图像生成 - arXivDaily 专题

2108.01806 2022-07-26 cs.CV cs.GR 62%

Neural Scene Decoration from a Single Photograph

Hong-Wing Pang, Yingshu Chen, Phuoc-Hieu Le, Binh-Son Hua, Duc Thanh Nguyen, Sai-Kit Yeung

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments ECCV 2022 paper. 14 pages of main content, 4 pages of references, and 11 pages of appendix

2112.02308 2022-07-25 cs.CV cs.GR 62%

MoFaNeRF: Morphable Facial Neural Radiance Field

Yiyu Zhuang, Hao Zhu, Xusen Sun, Xun Cao

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments accepted to ECCV2022; code available at http://github.com/zhuhao-nju/mofanerf

2004.02546 2022-07-05 cs.CV cs.GR 62%

GANSpace: Discovering Interpretable GAN Controls

Erik Härkönen, Aaron Hertzmann, Jaakko Lehtinen, Sylvain Paris

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Accepted to NeurIPS 2020

Journal ref Advances in Neural Information Processing Systems 33 (NeurIPS 2020), 9841-9850

2204.04455 2022-04-12 cs.GR cs.CV eess.IV 62%

Noise-based Enhancement for Foveated Rendering

Taimoor Tariq, Cara Tursun, Piotr Didyk

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments 14 pages including refences

2203.05349 2022-03-11 cs.MM cs.CV 62%

Two-stream Hierarchical Similarity Reasoning for Image-text Matching

Ran Chen, Hanli Wang, Lei Wang, Sam Kwong

专题命中文生图：text-to-image(abstract)；分类 cs.CV、cs.MM

2202.14020 2022-03-01 cs.CV cs.GR cs.LG 62%

State-of-the-Art in the Architecture, Methods and Applications of StyleGAN

Amit H. Bermano, Rinon Gal, Yuval Alaluf, Ron Mokady, Yotam Nitzan, Omer Tov, Or Patashnik, Daniel Cohen-Or

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

2202.13162 2022-03-01 cs.CV cs.AI cs.GR cs.LG 62%

Pix2NeRF: Unsupervised Conditional $π$-GAN for Single Image to Neural Radiance Fields Translation

Shengqu Cai, Anton Obukhov, Dengxin Dai, Luc Van Gool

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments 16 pages, 10 figures

2012.09855 2021-12-02 cs.CV cs.GR 62%

Infinite Nature: Perpetual View Generation of Natural Scenes from a Single Image

Andrew Liu, Richard Tucker, Varun Jampani, Ameesh Makadia, Noah Snavely, Angjoo Kanazawa

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments ICCV 2021 (oral); Project page: https://infinite-nature.github.io/; Video: https://www.youtube.com/watch?v=oXUf6anNAtc

2110.06688 2021-10-14 cs.CV cs.GR 62%

DeepVecFont: Synthesizing High-quality Vector Fonts via Dual-modality Learning

Yizhi Wang, Zhouhui Lian

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments SIGGRAPH Asia 2021 Technical Paper. Code: https://github.com/yizhiwang96/deepvecfont ; Homepage: https://yizhiwang96.github.io/deepvecfont_homepage/

2109.04654 2021-09-13 cs.GR cs.CV 62%

Per Garment Capture and Synthesis for Real-time Virtual Try-on

Toby Chong, I-Chao Shen, Nobuyuki Umetani, Takeo Igarashi

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Accepted to UIST2021. Project page: https://sites.google.com/view/deepmannequin/home

2006.16011 2021-03-30 cs.CV cs.GR 62%

Intrinsic Autoencoders for Joint Neural Rendering and Intrinsic Image Decomposition

Hassan Abu Alhaija, Siva Karthik Mustikovela, Justus Thies, Varun Jampani, Matthias Nießner, Andreas Geiger, Carsten Rother

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

2012.12884 2020-12-24 cs.CV cs.GR 62%

Vid2Actor: Free-viewpoint Animatable Person Synthesis from Video in the Wild

Chung-Yi Weng, Brian Curless, Ira Kemelmacher-Shlizerman

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Project Page: https://grail.cs.washington.edu/projects/vid2actor/ Supplementary Video: https://youtu.be/Zec8Us0v23o

1910.04639 2020-08-11 cs.CV cs.GR 62%

Visual Indeterminacy in GAN Art

Aaron Hertzmann

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Leonardo / SIGGRAPH 2020 Art Papers

Journal ref Leonardo, Volume 53, Issue 4, August 2020

2004.02867 2020-04-07 cs.CV cs.GR 62%

Rethinking Spatially-Adaptive Normalization

Zhentao Tan, Dongdong Chen, Qi Chu, Menglei Chai, Jing Liao, Mingming He, Lu Yuan, Nenghai Yu

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

1911.11378 2019-11-27 cs.LG cs.CV cs.MM eess.IV stat.ML 62%

Text2FaceGAN: Face Generation from Fine Grained Textual Descriptions

Osaid Rehman Nasir, Shailesh Kumar Jha, Manraj Singh Grover, Yi Yu, Ajit Kumar, Rajiv Ratn Shah

专题命中文生图：text-to-image(abstract)；分类 cs.CV、cs.MM

1908.02714 2019-08-12 cs.GR cs.CV 62%

Relighting Humans: Occlusion-Aware Inverse Rendering for Full-Body Human Images

Yoshihiro Kanamori, Yuki Endo

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Published at SIGGRAPH Asia 2018 (ACM Transactions on Graphics). Project page with codes, pretrained models, and human model lists is at http://kanamori.cs.tsukuba.ac.jp/projects/relighting_human/

1906.11478 2019-06-28 cs.CV cs.GR 62%

A Convolutional Decoder for Point Clouds using Adaptive Instance Normalization

Isaak Lim, Moritz Ibing, Leif Kobbelt

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments Symposium on Geometry Processing 2019

Journal ref Computer Graphics Forum 38 (5), 2019

1904.12620 2019-04-30 cs.CV cs.CR cs.IR cs.MM 62%

AnonymousNet: Natural Face De-Identification with Measurable Privacy

Tao Li, Lei Lin

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.MM

Comments CVPR-19 Workshop on Computer Vision: Challenges and Opportunities for Privacy and Security (CV-COPS 2019)

1808.06601 2018-12-04 cs.CV cs.GR cs.LG 62%

Video-to-Video Synthesis

Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Guilin Liu, Andrew Tao, Jan Kautz, Bryan Catanzaro

专题命中文生图：image synthesis(abstract)；分类 cs.CV、cs.GR

Comments In NeurIPS, 2018. Code, models, and more results are available at https://github.com/NVIDIA/vid2vid

2208.04139 2023-05-05 cs.CV cs.AI 61%

DALLE-URBAN: Capturing the urban design expertise of large text to image transformers

Sachith Seneviratne, Damith Senanayake, Sanka Rasnayaka, Rajith Vidanaarachchi, Jason Thompson

专题命中文生图：text-to-image(abstract)；分类 cs.CV；diffusion(comments)

Comments Accepted to DICTA 2022, released 11000+ environmental scene images generated by Stable Diffusion and 1000+ images generated by DALLE-2

2608.17564 2026-08-19 cs.CV cs.AI 新提交 57%

Where a New Concept Must Enter: Entry Point Gates Cross-Task Usability in Unified Multimodal Models

新概念必须进入之处：统一多模态模型中的入口门跨任务可用性

Zongyang Qiu, Yihan Wu, Kaixuan Fan, Bo Li, Hui Xiong

机构 * The Hong Kong University of Science and Technology (Guangzhou)（香港科技大学（广州））； Columbia University（哥伦比亚大学）； CUHK（香港中文大学）； University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结该研究通过分离统一多模态模型的理解与生成任务方向，发现跨任务可用性取决于概念绑定的入口层，提出的对齐目标可在极低损失下实现概念跨任务迁移。

Comments 27 pages, 10 figures

2608.15395 2026-08-18 cs.CV 新提交 57%

JoLT: Joint Latent Trajectories for Context-Guided High-Resolution Tiled Generation

JoLT：用于上下文引导的高分辨率分块生成的联合潜在轨迹

Mathis Koroglu, Guillaume Jeanneret, Hugo Caselles-Dupré, Matthieu Cord, Arnaud Dapogny

机构 * Obvious Research（奥布弗西斯研究公司）； Sorbonne Université（索邦大学）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结本文提出JoLT方法，通过联合去噪LR与HR潜在图像生成高分辨率图像，其生成的图像细节丰富、视觉效果佳，优于竞争基线，为艺术创作提供新方向。

Comments 25 pages, 10 figures, 7 tables. Accepted at the AI4VA Workshop at ECCV 2026. Project page: https://obvious-research.github.io/jolt/

2608.00410 2026-08-18 cs.AI cs.CL cs.CV 版本更新 57%

Where did the ambiguity go? Examining how multimodal models interpret polysemous words

歧义去了哪里？探究多模态模型如何解释多义词

Jasin Cekinmez, Addison J. Wu, Raja Marjieh, Thomas L. Griffiths

机构 * Princeton University（普林斯顿大学）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结该研究对比17个文本到图像模型和15个文本生成模型，发现多模态模型生成图像的词义多样性低于文本，揭示了基础模型在不同模态间意义表达的迁移 gap。

Comments Oral Presentation, Sci-FM Workshop @ COLM 2026

2608.13671 2026-08-17 cs.CV 新提交 57%

PROVE: Training-Free Prompt Recovery using Verifiable Evidence

PROVE：基于可验证证据的无训练提示词恢复

Rupayan Mallick, Mahsa Khoshnoodi, Sarah Adel Bargal

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结该研究提出无训练的黑盒提示词反转攻击PROVE，通过可验证场景描述重建提示词，在多数据集上优于基线方法，可用于版权保护相关研究。

2608.10908 2026-08-12 cs.CV cs.CL 新提交 57%

Order Matters: LVLMs as Judges for Temporal Reasoning in Image Sequences

顺序很重要：LVLMs作为图像序列时间推理的评判者

Martina Ianaro, Guilherme Fernandes, Maurizio Gabbrielli, Joao Magalhaes

机构 * University of Bologna（博洛尼亚大学）； NOVA School of Science and Technology（NOVA科技学院）； NOVA Laboratory for Computer Science and Informatics（NOVA计算机科学与信息实验室）

专题命中文生图：image synthesis(abstract)；分类 cs.CV

AI总结本研究发现大视觉语言模型（LVLMs）作为多模态评判者存在时间顺序判别缺陷，受首因、近因等结构性偏差影响，呼吁构建时间感知的视觉序列评估范式。

Comments 34 pages, camera-ready

2608.06264 2026-08-07 cs.CV cs.LG eess.IV 新提交 57%

OTLesMix: Wasserstein Barycenter and Optimal Transport Map for Synthetic Lesion Generation with Diverse Shapes and Locations

OTLesMix：用于生成形状与位置多样化合成病灶的Wasserstein重心与最优传输映射

Robin Trombetta, Carole Lartizien

机构 * Univ. Lyon（里昂大学）； INSA Lyon（里昂国立应用科学学院）； UCBL（里昂第一大学）； CNRS（法国国家科学研究中心）； Inserm（法国国家健康与医学研究院）； CREATIS

专题命中文生图：image synthesis(abstract)；分类 cs.CV

AI总结本研究提出OTLesMix方法，利用Wasserstein重心与最优传输映射生成多样化合成病灶，在三项脑病灶分割任务上使Dice分数提升2.9至6.6个百分点，性能优于现有混合类方法。

2608.05600 2026-08-07 cs.LG cs.AI cs.CV 新提交 57%

LC-GRPO: Bridging Train-Inference Gap for Flow-Based GRPO with Langevin Correction

LC-GRPO：通过朗之万校正弥合基于流的GRPO的训练-推理差距

Yingqing Guo, Hui Yuan, Zijian He, Mengdi Wang, Zheng Ding

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结 LC-GRPO 是带朗之万校正的基于流的 GRPO 框架，通过对齐推理的 ODE 欧拉步加朗之万校正，缩小流模型训练与推理的样本差距，在多任务上提升奖励优化并保留生成质量。

2608.05233 2026-08-07 cs.AI cs.CV 新提交 57%

Coherence-Oriented Dream Scene Visualisation

面向连贯性的梦境场景可视化

Azra Açıl, Simon Colton

机构 * School of Electronic Engineering and Computer Science, Queen Mary University of London（伦敦玛丽女王大学电子工程与计算机科学学院）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结该研究提出DSV系统，通过大型语言模型拆分梦境描述为四部分，结合文本到图像模型生成连贯的四面板图像序列，经DreamBank数据集50次可视化评估，用CLIP等模型指标验证了其质量、保真度与连贯性。

Comments short paper accepted at ICCC 2026

2608.00994 2026-08-04 cs.CV cs.CL 新提交 57%

Entity-Faithful Repair of Synthetic Supervision for Zero-Shot Image Captioning

零样本图像描述中合成监督的实体忠实修复

Zhiyue Liu, Wenkai Zhou, Jian Qin, Qipeng Jiang

机构 * Guangxi University（广西大学）； School of Computer, Electronics and Information（计算机与电子信息学院）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结针对零样本图像描述中合成监督的实体级错位问题，提出即插即用框架ReCap，通过实体级重对齐与自适应加权策略优化合成数据，在两类基准上实现最优性能。

Comments Accepted to the 34th ACM International Conference on Multimedia (ACM MM 2026). 16 pages, 7 figures

2608.00559 2026-08-04 cs.CV 新提交 57%

Test-Time Curriculum for Open-Set AIGC Detection

开放集AIGC检测的测试时课程

Yiqian Zhang, Zheyuan Gu, Xiangzhao Hao, Zefeng Zhang, Jingjia Mao, Jiahao Hu, Jiaxu Miao, Jun Yu, Zhenyu Zhang, Shuohuan Wang, Yu Sun

机构 * Baidu Inc.（百度公司）

专题命中文生图：text-to-image(abstract)；分类 cs.CV

AI总结本研究针对开放集AIGC检测的分布偏移问题，提出Test-Time Curriculum框架，结合跨尺度伪标签细化技术，构建AIGCGuard基准，经实验验证可显著提升未见生成器偏移下的检测性能。

1. 文生图 3480 篇

Neural Scene Decoration from a Single Photograph

MoFaNeRF: Morphable Facial Neural Radiance Field

GANSpace: Discovering Interpretable GAN Controls

Noise-based Enhancement for Foveated Rendering

Two-stream Hierarchical Similarity Reasoning for Image-text Matching

State-of-the-Art in the Architecture, Methods and Applications of StyleGAN

Pix2NeRF: Unsupervised Conditional $π$-GAN for Single Image to Neural Radiance Fields Translation

Infinite Nature: Perpetual View Generation of Natural Scenes from a Single Image

DeepVecFont: Synthesizing High-quality Vector Fonts via Dual-modality Learning

Per Garment Capture and Synthesis for Real-time Virtual Try-on

Intrinsic Autoencoders for Joint Neural Rendering and Intrinsic Image Decomposition

Vid2Actor: Free-viewpoint Animatable Person Synthesis from Video in the Wild

Visual Indeterminacy in GAN Art

Rethinking Spatially-Adaptive Normalization

Text2FaceGAN: Face Generation from Fine Grained Textual Descriptions

Relighting Humans: Occlusion-Aware Inverse Rendering for Full-Body Human Images

A Convolutional Decoder for Point Clouds using Adaptive Instance Normalization

AnonymousNet: Natural Face De-Identification with Measurable Privacy

Video-to-Video Synthesis

DALLE-URBAN: Capturing the urban design expertise of large text to image transformers

Where a New Concept Must Enter: Entry Point Gates Cross-Task Usability in Unified Multimodal Models

JoLT: Joint Latent Trajectories for Context-Guided High-Resolution Tiled Generation

Where did the ambiguity go? Examining how multimodal models interpret polysemous words

PROVE: Training-Free Prompt Recovery using Verifiable Evidence

Order Matters: LVLMs as Judges for Temporal Reasoning in Image Sequences

OTLesMix: Wasserstein Barycenter and Optimal Transport Map for Synthetic Lesion Generation with Diverse Shapes and Locations

LC-GRPO: Bridging Train-Inference Gap for Flow-Based GRPO with Langevin Correction

Coherence-Oriented Dream Scene Visualisation

Entity-Faithful Repair of Synthetic Supervision for Zero-Shot Image Captioning

Test-Time Curriculum for Open-Set AIGC Detection