Joonmyung Choi

I am an Integrated M.S. & Ph.D. student in Artificial Intelligence at Korea University, advised by Prof. Hyunwoo J. Kim (MLV Lab).

My research centers on efficient deep learning, with a focus on token compression that prunes and merges redundant tokens to reduce the computational cost of transformers. I apply these techniques to video, multi-modal, and document understanding, making large models efficient enough for real-world deployment.

News

Feb 2026	Ours paper DocPrune is accepted to CVPR 2026.
Jun 2025	Ours paper Representation Shift is accepted to ICCV 2025.
Feb 2025	Ours paper EfficientViM is accepted to CVPR 2025.
Feb 2024	Two Papers, vid-TLDR, MCTF is accepted to CVPR 2024.
Aug 2023	Ours paper Concept Bottleneck is accepted to MICCAI-W 2023 (Oral).
Jul 2023	Ours paper RPO is accepted to ICCV 2023.
Feb 2023	Ours paper MELTR is accepted to CVPR 2023.
Sep 2022	Ours paper TokenMixup is accepted to NeurIPS 2022.
Mar 2022	Ours paper Video-Text Representation Learning is accepted to CVPR 2022.

↓ scroll for more

Selected Publications

(*) denotes equal contribution. Author names with bold indicate me.

CVPR 2026

DocPrune: Efficient Document Question Answering via Background, Question, and Comprehension-aware Token Pruning

Joonmyung Choi, S. Lee, J. Kim, S. Kim, D. Ko, J. Kil, and Hyunwoo J. Kim

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026

PDF

@inproceedings{choi2026docprune,
  title     = {DocPrune: Efficient Document Question Answering via Background, Question, and Comprehension-aware Token Pruning},
  author    = {Choi, Joonmyung and Lee, S. and Kim, J. and Kim, S. and Ko, D. and Kil, J. and Kim, Hyunwoo J.},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026},
}

ICCV 2025

Representation Shift: Unifying Token Compression with FlashAttention

Joonmyung Choi*, S. Lee*, B. Ko, E. Kim, J. Kil, and Hyunwoo J. Kim

IEEE/CVF International Conference on Computer Vision (ICCV), 2025

PDF Code

@inproceedings{choi2025representation,
  title     = {Representation Shift: Unifying Token Compression with FlashAttention},
  author    = {Choi, Joonmyung and Lee, S. and Ko, B. and Kim, E. and Kil, J. and Kim, Hyunwoo J.},
  booktitle = {IEEE/CVF International Conference on Computer Vision (ICCV)},
  year      = {2025},
}

CVPR 2024

vid-TLDR: Training Free Token Merging for Light-weight Video Transformer

Joonmyung Choi*, S. Lee*, J. Chu, M. Choi, and Hyunwoo J. Kim

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

PDF Code

@inproceedings{choi2024vidtldr,
  title     = {vid-TLDR: Training Free Token Merging for Light-weight Video Transformer},
  author    = {Choi, Joonmyung and Lee, S. and Chu, J. and Choi, M. and Kim, Hyunwoo J.},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2024},
}

CVPR 2024

Multi-criteria Token Fusion with One-step-ahead Attention for Efficient Vision Transformers

S. Lee*, Joonmyung Choi*, and Hyunwoo J. Kim

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

PDF Code

@inproceedings{lee2024mctf,
  title     = {Multi-criteria Token Fusion with One-step-ahead Attention for Efficient Vision Transformers},
  author    = {Lee, S. and Choi, Joonmyung and Kim, Hyunwoo J.},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2024},
}

CVPR 2023

MeLTR: Meta Loss Transformer for Learning to Fine-tune Video Foundation Models

D. Ko*, Joonmyung Choi*, H. Choi, K. On, B. Roh, and Hyunwoo J. Kim

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

PDF Code

@inproceedings{ko2023meltr,
  title     = {MeLTR: Meta Loss Transformer for Learning to Fine-tune Video Foundation Models},
  author    = {Ko, D. and Choi, Joonmyung and Choi, H. and On, K. and Roh, B. and Kim, Hyunwoo J.},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2023},
}

NeurIPS 2022

TokenMixup: Efficient Attention-guided Token-level Data Augmentation for Transformers

H. Choi*, Joonmyung Choi*, and Hyunwoo J. Kim

Advances in Neural Information Processing Systems (NeurIPS), 2022

PDF Code

@inproceedings{choi2022tokenmixup,
  title     = {TokenMixup: Efficient Attention-guided Token-level Data Augmentation for Transformers},
  author    = {Choi, H. and Choi, Joonmyung and Kim, Hyunwoo J.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2022},
}

Full list of publications →

Experience

Machine Learning & Vision Lab (MLV Lab), Korea University	Researcher	Mar 2021 – Current
LG CnS	Intern	Jul 2019 – Aug 2019
Data Marketing Korea	Intern	Jan 2019 – Feb 2019
COSCOI	Researcher	Aug 2016 – Oct 2018