated instruction synthesis, verification, and evaluation.
Sijun Tan, Siyuan Zhuang, Kyle Montgomery +5 authors
JudgeBench is a benchmark for evaluating LLM-based judges using objective correctness for challenging tasks like knowledge, reasoning, math, and coding.
Zhuoqun Li, Xuanang Chen, Haiyang Yu +7 authors
StructRAG enhances LLMs by identifying optimal structure types, reconstructing documents into these structures, and inferring answers for knowledge-intensive reasoning tasks, achieving state-of-the-art performance.
Fengji Zhang, Linquan Wu, Huiyu Bai +6 authors
HumanEval-V provides a comprehensive benchmark for evaluating LMMs' diagram interpretation and reasoning abilities, particularly in coding contexts, revealing significant areas for improvement.
Zhiyuan Zhao, Hengrui Kang, Bin Wang +1 authors
DocLayout-YOLO improves document layout analysis speed and accuracy using mesh-candidate bestfit pre-training and global-to-local controllable receptive modules.
Guorui Zheng, Xidong Wang, Juhao Liang +3 authors
A novel MoE-based routing method and language family experts enhance the generalization of multilingual medical LLMs to low-resource languages.
Lijie Fan, Tianhong Li, Siyang Qin +6 authors
Autoregressive models for text-to-image generation achieve better visual quality with continuous tokens and random generation order compared to discrete tokens and raster order.
Chengyue Wu, Xiaokang Chen, Zhiyu Wu +8 authors
Janus, an autoregressive framework with separate visual encoding pathways within a unified transformer architecture, enhances performance in unified multimodal understanding and generation.
Zihan Zhou, Chong Li, Xinyi Chen +11 authors
A divide-and-conquer framework addresses long-range information loss in large language models by splittingdocuments into chunks and using structured protocols and calibration mechanisms to improve accuracy and performance.
Jiacheng Chen, Tianhao Liang, Sherman Siu +13 authors
MEGA-Bench evaluates vision-language models across over 500 real-world tasks with diverse output formats, providing a comprehensive capability report.
Yang Bai, Yang Zhou, Jun Zhou +3 authors
VITask enhances vision language models with task-specific models using exemplar prompting, response distribution alignment, and contrastive response tuning to improve task-specific performance across medical diagnosis datasets.
Fan Bu, Yuhao Zhang, Xidong Wang +3 authors
The paper outlines a roadmap and introduces a benchmark for advancing speech LLMs to integrate non-semantic information and abstract acoustic knowledge, highlighting current limitations and future directions.
Shwai He, Guoheng Sun, Zheyu Shen +1 authors
Investigation reveals significant redundancy in Transformer layers, including Attention and MLP, which can be pruned to enhance speed without substantial performance loss.
Litu Rout, Yujia Chen, Nataniel Ruiz +3 authors
Proposed RF inversion method using dynamic optimal control achieves top performance in image inversion and editing compared to existing diffusion model methods.
Bofei Gao, Feifan Song, Zhe Yang +17 authors
A new benchmark evaluates LLMs on Olympiad-level mathematics, revealing challenges even for advanced models.
Junpeng Liu, Tianyue Ou, Yifan Song +6 authors
MultiUI, a dataset from webpage UIs, enhances multimodal models' ability to handle text-rich visual environments, improving performance on web and non-web tasks.
Sicong Leng, Yun Xing, Zesen Cheng +7 authors
Investigation into hallucinations in large multimodal models highlights vulnerabilities in modality integration and suggests strategies for improved reliability.
Genta Indra Winata, Frederikus Hudi, Patrick Amadeus Irawan +48 authors
WorldCuisines, a large-scale multilingual and multicultural visual question answering benchmark, evaluates Vision Language Models' understanding of cuisine-specific knowledge across various languages and cultural contexts.
Shi Yu, Chaoyue Tang, Bokai Xu +8 authors
VisRAG, a vision-language model-based RAG pipeline, enhances document generation by directly embedding multi-modal documents and outperforms traditional text-based RAG systems.
Jian Xie, Kexun Zhang, Jiangjie Chen +5 authors
Research identifies constraints and question influence as critical factors hindering the planning capabilities of current language agents.
Nimrod Shabtay, Felipe Maia Polo, Sivan Doveh +8 authors
LiveXiv is a scalable, evolving benchmark using scientific ArXiv papers to generate visual question-answer pairs, providing a contamination-free evaluation for large multi-modal models.
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