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
Joo Chan Lee, Daniel Rho, Xiangyu Sun +2 authors
A framework for efficient 3D scene representation using a learnable mask strategy, grid-based neural fields, and residual vector quantization to reduce storage and improve rendering speed while maintaining high-quality output.
Zhiyu Tan, Xiaomeng Yang, Luozheng Qin +1 authors
A new, high-quality dataset called VidGen-1M improves training for text-to-video models by ensuring better video quality, detailed captions, and temporal consistency.
Zilyu Ye, Jinxiu Liu, Ruotian Peng +9 authors
A new dataset Openstory++ and benchmark Cohere-Bench improve image generation by enhancing instance feature labeling and evaluating consistency in long multimodal contexts.
Armel Zebaze, Benoît Sagot, Rachel Bawden
In-context translation using sentence embeddings improves machine translation performance, particularly for low-resource languages, by balancing selection pool diversity and quality.
Dong Huo, Zixin Guo, Xinxin Zuo +6 authors
TexGen, a novel framework using a text-to-image diffusion model, generates high-quality, viewpoint-consistent 3D textures from textual descriptions, improving upon current methods.
Orest Kupyn, Eugene Khvedchenia, Christian Rupprecht
A large synthetic dataset using diffusion models for human head detection and 3D mesh estimation is introduced, enabling a new model to achieve strong performance on real images across various tasks.
Jiazhi Guan, Zhiliang Xu, Hang Zhou +10 authors
ReSyncer is a unified framework that uses a style-injected Transformer and a re-configured Style-based generator to produce high-fidelity lip-synced videos with various appealing properties.
Jack Lu, Ryan Teehan, Mengye Ren
ProCreate enhances sample diversity and creativity in diffusion-based image generation by repelling generated embeddings from reference images and prevents data reproduction.
Benno Weck, Ilaria Manco, Emmanouil Benetos +3 authors
MuChoMusic is a benchmark that evaluates multimodal language models focusing on audio understanding through multiple-choice questions covering various musical and contextual dimensions.
Beomseok Lee, Ioan Calapodescu, Marco Gaido +2 authors
A multilingual spoken language understanding dataset developed for evaluating foundation models across languages and tasks, offering scenarios for zero-shot, few-shot, and full fine-tuning.
Hugo Blanc, Jean-Emmanuel Deschaud, Alexis Paljic
The work introduces a physically consistent volumetric rendering method using differentiable ray casting of irregularly distributed Gaussians, achieving high-quality rendering with fast inference.
Jiaxi Yang, Binyuan Hui, Min Yang +3 authors
A synthetic data approach combining strong and weak models improves open-source text-to-SQL models' performance and domain generalization.
Boxi Cao, Mengjie Ren, Hongyu Lin +4 authors
StructEval provides a comprehensive, robust, and consistent evaluation framework for large language models by assessing multiple cognitive levels and critical concepts, reducing data contamination and bias.
Ling Yue, Sixue Xing, Yingzhou Lu +1 authors
BioMamba, a pre-trained model based on Mamba architecture, excels in biomedical text mining, outperforming other models in perplexity and cross-entropy loss.
Ruining Li, Chuanxia Zheng, Christian Rupprecht +1 authors
Puppet-Master is an interactive video generative model that uses fine-tuned diffusion with part-level motion and all-to-first attention, creating realistic videos based on sparse motion control.
Saman Kazemkhani, Aarav Pandya, Daphne Cornelisse +2 authors
GPUDrive, a GPU-accelerated multi-agent simulator, enables efficient training of reinforcement learning agents on the Waymo Motion dataset, reducing training time significantly.
Zhuofan Zhang, Ziyu Zhu, Pengxiang Li +6 authors
A new dataset and task for sequential task-oriented grounding in 3D environments are introduced to advance embodied AI, showcasing challenges for current models in dynamic, instruction-following tasks.
Karn N. Watcharasupat, Chih-Wei Wu, Iroro Orife
The query-based Banquet model outperforms the dedicated-decoder Bandit model in a four-stem Cinematic Audio Source Separation task by better aligning features through a band-agnostic FiLM layer.
Tomoyuki Suzuki, Kotaro Kikuchi, Kota Yamaguchi
This paper presents an approach to decomposing animated graphics into sprites, a set of basic elements or layers. Our approach builds on the optimization of sprite parameters to fit the raster video. For efficiency, we assume static textures for sprites to reduce the search space while preventing artifacts using a texture prior model. To further speed up the optimization, we introduce the initialization of the sprite parameters utilizing a pre-trained video object segmentation model and user input of single frame annotations. For our study, we construct the Crello Animation dataset from an online design service and define quantitative metrics to measure the quality of the extracted sprites. Experiments show that our method significantly outperforms baselines for similar decomposition tasks in terms of the quality/efficiency tradeoff.
Yuxuan Wang, Alan Yuille, Zhuowan Li +1 authors
An ExoViP method enhances vision-language programming by incorporating verification modules to correct planning and execution errors, improving compositional reasoning tasks.
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