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
Xuan Gao, Haiyao Xiao, Chenglai Zhong +3 authors
PortraitGen achieves consistent and expressive stylization in portrait video editing by using a 3D Gaussian field and a Neural Gaussian Texture mechanism, enabling sophisticated style editing and high rendering speed.
Ankit Dhiman, Manan Shah, Rishubh Parihar +3 authors
A diffusion-based generative model issues high-fidelity mirror reflections using a novel depth-conditioned inpainting method on a large synthetic dataset, enabling user control over mirror placement.
Yukun Huang, Jianan Wang, Ailing Zeng +3 authors
DreamWaltz-G generates and animates high-quality 3D avatars from text using skeleton-guided score distillation and hybrid 3D Gaussian avatar representation, enhancing consistency and expressiveness.
Michal Nazarczuk, Thomas Tanay, Sibi Catley-Chandar +3 authors
A new dataset and benchmark, Sparse Rendering (SpaRe), is introduced for sparse or few-shot neural rendering, providing high-quality synthetic scenes for reproducible evaluation and comparison of performance.
Chris Careaga, Yağız Aksoy
A model decomposes images into diffuse albedo, colorful shading, and specular components, facilitating illumination-aware image editing despite limited data.
Xinghua Zhang, Haiyang Yu, Yongbin Li +3 authors
Conversation Analysis task is systematized into four key steps, leveraging large language models to bridge the gap between research and business by focusing on causality and strategic tasks.
Aiping Zhang, Zongsheng Yue, Renjing Pei +2 authors
The proposed one-step super-resolution model enhances efficiency and quality by using a degradation-guided LoRA module and an online negative sample generation strategy in combination with classifier-free guidance.
Yehonathan Litman, Or Patashnik, Kangle Deng +4 authors
The study introduces MaterialFusion, an inverse rendering pipeline that utilizes a 2D diffusion model, StableMaterial, to enhance the accuracy of albedo and material property estimation, thereby improving relightability of rendered objects under new lighting conditions.
Ming Wang, Yuanzhong Liu, Xiaoyu Liang +8 authors
LangGPT and Minstrel enhance LLM performance by providing structural prompts that are more user-friendly and adaptable for non-AI experts.
Benjamin Clavié, Antoine Chaffin, Griffin Adams
A clustering-based token pooling method significantly reduces the storage and memory requirements of ColBERT indexes without compromising retrieval performance.
Huy Nhat Phan, Phong X. Nguyen, Nghi D. Q. Bui
HyperAgent, a multi-agent system, achieves state-of-the-art performance across diverse software engineering tasks using four specialized agents, potentially revolutionizing AI-assisted development.
Pei Xu, Ruocheng Wang
The approach uses cooperative learning of individual hand policies for bimanual control, enhancing training efficiency and demonstrated in guitar-playing tasks.
Juan Ossa, Eren Doğan, Alex Birch +1 authors
Changes to SDXL were implemented in the training of NovelAI Diffusion V3 to improve anime image generation.
Penghao Wang, Zhirui Zhang, Liao Wang +5 authors
A novel approach, V³, enables high-quality mobile rendering and streaming of volumetric videos by streaming dynamic Gaussians as 2D videos, using hardware video codecs and a two-stage training strategy to optimize motion and appearance.
Hui En Pang, Shuai Liu, Zhongang Cai +3 authors
Disco4D is a Gaussian Splatting framework for 4D human generation and animation from a single image, which disentangles clothing and body models, uses diffusion models for enhanced generation, and supports dynamic human animation.
Faria Huq, Abdus Samee, David Chuan-en Lin +2 authors
Video has become a popular media form for information sharing and consumption. However, taking notes while watching a video requires significant time and effort. To address this, we propose a novel interactive system, NoTeeline, for taking real-time, personalized notes. NoTeeline lets users quickly jot down keypoints (micronotes), which are automatically expanded into full-fledged notes that capture the content of the user's micronotes and are consistent with the user's writing style. In a within-subjects study (N=12), we found that NoTeeline helps users create high-quality notes that capture the essence of their micronotes with a higher factual correctness (93.2%) while accurately reflecting their writing style. While using NoTeeline, participants experienced significantly reduced mental effort, captured satisfactory notes while writing 47% less text, and completed notetaking with 43.9% less time compared to a manual notetaking baseline.
Benjamin Feuer, Micah Goldblum, Teresa Datta +5 authors
SOS-Bench evaluates the effectiveness of preference optimization methods in aligning LLMs, finding that LLM-judgment preferences do not correlate well with safety, knowledge, and instruction following, and that supervised fine-tuning has the greatest impact on alignment.
Maria Rigaki, Carlos Catania, Sebastian Garcia
Hackphyr, a locally fine-tuned LLM with 7 billion parameters, performs comparably to larger commercial models like GPT-4 and outperforms GPT-3.5-turbo and Q-learning agents in cybersecurity tasks using a custom dataset and runs on a single GPU.
Chen Tessler, Yunrong Guo, Ofir Nabati +2 authors
MaskedMimic unifies physics-based character control by formulating it as a motion inpainting problem, enabling versatile and intuitive character animation across various modalities and tasks.
Zahra Sepasdar, Sushant Gautam, Cise Midoglu +2 authors
Structured-GraphRAG, a framework using multiple knowledge graphs, enhances information retrieval in natural language queries, improving efficiency and reducing response times over traditional methods.
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