图像生成 - arXivDaily 专题

1905.09400 2019-05-24 cs.CV 57%

AttentionRNN: A Structured Spatial Attention Mechanism

Siddhesh Khandelwal, Leonid Sigal

专题命中效率与蒸馏：image generation(abstract)；分类 cs.CV

1904.09763 2019-05-06 cs.CV 57%

Water-Filling: An Efficient Algorithm for Digitized Document Shadow Removal

Seungjun Jung, Muhammad Abul Hasan, Changick Kim

专题命中效率与蒸馏：diffusion(abstract)；分类 cs.CV

Comments Accepted at Asian Conference on Computer Vision (2018)

1812.09916 2019-02-11 cs.LG cs.CV stat.ML 57%

Improving MMD-GAN Training with Repulsive Loss Function

Wei Wang, Yuan Sun, Saman Halgamuge

专题命中效率与蒸馏：image generation(abstract)；分类 cs.CV

Comments Published as a conference paper at ICLR 2019

1810.01367 2018-10-23 cs.LG cs.CV stat.ML 57%

FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models

Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, David Duvenaud

专题命中效率与蒸馏：image generation(abstract)；分类 cs.CV

Comments 8 Pages, 6 figures

1806.06710 2018-06-19 cs.GR 57%

End-to-end Sampling Patterns

Thomas Leimkühler, Gurprit Singh, Karol Myszkowski, Hans-Peter Seidel, Tobias Ritschel

专题命中效率与蒸馏：image synthesis(abstract)；分类 cs.GR

1711.10742 2017-11-30 cs.CV 57%

Pipeline Generative Adversarial Networks for Facial Images Generation with Multiple Attributes

Ziqiang Zheng, Zhibin Yu, Haiyong Zheng, Chao Wang, Nan Wang

专题命中效率与蒸馏：image generation(abstract)；分类 cs.CV

Comments 9 pages, 10 figures

1704.08992 2017-05-01 cs.CV 57%

A Unified Approach of Multi-scale Deep and Hand-crafted Features for Defocus Estimation

Jinsun Park, Yu-Wing Tai, Donghyeon Cho, In So Kweon

专题命中效率与蒸馏：image generation(abstract)；分类 cs.CV

Comments 10 pages, 14 figures. To appear in CVPR 2017. Project page : https://github.com/zzangjinsun/DHDE_CVPR17

1701.05595 2017-01-23 cs.CV 57%

Fast and Efficient Skin Detection for Facial Detection

Mohammad Reza Mahmoodi

专题命中效率与蒸馏：diffusion(abstract)；分类 cs.CV

1609.02072 2016-09-08 cs.GR 57%

Sampling BSSRDFs with non-perpendicular incidence

Etienne Ferrier

专题命中效率与蒸馏：diffusion(abstract)；分类 cs.GR

Comments Supervised by Prof. Wenzel Jakob (EPFL). Contains 9 pages, 13 figures, 4 tables, 3 algorithms

1406.6273 2014-06-25 cs.CV 57%

Image Completion for View Synthesis Using Markov Random Fields and Efficient Belief Propagation

Julian Habigt, Klaus Diepold

专题命中效率与蒸馏：inpainting(abstract)；分类 cs.CV

Comments Published version: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6738439

Journal ref Proc. 20th IEEE International Conference on Image Processing (2013) 2131-2134

2410.15342 2025-03-10 cs.SD cs.LG eess.AS 56%

ConSinger: Efficient High-Fidelity Singing Voice Generation with Minimal Steps

Yulin Song, Guorui Sang, Jing Yu, Chuangbai Xiao

专题命中效率与蒸馏：diffusion(abstract,comments)

Comments Singing voice synthesis, Consistency models, Shallow Diffusion Mechanism; Accepted by ICASSP 2025

2403.15760 2024-08-20 cs.AI cs.DC 56%

An Upload-Efficient Scheme for Transferring Knowledge From a Server-Side Pre-trained Generator to Clients in Heterogeneous Federated Learning

Jianqing Zhang, Yang Liu, Yang Hua, Jian Cao

专题命中效率与蒸馏：diffusion(abstract,comments)

Comments Accepted by CVPR2024. We have incorporated additional analysis for the Stable Diffusion experiments in Appendix A

gr-qc/9912004 2009-11-30 gr-qc 56%

An Efficient Method for Fully Relativistic Simulations of Coalescing Binary Neutron Stars

Walter Landry, Saul A. Teukolsky

专题命中效率与蒸馏：diffusion(abstract,comments)

Comments 10 pages, 24 figures, fixed some typos, added a paragraph about the diffusion method and time stepping

2608.19906 2026-08-21 cs.LG 新提交 50%

PETA:Parameter-Efficient Test-Time Adaptation for Virtual Screening

PETA：用于虚拟筛选的参数高效测试时自适应方法

Jia-Qi Lin, Yinghua Yao, Chang-Dong Wang, Yew-Soon Ong, Yuangang Pan

专题命中效率与蒸馏：diffusion(abstract)

AI总结本研究提出参数高效测试时自适应框架PETA，仅更新预训练虚拟筛选模型约0.03%的LayerNorm参数，在测试时通过构建特定负样本等方式自适应，性能优于预训练及全重训练基线。

2608.19209 2026-08-21 physics.comp-ph physics.chem-ph 新提交 50%

Physics-Informed Neural Networks as Fast Surrogate Models for Electrochemical Flow Reactors

用于电化学流动反应器快速替代模型的物理信息神经网络

Eric Fernández-García, Miguel Modestino, Sergio Maldonado

专题命中效率与蒸馏：diffusion(abstract)

AI总结本研究提出物理信息神经网络（PINN），将其作为电化学流动反应器的快速替代模型，经验证精度高、速度快，为低计算成本的数字孪生建模提供基础。

2608.19116 2026-08-20 cond-mat.mtrl-sci 新提交 50%

JANUS: A Multi-modal Foundation Neural Sampler for Disordered Materials

JANUS：面向无序材料的多模态基础神经采样器

Denis Blessing, Mouyang Cheng, Maximilian Schebek, Jutta Rogal, Mingda Li, Carles Domingo-Enrich, Yuanqi Du

专题命中效率与蒸馏：diffusion(abstract)

AI总结研究人员开发JANUS多模态神经采样器，用于无序材料的热力学采样，其以远少于传统方法的能量评估次数重现平衡性质，可用于合金逆设计及半导体缺陷探索，为相关研究提供了基础。

2608.18954 2026-08-20 astro-ph.HE 新提交 50%

Constraining Cosmic-Ray Acceleration and Escape in Middle-Aged Supernova Remnants with GeV-TeV Gamma-Ray Observations

利用GeV-TeV伽马射线观测约束中年超新星遗迹中的宇宙射线加速与逃逸

Siyu Chen, Bing Theodore Zhang, Yi Xing, Siming Liu

专题命中效率与蒸馏：diffusion(abstract)

AI总结本研究通过对四个中年超新星遗迹的伽马射线辐射开展时变研究，结合多望远镜观测，约束了宇宙射线加速与逃逸参数，揭示逃逸宇宙射线对其甚高能辐射及中微子通量的重要影响。

Comments 11 pages, 11 figures

2608.17022 2026-08-19 math.PR math.AP 新提交 50%

Sharp hypocoercive convergence estimates for underdamped Langevin dynamics with specular reflection

Hengrong Du, Qi Feng, Lingjiong Zhu

专题命中效率与蒸馏：diffusion(abstract)

Comments 32 pages

2509.05799 2026-08-18 physics.plasm-ph cs.AI cs.LG physics.comp-ph 50%

Hybrid Fourier Neural Operator-Plasma Fluid Model for Fast and Accurate Multiscale Simulations of High Power Microwave Breakdown

Kalp Pandya, Pratik Ghosh, Ajeya Mandikal, Shivam Gandha, Bhaskar Chaudhury

机构 * Group in Computational Science and HPC（计算科学与高性能计算组）； DA-IICT ； Dhirubhai Ambani University（迪鲁布希阿尼大学）； Smart Energy Learning Center（智能能源学习中心）； DAU

专题命中效率与蒸馏：diffusion(abstract)

Journal ref Engineering Applications of Artificial Intelligence, Volume 167, Part 2, 1 March 2026, 113889

2608.13932 2026-08-17 cs.LG 新提交 50%

Post-training Quantization for Hybrid Iterative Generative Models

混合迭代生成模型的训练后量化

Jing Gao, Junyi Wu, Wei Wang, Yan Yan, Yao Zhao

机构 * Institute of Information Science, Beijing Jiaotong University（北京交通大学信息科学研究所）； University of Illinois Chicago（伊利诺伊大学芝加哥分校）

专题命中效率与蒸馏：diffusion(abstract)

AI总结针对混合迭代生成模型训练后量化易引发模型崩溃的问题，提出HyGenQ框架，通过分层聚类解耦与缩放重校准，成功将其量化至8位精度且性能优于现有方法。

2607.27431 2026-08-14 cs.LG cs.AI 版本更新 50%

SE(3)-MeanFlow: Few-Step Protein Backbone Generation on Lie Groups

SE(3)-MeanFlow：李群上的少步蛋白质主链生成

Yikun Bai, Binghang Lu, Yikai Liu, Elaheh Akbari, Soheil Kolouri, Linxuan Wang, Ping He, Shuchan Wang, Ruqi Zhang, Guang Lin

机构 * Purdue University（普渡大学）； Vanderbilt University（范德堡大学）； Freie Universität Berlin（柏林自由大学）

专题命中效率与蒸馏：diffusion(abstract)

AI总结 SE(3)-MeanFlow将MeanFlow扩展到李群几何，通过新训练目标实现少步蛋白质主链生成，性能优于多倍采样的流匹配基线，适配高通量设计需求。

2511.06239 2026-08-14 stat.ML cs.LG 版本更新 50%

Functional Adjoint Sampler: Scalable Sampling on Infinite Dimensional Spaces

功能伴随采样器：无穷维空间上的可扩展采样方法

Byoungwoo Park, Juho Lee, Guan-Horng Liu

机构 * FAIR at Meta（Meta 的 FAIR）

专题命中效率与蒸馏：diffusion(abstract)

AI总结针对无穷维函数空间吉布斯分布采样的理论缺口，提出基于随机最优控制的功能伴随采样器FAS，在合成势与真实分子系统上实现了更优的转移路径采样性能。

2608.11019 2026-08-12 cs.LG 新提交 50%

DEFT: Data-Efficient Frequency-domain Top-k Sampling via Inverse Discrete Fourier Transform for Spatiotemporal Dynamical Systems Modeling

DEFT：用于时空动力系统建模的、基于逆离散傅里叶变换的数据高效频域Top-k采样方法

Hengbo Xiao, Jiale Liu, Jiahao Song, Guannan He

机构 * Peking University（北京大学）

专题命中效率与蒸馏：diffusion(abstract)

AI总结 DEFT是一种频域数据采样方法，通过逆离散傅里叶变换生成物理一致的训练数据，可减少数据需求且提升泛化能力，在多个PDE及电池退化系统建模任务中表现优异。

2608.10384 2026-08-12 cs.LG 新提交 50%

Generator-Guided Inverse Sampling for Lévy-Driven Generative Models

用于Lévy驱动生成模型的生成器引导逆采样

Tianfu Qi, Jun Wang, Jun Zhang

机构 * University of Electronic Science and Technology of China（电子科技大学）； Hong Kong University of Science and Technology（香港科技大学）

专题命中效率与蒸馏：diffusion(abstract)

AI总结本文针对Lévy驱动生成模型的非局域反向过程挑战，提出生成器引导的结构化逆采样器，将动力学分解为三类分量，结合神经网络与解析分布实现高效采样，在OFDM-SISO信道估计中表现出稳健性能与良好权衡。

2504.21784 2026-08-12 math.NA cs.NA 版本更新 50%

Efficient, Nonlinear Second Moment Methods for Multigroup Thermal Radiative Transfer

高效非线性二阶矩方法用于多群热辐射传输

Samuel Olivier, James S. Warsa, HyeongKae Park

专题命中效率与蒸馏：diffusion(abstract)

AI总结本文提出一种高效的非线性加速框架，用于多群热辐射传输，通过高阶离散 ordinates 传输与灰度扩散低阶系统结合，实现稳定求解并提升鲁棒性。

2608.08638 2026-08-11 cs.SD cs.AI cs.CL 新提交 50%

CuteTTS: Efficient and High-Quality Speech Synthesis via Autoregressive Modeling of Continuous Latents

CuteTTS：基于连续隐变量自回归建模的高效高保真语音合成

Yuqian Zhang, Yao Shi, Kexin Huang, Botian Jiang, Zhe Xu, Yiwei Zhao, Min Liang, Shuang Chen, Xipeng Qiu

机构 * Shanghai Innovation Institute（上海创新研究院）； OPPO AI Center（OPPO人工智能中心）； Fudan University（复旦大学）

专题命中效率与蒸馏：diffusion(abstract)

AI总结 CuteTTS是结合语义对齐因果VAE隐变量等的连续自回归TTS系统，经引导步蒸馏后，在保持相当质量的同时降低了延迟，实现了高保真与低延迟的平衡。

Comments 20 pages, 6 figures, 10 tables. Technical report

2608.08129 2026-08-11 cs.HC 新提交 50%

Understanding Security and Privacy Perceptions of Content Creators Regarding AI Labels of AI-Generated Content

理解内容创作者对AI生成内容的AI标签的安全与隐私认知

Shuning Zhang, Hui Wang, Rongjun Ma, Xin Yi, Kanye Ye Wang, Robert Xiao, Hewu Li

专题命中效率与蒸馏：image generation(abstract)

AI总结本文通过对21位AIGC创作者访谈及多平台图像测试，揭示创作者对AI标签的认知偏差与规避行为，提出应开发适配创作者动机的隐式AI标签。

2608.07648 2026-08-11 stat.ML cond-mat.dis-nn cond-mat.stat-mech cs.LG physics.comp-ph 新提交 50%

Leveraging generative models to assist Monte Carlo sampling

利用生成模型辅助蒙特卡洛采样

Marylou Gabrié

专题命中效率与蒸馏：diffusion(abstract)

AI总结本教程综述介绍了机器学习与计算统计物理交叉领域的新范式——用生成模型辅助蒙特卡洛采样，探讨相关方法方向及优劣势，为相关领域研究者提供入门基础。

2603.00040 2026-08-11 cs.LG cs.AI 版本更新 50%

Attn-QAT: 4-Bit Attention With Quantization-Aware Training

Attn-QAT：4位注意力与量化感知训练

Peiyuan Zhang, Matthew Noto, Wenxuan Tan, Chengquan Jiang, Will Lin, Wei Zhou, Hao Zhang

机构 * Stanford University（斯坦福大学）； University of Wisconsin-Madison（威斯康星大学麦迪逊分校）； Georgia Institute of Technology（佐治亚理工学院）

专题命中效率与蒸馏：diffusion(abstract)

AI总结 Attn-QAT通过稳定4位量化感知训练实现高效的注意力计算，提升模型推理速度并保持性能。

2608.07329 2026-08-10 math.OC math.AP 新提交 50%

Sampling and Optimization meet Enhanced Flows

采样与优化方法结合增强流

Yuan Gao, Siming He, Eitan Tadmor

专题命中效率与蒸馏：diffusion(abstract)

AI总结本文针对采样与优化中核心的Gibbs概率测度计算问题，引入带增强耗散的输运-扩散动力学，设计了对应的采样数值算法与粒子系统。

1. 效率与蒸馏 2551 篇

AttentionRNN: A Structured Spatial Attention Mechanism

Water-Filling: An Efficient Algorithm for Digitized Document Shadow Removal

Improving MMD-GAN Training with Repulsive Loss Function

FFJORD: Free-form Continuous Dynamics for Scalable Reversible Generative Models

End-to-end Sampling Patterns

Pipeline Generative Adversarial Networks for Facial Images Generation with Multiple Attributes

A Unified Approach of Multi-scale Deep and Hand-crafted Features for Defocus Estimation

Fast and Efficient Skin Detection for Facial Detection

Sampling BSSRDFs with non-perpendicular incidence

Image Completion for View Synthesis Using Markov Random Fields and Efficient Belief Propagation

ConSinger: Efficient High-Fidelity Singing Voice Generation with Minimal Steps

An Upload-Efficient Scheme for Transferring Knowledge From a Server-Side Pre-trained Generator to Clients in Heterogeneous Federated Learning

An Efficient Method for Fully Relativistic Simulations of Coalescing Binary Neutron Stars

PETA:Parameter-Efficient Test-Time Adaptation for Virtual Screening

Physics-Informed Neural Networks as Fast Surrogate Models for Electrochemical Flow Reactors

JANUS: A Multi-modal Foundation Neural Sampler for Disordered Materials

Constraining Cosmic-Ray Acceleration and Escape in Middle-Aged Supernova Remnants with GeV-TeV Gamma-Ray Observations

Sharp hypocoercive convergence estimates for underdamped Langevin dynamics with specular reflection

Hybrid Fourier Neural Operator-Plasma Fluid Model for Fast and Accurate Multiscale Simulations of High Power Microwave Breakdown

Post-training Quantization for Hybrid Iterative Generative Models

SE(3)-MeanFlow: Few-Step Protein Backbone Generation on Lie Groups

Functional Adjoint Sampler: Scalable Sampling on Infinite Dimensional Spaces

DEFT: Data-Efficient Frequency-domain Top-k Sampling via Inverse Discrete Fourier Transform for Spatiotemporal Dynamical Systems Modeling

Generator-Guided Inverse Sampling for Lévy-Driven Generative Models

Efficient, Nonlinear Second Moment Methods for Multigroup Thermal Radiative Transfer

CuteTTS: Efficient and High-Quality Speech Synthesis via Autoregressive Modeling of Continuous Latents

Understanding Security and Privacy Perceptions of Content Creators Regarding AI Labels of AI-Generated Content

Leveraging generative models to assist Monte Carlo sampling

Attn-QAT: 4-Bit Attention With Quantization-Aware Training

Sampling and Optimization meet Enhanced Flows