图像生成 - arXivDaily 专题

2412.16906 2026-03-31 cs.CV 86%

Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Text-to-Image Generation

自修正流蒸馏用于一致的一步和少步文本到图像生成

Quan Dao, Hao Phung, Trung Dao, Dimitris Metaxas, Anh Tran

专题命中效率与蒸馏：image generation(title)；text-to-image(title)；diffusion(abstract)；分类 cs.CV

AI总结本文提出自修正流蒸馏方法，结合一致性模型和对抗训练，提升文本到图像生成的一致性与效率，实验验证其在CelebA-HQ和COCO数据集上的优越性能。

Comments Accepted to AAAI 2025. Code: https://github.com/hao-pt/SCFlow.git

2602.09014 2026-02-10 cs.CV cs.AI 86%

ArcFlow: Unleashing 2-Step Text-to-Image Generation via High-Precision Non-Linear Flow Distillation

ArcFlow：通过高精度非线性流蒸馏实现两步文本到图像生成

Zihan Yang, Shuyuan Tu, Licheng Zhang, Qi Dai, Yu-Gang Jiang, Zuxuan Wu

机构 * Fudan University（复旦大学）； Microsoft Research Asia（微软亚洲研究院）

专题命中效率与蒸馏：image generation(title)；text-to-image(title)；diffusion(abstract)；分类 cs.CV

AI总结 ArcFlow通过高精度非线性流蒸馏实现两步文本到图像生成，提升生成速度并保持质量。

2405.17873 2024-05-31 cs.CV cs.AI 86%

MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization

Tianchen Zhao, Xuefei Ning, Tongcheng Fang, Enshu Liu, Guyue Huang, Zinan Lin, Shengen Yan, Guohao Dai, Yu Wang

专题命中效率与蒸馏：diffusion(title,abstract)；text-to-image(title)；分类 cs.CV

Comments Project Page: https://a-suozhang.xyz/mixdq.github.io/

2304.01515 2023-04-05 cs.LG cs.CL cs.CV 86%

Text-Conditioned Sampling Framework for Text-to-Image Generation with Masked Generative Models

Jaewoong Lee, Sangwon Jang, Jaehyeong Jo, Jaehong Yoon, Yunji Kim, Jin-Hwa Kim, Jung-Woo Ha, Sung Ju Hwang

专题命中效率与蒸馏：image generation(title)；text-to-image(title)；diffusion(abstract)；分类 cs.CV

2608.09226 2026-08-11 cs.CV cs.AI 新提交 85%

RL-Native Distillation: Exploiting Scored Trajectories for Few-Step Image Generation

原生强化学习蒸馏：利用带分数轨迹实现少步图像生成

Yuhan Li, Fangao Zeng, Sicong Kang, Mengfei Xu, Hao Zhou, Wei Li, Pipei Huang, Bingbing Ni

机构 * Shanghai Jiao Tong University（上海交通大学）； Taobao & Tmall group of Alibaba（阿里巴巴淘宝天猫集团）

专题命中效率与蒸馏：image generation(title,abstract)；text-to-image(abstract)；diffusion(abstract)；分类 cs.CV

AI总结该研究提出REST框架，将RL与蒸馏联合训练，通过AMD优化，实现少步无CFG图像生成，性能优于40步RL教师，成本低、迭代少。

2509.22925 2026-03-20 cs.CV cs.AI cs.LG 85%

Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings

Soft-Di[M]O: 通过软嵌入提升一步离散图像生成

Yuanzhi Zhu, Xi Wang, Stéphane Lathuilière, Vicky Kalogeiton

机构 * LIX, École Polytechnique, CNRS, IPP（IX实验室，巴黎高等理工学院，法国国家科学研究中心，IPP）； Inria at Univ. Grenoble Alpes, CNRS, LJK（里尔大学，法国国家科学研究中心，LJK）

专题命中效率与蒸馏：image generation(title)；text-to-image(abstract)；diffusion(abstract)；image synthesis(abstract)

AI总结本文提出Soft-Di[M]O框架，通过引入软嵌入解决一步生成器中离散token阻断梯度的问题，实现端到端训练并支持GAN精炼、可微奖励微调和TTEO。

Comments ICLR 2026

2510.16983 2025-10-21 cs.CV cs.AI cs.LG 85%

One-step Diffusion Models with Bregman Density Ratio Matching

Yuanzhi Zhu, Eleftherios Tsonis, Lucas Degeorge, Vicky Kalogeiton

机构 * LIX, École Polytechnique, CNRS, IPP（巴黎政治学院LIX研究所，法国国家科学研究中心，IPP）； LIGM, École Nationale des Ponts et Chaussées, CNRS, IPP（巴黎综合理工学院LIGM研究所，法国国家科学研究中心，IPP）

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

Comments work in progress

2503.09641 2025-09-30 cs.GR 85%

SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation

Junsong Chen, Shuchen Xue, Yuyang Zhao, Jincheng Yu, Sayak Paul, Junyu Chen, Han Cai, Song Han, Enze Xie

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.GR

Comments 22 pages, 11 figures, 8 tables, In submission

2507.11554 2025-08-05 cs.CV cs.AI 85%

Inversion-DPO: Precise and Efficient Post-Training for Diffusion Models

Zejian Li, Yize Li, Chenye Meng, Zhongni Liu, Yang Ling, Shengyuan Zhang, Guang Yang, Changyuan Yang, Zhiyuan Yang, Lingyun Sun

机构 * Zhejiang University（浙江大学）； University of Electronic Science and Technology of China（电子科技大学）； Peking University（北京大学）； Alibaba Group（阿里巴巴集团）

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

Comments Accepted by ACM MM25

2507.04947 2025-07-08 cs.CV cs.AI 85%

DC-AR: Efficient Masked Autoregressive Image Generation with Deep Compression Hybrid Tokenizer

Yecheng Wu, Junyu Chen, Zhuoyang Zhang, Enze Xie, Jincheng Yu, Junsong Chen, Jinyi Hu, Yao Lu, Song Han, Han Cai

机构 * NVIDIA

专题命中效率与蒸馏：image generation(title,abstract)；text-to-image(abstract)；diffusion(abstract)；分类 cs.CV

Comments ICCV 2025

2506.18226 2025-06-24 cs.CV cs.AI 85%

Make It Efficient: Dynamic Sparse Attention for Autoregressive Image Generation

Xunzhi Xiang, Qi Fan

机构 * Nanjing University, School of Intelliger Science and Technology（南京大学智能科学与技术学院）

专题命中效率与蒸馏：image generation(title,abstract)；text-to-image(abstract)；image synthesis(abstract)；分类 cs.CV

2503.13070 2025-06-10 cs.CV 85%

Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation

Yihong Luo, Tianyang Hu, Weijian Luo, Kenji Kawaguchi, Jing Tang

机构 * HKUST（香港科技大学）； NUS（国立大学新加坡）； Xiaohongshu Inc（小红书公司）； HKUST (GZ)（香港科技大学（广州））

专题命中效率与蒸馏：image generation(title,abstract)；text-to-image(abstract)；diffusion(abstract)；分类 cs.CV

2506.04421 2025-06-06 cs.CV cs.AI cs.LG 85%

HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation

Hermann Kumbong, Xian Liu, Tsung-Yi Lin, Ming-Yu Liu, Xihui Liu, Ziwei Liu, Daniel Y. Fu, Christopher Ré, David W. Romero

机构 * Stanford University（斯坦福大学）； NVIDIA ； CUHK（中国香港大学）； HKU（香港大学）； NTU（国立新加坡大学）； UCSD（加州大学圣地亚哥分校）； Together AI

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

Comments Accepted to CVPR 2025. Project Page: https://research.nvidia.com/labs/dir/hmar/

2501.18427 2025-05-20 cs.CV 85%

SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer

Enze Xie, Junsong Chen, Yuyang Zhao, Jincheng Yu, Ligeng Zhu, Chengyue Wu, Yujun Lin, Zhekai Zhang, Muyang Li, Junyu Chen, Han Cai, Bingchen Liu, Daquan Zhou, Song Han

机构 * NVIDIA ； MIT（麻省理工学院）； Tsinghua University（清华大学）； Playground ； Peking University（北京大学）； The University of Hong Kong（香港大学）

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2412.06661 2025-05-08 cs.CV 85%

Efficiency Meets Fidelity: A Novel Quantization Framework for Stable Diffusion

Shuaiting Li, Juncan Deng, Zeyu Wang, Kedong Xu, Rongtao Deng, Hong Gu, Haibin Shen, Kejie Huang

机构 * Zhejiang University（浙江大学）； vivo Mobile Communication Co., Ltd（vivo移动通信有限公司）

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2503.20744 2025-03-27 cs.CV cs.AI 85%

High Quality Diffusion Distillation on a Single GPU with Relative and Absolute Position Matching

Guoqiang Zhang, Kenta Niwa, J. P. Lewis, Cedric Mesnage, W. Bastiaan Kleijn

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2503.15457 2025-03-20 cs.CV cs.AI cs.LG 85%

Di$\mathtt{[M]}$O: Distilling Masked Diffusion Models into One-step Generator

Yuanzhi Zhu, Xi Wang, Stéphane Lathuilière, Vicky Kalogeiton

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2503.14868 2025-03-20 cs.CV cs.AI 85%

Efficient Personalization of Quantized Diffusion Model without Backpropagation

Hoigi Seo, Wongi Jeong, Kyungryeol Lee, Se Young Chun

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

2503.06674 2025-03-13 cs.CV 85%

Learning Few-Step Diffusion Models by Trajectory Distribution Matching

Yihong Luo, Tianyang Hu, Jiacheng Sun, Yujun Cai, Jing Tang

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

Comments Project page: https://tdm-t2x.github.io/

2502.15681 2025-03-11 cs.LG cs.AI cs.CV 85%

One-step Diffusion Models with $f$-Divergence Distribution Matching

Yilun Xu, Weili Nie, Arash Vahdat

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2403.01505 2025-03-06 cs.CV 85%

SCott: Accelerating Diffusion Models with Stochastic Consistency Distillation

Hongjian Liu, Qingsong Xie, TianXiang Ye, Zhijie Deng, Chen Chen, Shixiang Tang, Xueyang Fu, Haonan Lu, Zheng-jun Zha

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

Comments 22 pages, 16 figures

2411.15393 2024-11-26 cs.CV cs.AI 85%

Gradient-Free Classifier Guidance for Diffusion Model Sampling

Rahul Shenoy, Zhihong Pan, Kaushik Balakrishnan, Qisen Cheng, Yongmoon Jeon, Heejune Yang, Jaewon Kim

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2311.14097 2024-03-29 cs.CV 85%

ACT-Diffusion: Efficient Adversarial Consistency Training for One-step Diffusion Models

Fei Kong, Jinhao Duan, Lichao Sun, Hao Cheng, Renjing Xu, Hengtao Shen, Xiaofeng Zhu, Xiaoshuang Shi, Kaidi Xu

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

Comments To appear in CVPR 2024

2402.13929 2024-03-05 cs.CV cs.AI cs.LG 85%

SDXL-Lightning: Progressive Adversarial Diffusion Distillation

Shanchuan Lin, Anran Wang, Xiao Yang

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)；分类 cs.CV

2603.22943 2026-03-25 cs.AI 85%

PersonalQ: Select, Quantize, and Serve Personalized Diffusion Models for Efficient Inference

PersonalQ: 选择、量化并服务个性化扩散模型以实现高效推理

Qirui Wang, Qi Guo, Yiding Sun, Junkai Yang, Dongxu Zhang, Shanmin Pang, Qing Guo

机构 * School of Software Engineering, Xi'an Jiaotong University（西安交通大学软件工程学院）； Nankai University（南开大学）

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；text-to-image(abstract)

AI总结 PersonalQ通过触发令牌连接检查点选择与量化，提升意图对齐并优化压缩质量，实现高效个性化扩散模型服务。

Comments Accepted in ICME 2026

2111.09267 2022-05-10 cs.CV cs.AI 84%

DiverGAN: An Efficient and Effective Single-Stage Framework for Diverse Text-to-Image Generation

Zhenxing Zhang, Lambert Schomaker

专题命中效率与蒸馏：image generation(title)；text-to-image(title)；分类 cs.CV

Journal ref Neurocomputing 2022

2508.10931 2026-02-20 cs.CV cs.GR 84%

VSF: Simple, Efficient, and Effective Negative Guidance in Few-Step Image Generation Models By Value Sign Flip

VSF: 在少步图像生成模型中通过值符号翻转实现简单、高效且有效的负引导

Wenqi Guo, Shan Du

机构 * Department of CMPS University of British Columbia（计算机科学与工程系，不列颠哥伦比亚大学）； Department of MEOW, Weathon Software（MEOW系，Weathon软件公司）

专题命中效率与蒸馏：image generation(title,abstract)；diffusion(abstract)；分类 cs.CV、cs.GR

AI总结 VSF通过值符号翻转在少步图像生成模型中实现高效负提示引导，提升生成质量与稳定性。

2410.20898 2025-06-06 cs.CV cs.AI cs.LG cs.MM 84%

David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training

Weijian Luo, Colin Zhang, Debing Zhang, Zhengyang Geng

专题命中效率与蒸馏：diffusion(title,abstract)；text-to-image(abstract)；分类 cs.CV、cs.MM

Comments Revision: paper accepted by the ICML2025 main conference

2504.07998 2025-04-14 cs.GR cs.AI cs.AR cs.CV 84%

CDM-QTA: Quantized Training Acceleration for Efficient LoRA Fine-Tuning of Diffusion Model

Jinming Lu, Minghao She, Wendong Mao, Zhongfeng Wang

专题命中效率与蒸馏：diffusion(title,abstract)；image generation(abstract)；分类 cs.CV、cs.GR

Comments ISCAS 2025

2405.20675 2024-06-03 cs.CV cs.AI cs.LG cs.MM 84%

Adv-KD: Adversarial Knowledge Distillation for Faster Diffusion Sampling

Kidist Amde Mekonnen, Nicola Dall'Asen, Paolo Rota

专题命中效率与蒸馏：diffusion(title,abstract)；image synthesis(abstract)；分类 cs.CV、cs.MM

Comments 7 pages, 11 figures, ELLIS Doctoral Symposium 2023 in Helsinki, Finland

1. 效率与蒸馏 2536 篇

Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Text-to-Image Generation

ArcFlow: Unleashing 2-Step Text-to-Image Generation via High-Precision Non-Linear Flow Distillation

MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization

Text-Conditioned Sampling Framework for Text-to-Image Generation with Masked Generative Models

RL-Native Distillation: Exploiting Scored Trajectories for Few-Step Image Generation

Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings

One-step Diffusion Models with Bregman Density Ratio Matching

SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation

Inversion-DPO: Precise and Efficient Post-Training for Diffusion Models

DC-AR: Efficient Masked Autoregressive Image Generation with Deep Compression Hybrid Tokenizer

Make It Efficient: Dynamic Sparse Attention for Autoregressive Image Generation

Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation

HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation

SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer

Efficiency Meets Fidelity: A Novel Quantization Framework for Stable Diffusion

High Quality Diffusion Distillation on a Single GPU with Relative and Absolute Position Matching

Di$\mathtt{[M]}$O: Distilling Masked Diffusion Models into One-step Generator

Efficient Personalization of Quantized Diffusion Model without Backpropagation

Learning Few-Step Diffusion Models by Trajectory Distribution Matching

One-step Diffusion Models with $f$-Divergence Distribution Matching

SCott: Accelerating Diffusion Models with Stochastic Consistency Distillation

Gradient-Free Classifier Guidance for Diffusion Model Sampling

ACT-Diffusion: Efficient Adversarial Consistency Training for One-step Diffusion Models

SDXL-Lightning: Progressive Adversarial Diffusion Distillation

PersonalQ: Select, Quantize, and Serve Personalized Diffusion Models for Efficient Inference

DiverGAN: An Efficient and Effective Single-Stage Framework for Diverse Text-to-Image Generation

VSF: Simple, Efficient, and Effective Negative Guidance in Few-Step Image Generation Models By Value Sign Flip

David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training

CDM-QTA: Quantized Training Acceleration for Efficient LoRA Fine-Tuning of Diffusion Model

Adv-KD: Adversarial Knowledge Distillation for Faster Diffusion Sampling