Dogyun Park

M.S & Ph.D Integrated Student in MLV Lab, advised by Prof. Hyunwoo J. Kim. Department of Computer Science and Engineering at Korea University, Seoul, Republic of Korea.

My research interests are Generative AI in computer vision, especially in making efficient and effective vision generative models. My goal is to push the boundaries of generative AI to enable more creative and efficient solutions for real-world applications, from content creation to scientific simulations.

I am actively seeking opportunities to contribute to impactful projects in the field of Generative AI. If you are interested in collaboration or have opportunities that align with my expertise, please feel free to reach out to me via contact email.

news

Sep 24, 2025	We’re excited to announce that our new preprint Sprint: Sparse-Dense Residual Fusion for Efficient Diffusion Transformers is now available on arXiv. Our proposed SPRINT 🚀 framework achieves up to 9.8× faster DiT training while maintaining high generation quality.
Sep 18, 2025	Our Blockwise Flow Matching has been accepted to NeurIPS 2025! 🎉 BFM reduces the inference cost of standard DiTs by up to 4.9×, while maintaining comparable generation quality.
May 31, 2025	Excited to share that I’ve started my Research Internship at 👻 Snap Inc., joining the Creative Vision Team! I’ll be working on efficiency of generative AI and video diffusion models.

selected publications [full list]

(*) denotes equal contribution

arxiv

Sprint: Sparse-Dense Residual Fusion for Efficient Diffusion Transformers

Dogyun Park, Haji-Ali Moayed, Yanyu Li, Willi Menapace, Sergey Tulyakov, Hyunwoo J Kim, Aliaksandr Siarohin, and Anil Kag

arXiv preprint arXiv:2510.21986

Bib PDF Project

@article{park2025sprint,
  project = {https://dogyunpark.github.io/bfm/},
  title = {Sprint: Sparse-Dense Residual Fusion for Efficient Diffusion Transformers},
  author = {Park, Dogyun and Moayed, Haji-Ali and Li, Yanyu and Menapace, Willi and Tulyakov, Sergey and Kim, Hyunwoo J and Siarohin, Aliaksandr and Kag, Anil},
  journal = {arXiv preprint arXiv:2510.21986},
  year = {2025},
}

NeurIPS

Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation

Dogyun Park, Taehoon Lee, Minseok Joo, and Hyunwoo J Kim

Neural Information Processing Systems, NeurIPS

Bib PDF Code

@article{park2025blockwise,
  title = {Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation},
  author = {Park, Dogyun and Lee, Taehoon and Joo, Minseok and Kim, Hyunwoo J},
  journal = {Neural Information Processing Systems, NeurIPS},
  year = {2025},
}

NeurIPS

Constant Acceleration Flow

Dogyun Park, Sojin Lee, Sihyeon Kim, Taehoon Lee, Youngjoon Hong, and Hyunwoo J Kim

Neural Information Processing Systems, NeurIPS

Bib PDF Code

@article{park2024constant,
  title = {Constant Acceleration Flow},
  author = {Park, Dogyun and Lee, Sojin and Kim, Sihyeon and Lee, Taehoon and Hong, Youngjoon and Kim, Hyunwoo J},
  journal = {Neural Information Processing Systems, NeurIPS},
  year = {2024},
}

ECCVOral

Diffusion prior-based amortized variational inference for noisy inverse problems

Sojin Lee*, Dogyun Park*, Inho Kong, and Hyunwoo J Kim

European Conference on Computer Vision, ECCV

Oral Presentation [Top 2.3%]

Bib PDF Code

@article{lee2024diffusion,
  title = {Diffusion prior-based amortized variational inference for noisy inverse problems},
  author = {Lee, Sojin and Park, Dogyun and Kong, Inho and Kim, Hyunwoo J},
  award = {Oral},
  honor = {Oral Presentation [Top 2.3%]},
  journal = {European Conference on Computer Vision, ECCV},
  year = {2024},
}

ICLR

DDMI: Domain-Agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations

Dogyun Park, Sihyeon Kim, Sojin Lee, and Hyunwoo J Kim

International Conference on Learning Representations, ICLR

Bib PDF Code Project

@article{park2024ddmi,
  project = {https://dogyunpark.github.io/ddmi/},
  title = {DDMI: Domain-Agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations},
  author = {Park, Dogyun and Kim, Sihyeon and Lee, Sojin and Kim, Hyunwoo J},
  journal = {International Conference on Learning Representations, ICLR},
  year = {2024},
}

ICCV

Probabilistic Precision and Recall Towards Reliable Evaluation of Generative Models

Dogyun Park, and Suhyun Kim

International Conference on Computer Vision, ICCV

Bib PDF

@article{park2023probabilistic,
  title = {Probabilistic Precision and Recall Towards Reliable Evaluation of Generative Models},
  author = {Park, Dogyun and Kim, Suhyun},
  journal = {International Conference on Computer Vision, ICCV},
  pages = {20099--20109},
  year = {2023},
}