Title: (ECCV 2024) Nickel and Diming Your GAN: A Dual-Method Approach to Enhancing GAN Efficiency via Knowledge Distillation
Website: https://sangyeopyeo.github.io/Nickel_and_Diming_Your_GAN/
Authors: Sangyeop Yeo, Yoojin Jang, Jaejun Yoo

Thank you for your interest! The code has been pre-released, but we are currently refining it, and the trained weights will be available soon.

I'm sangyeop. I obtained B.S.(2021) degree from Electrical and Computer Engineering at Ajou University.

Currently, I'm a M.S. & Ph.D. course student at Laboratory of Advanced Imaging Technology (LAIT) in the Artificial Intelligence Graduate School (AIGS) at Ulsan National Institute of Science and Technology (UNIST), under the supervision of Prof.Jaejun Yoo.

My main research area lies at the intersection of multi-modal generative models and efficient, personalized generative models.

I am very interested in understanding and expressing the world we live in. Generative models are a powerful way to model our world, and I have a strong interest in them to better express the real world. To achieve this, we need a deep understanding of the various modalities that make up the world, such as images, videos, 3D representations, language, and audio. I am particularly focused on multi-modal generative models that can effectively integrate these elements, and I also have a strong interest in foundation models to further enrich this expression.

Even if we model the world excellently, if only a few people can benefit from it, it cannot reach its true value. Therefore, I am also deeply interested in personalized generative models. In particular, my research focuses on computationally efficient and data-efficient generative models to address the cost challenges that pose significant barriers to personalization. Lightweight generative models and few-shot learning and generation enable generative models to perform well in personal environments where computational resources and data are limited. Additionally, I am interested in the controllability and manipulability of generative models to enable users to use them effectively for personalized generative models.

If you have any questions about me, feel free to ask anytime.

Curriculum Vitae

• Multi-modal generative models
  • images, videos, 3D representations, language, and audio
• Efficient deep generative models
  • lightweight models, and few-shot learning and generation
• Controllable generative models
  • manipulation, and editing
• Foundation models, and Personalization

• LAIT - https://github.com/LAIT-CVLab

Nickel and Diming Your GAN: A Dual-Method Approach to Enhancing GAN Efficiency via Knowledge Distillation
Sangyeop Yeo, Yoojin Jang, Jaejun Yoo
European Conference on Computer Vision (ECCV), October 2024
project page (First)

Can We Find Strong Lottery Tickets in Generative Models?
Sangyeop Yeo, Yoojin Jang, Jy-yong Sohn, Dongyoon Han, Jaejun Yoo
Association for the Advancement of Artificial Intelligence (AAAI), February 2023
project page (First, Oral)

Email: sangyeop377@gmail.com
LinkdIn: https://www.linkedin.com/in/SangyeopYeo/