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
Sumit Chaturvedi, Mengwei Ren, Yannick Hold-Geoffroy +3 authors
SynthLight, a diffusion model, achieves realistic portrait relighting by combining synthetic data generation, multi-task training, and classifier-free guidance, producing photo-realistic illumination effects while preserving subject identity.
Yabo Zhang, Xinpeng Zhou, Yihan Zeng +3 authors
FramePainter, an image-to-video generation model, efficiently handles interactive image editing with sparse control and matching attention for seamless and coherent edits.
Vighnesh Subramaniam, Yilun Du, Joshua B. Tenenbaum +3 authors
A multi-agent society of language models self-improves through interactions, enabling specialization and better performance over rounds of fine-tuning compared to single-agent methods.
Zhaocheng Liu, Quan Tu, Wen Ye +7 authors
Experiments with a patient simulator show that the quality of the inquiry phase significantly affects diagnostic effectiveness in online medical consultations.
Richard Zhuang, Akshat Gupta, Richard Yang +3 authors
PokerBench evaluates the poker-playing abilities of large language models, demonstrating that fine-tuning is necessary for these models to perform optimally in such complex games.
Hwan Heo, Jangyeong Kim, Seongyeong Lee +3 authors
A two-stage framework, CaPa, uses a 3D latent diffusion model for geometry and a Spatially Decoupled Attention for textures, generating high-quality 3D assets efficiently.
Dongwon Kim, Ju He, Qihang Yu +4 authors
TA-TiTok, a text-aware transformer-based tokenizer, integrates textual information during decoding, improving training efficiency and enabling MaskGen models to achieve comparable performance using open datasets.
Abhilasha Ravichander, Shrusti Ghela, David Wadden +1 authors
HALoGEN is a benchmark for identifying and categorizing hallucinations in generative language models by verifying atomic units against knowledge sources.
Chenyang Si, Weichen Fan, Zhengyao Lv +3 authors
RepVideo enhances video generation by stabilizing semantic representations through enriched feature accumulation, improving spatial accuracy and temporal coherence in text-to-video diffusion models.
Liping Yuan, Jiawei Wang, Haomiao Sun +2 authors
Tarsier2, a large vision-language model, achieves state-of-the-art performance in video description and understanding by scaling pre-training data, fine-grained temporal alignment, and using DPO training.
Tianjin Huang, Ziquan Zhu, Gaojie Jin +3 authors
SPAM optimizer addresses training instability in LLMs by mitigating gradient spikes, enhancing performance and efficiency across various tasks.
Junjie He, Yuxiang Tuo, Binghui Chen +3 authors
AnyStory uses a unified approach with ReferenceNet and CLIP for encoding and routing subject features to achieve high-fidelity personalized text-to-image generation for multiple subjects.
Yuzhou Huang, Ziyang Yuan, Quande Liu +6 authors
ConceptMaster addresses identity decoupling and data scarcity in text-to-video generation using decoupled multi-concept embeddings and a comprehensive data pipeline, outperforming previous methods.
Tsai-Shien Chen, Aliaksandr Siarohin, Willi Menapace +7 authors
Video Alchemist uses a Diffusion Transformer with cross-attention to enable multi-subject personalization without test-time optimization, outperforming existing methods.
Xingyu Fu, Minqian Liu, Zhengyuan Yang +6 authors
ReFocus enhances multimodal LLMs for structured image understanding tasks by enabling visual edits and reasoning improvements through generated Python code, outperforming GPT-4 and other models.
Xingchen Liu, Piyush Tayal, Jianyuan Wang +10 authors
The introduction of uCO3D, a comprehensive 3D dataset with high-resolution videos and detailed annotations, enhances the performance of 3D deep learning and generative AI models.
David Noever, Forrest McKee
Research evaluates LLM security boundaries by embedding executable EICAR test files in JPEG images across multiple platforms, demonstrating consistent obfuscation and extraction capabilities.
Jingyuan Chen, Fuchen Long, Jie An +4 authors
Ouroboros-Diffusion improves long video generation by introducing latent sampling, Subject-Aware Cross-Frame Attention (SACFA), and self-recurrent guidance to enhance structural and content consistency.
Yunlong Tang, Junjia Guo, Pinxin Liu +14 authors
Generative AI, including large language models, multimodal models, and diffusion models, is transforming Cel-animation by automating labor-intensive tasks and improving accessibility for creators.
Xiangru Tang, Tianyu Hu, Muyang Ye +9 authors
ChemAgent, a framework incorporating a dynamic memory system, enhances LLMs in chemical reasoning tasks by decomposing them into sub-tasks and leveraging a library for improved accuracy and efficiency.
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