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。
Renrui Zhang, Zhengkai Jiang, Ziyu Guo +5 authors
A training-free and fine-tuning variant of the Segment Anything Model (SAM), PerSAM and PerSAM-F, achieves personalized image and video segmentation using a single reference image and minimal fine-tuning, improving performance on personalized and dreambooth applications.
18 篇论文 · 按点赞排序
Chen-Yu Lee, Chun-Liang Li, Hao Zhang +13 authors
FormNetV2 uses a centralized multimodal graph contrastive learning approach to improve self-supervised pre-training and achieves state-of-the-art performance on form understanding benchmarks with a compact model.
Connor Z. Lin, Koki Nagano, Jan Kautz +5 authors
A novel method for constructing implicit 3D morphable face models that combines the benefits of neural implicit representations and 3D morphable models, offering photo-realism, generalizability, and intuitive editing capabilities.
Shujian Zhang, Chengyue Gong, Lemeng Wu +2 authors
AutoML-GPT uses LLMs to automate the entire AI training pipeline from data processing to hyperparameter tuning, achieving high performance across various domains.
Heewoo Jun, Alex Nichol
Shap-E, a conditional generative model, generates 3D assets by training an encoder and a conditional diffusion model, achieving faster convergence and high-quality samples compared to explicit models.
Jun-Kun Chen, Jipeng Lyu, Yu-Xiong Wang
NeuralEditor introduces a novel rendering scheme for NeRFs using point clouds and K-D tree-guided density-adaptive voxels, enabling state-of-the-art shape editing and scene morphing with support for zero-shot inference and fine-tuning.
Telmo Pessoa Pires, Robin M. Schmidt, Yi-Hsiu Liao +1 authors
Language-Specific Transformer Layers enhance multilingual machine translation quality without increasing computational cost or parameters.
Wen Xiao, Yujia Xie, Giuseppe Carenini +1 authors
A tri-agent pipeline for enhancing customizability of large language model outputs includes a generator, an instructor, and an editor, with the instructor being optimized through reinforcement learning guided by the editor.
Saurabh Pujar, Luca Buratti, Xiaojie Guo +8 authors
A transformer-based model named Ansible Wisdom generates Ansible-YAML from natural language with performance similar to existing models, using a new dataset and novel performance metrics specifically for Ansible.
An-Chieh Cheng, Xueting Li, Sifei Liu +1 authors
Texture UV Radiance Fields (TUVF) enable controllable and transferable texture generation for 3D models by learning textures in a UV sphere space.
Tobias Kirschstein, Shenhan Qian, Simon Giebenhain +2 authors
A new multi-view capture setup and Dynamic Neural Radiance Fields using Hash Ensembles (NeRSemble) enable high-fidelity reconstruction and re-rendering of human heads over time.
Philipp Wu, Arjun Majumdar, Kevin Stone +4 authors
Masked Trajectory Models (MTM) are versatile sequential decision-making models that can perform tasks such as forward dynamics modeling, inverse dynamics modeling, and offline RL using generic masks and representations, matching or outperforming specialized methods.
Tizian Zeltner, Fabrice Rousselle, Andrea Weidlich +7 authors
Real-time rendering of complex scenes is achieved using learned hierarchical textures interpreted by neural decoders, supported by hardware-accelerated tensor operations in ray tracing shaders.
Basile Van Hoorick, Pavel Tokmakov, Simon Stent +2 authors
TCOW is a benchmark and model for visual tracking in cluttered environments with heavy occlusions and containment, using a mixture of synthetic and real datasets to evaluate transformer-based video models' performance in understanding object permanence.
Yijin Liu, Xianfeng Zeng, Fandong Meng +1 authors
BranchNorm dynamically adjusts the scaling of the non-residual branch in Transformers to improve training stability and convergence in deep models.
Yue Wu, So Yeon Min, Yonatan Bisk +5 authors
The PET framework leverages pre-trained large language models to simplify control tasks in non-text environments, improving generalization on instruction following benchmarks.
Deepak Narayanan, Keshav Santhanam, Peter Henderson +3 authors
A new metric called idealized runtime is proposed to compare inference efficiency across large language models, facilitating analysis of tradeoffs between runtime performance and model capabilities.
Zhiqing Sun, Yikang Shen, Qinhong Zhou +5 authors
SELF-ALIGN enhances AI assistants with minimal human supervision by leveraging LLMs for synthetic prompts, in-context learning, and self-alignment to improve helpfulness, ethics, and reliability.
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