Manjin Kim

I’m Manjin Kim, a Senior Researcher at Qualcomm AI Research. I completed my Ph.D. in 2026 at Pohang University of Science and Technology (POSTECH), South Korea, under the supervision of Professor Minsu Cho in the computer vision lab. My research interest is in learning video representation and its applications.

Research Interests

Tempered Self-Similarity Alignment for Physically Plausible Video Generation
Manjin Kim, Suha Kwak, and Minsu Cho, CVPR Workshop on Video Generative Models: Benchmarks and Evaluation 2026.
Exploring Multi-Order Self-Similarity for Motion Understanding
Manjin Kim*, Heeseung Kwon*, Karteek Alahari, and Minsu Cho (* equal contribution), NeurIPS 2026.
Generic Event Boundary Detection via Denoising Diffusion
Jaejun Hwang*, Dayoung Gong*, Manjin Kim, and Minsu Cho (* equal contribution), ICCV 2025.
Learning correlation structures for vision transformers [project page] [code]
Manjin Kim, Paul Hongsuck Seo*, Cordelia Schmid, and Minsu Cho* (* corresponding authors), CVPR 2024.
Future transformer for long-term action anticipation [project page] [code]
Dayoung Gong, Joonseok Lee, Manjin Kim, Seongjong Ha, and Minsu Cho, CVPR 2022.
Relational self-attention: what’s missing in attention for video understanding [project page] [code]
Manjin Kim*, Heeseung Kwon*, Chunyu Wang, Suha Kwak, and Minsu Cho (* equal contribution), NeurIPS 2021.
Learning self-similarity in space and time as as generalized motion for video action recognition [project page] [code]
Heeseung Kwon*, Manjin Kim*, Suha Kwak, and Minsu Cho (* equal contribution), ICCV 2021.
MotionSqueeze: neural motion feature learning for video understanding [project page] [code]
Heeseung Kwon, Manjin Kim, Suha Kwak, and Minsu Cho, ECCV 2020.

Senior Researcher, Qualcomm AI Research, Seoul, Korea (Mar. 2026 - present)
Student Researcher, Google Research, France (Jul. 2022 - Jan. 2023)
- Developed a multimodal long-form video captioning system.
- Host: Paul Hongsuck Seo
Research Intern, Microsoft Research Asia (MSRA), remote (Dec. 2020 - June. 2021)
- Developed a dynamic neural feature transform method, called Relational Self-Attention.
- Mentor: Chunyu Wang
Research Intern, LG CNS, Korea (Jun. 2018 - Aug. 2018)
- Developed a video data augmentation system using CycleGAN.

Motion-centric video representation learning (with Google Research)
- Developed vision transformers, dubbed StructViT, that effectively learn spatial(-temporal) correlation structures for both image and video understanding.
Online action detection in streamed videos (with POSCO ICT)
- Developed a real-time action recognition system for video surveillance.
Sentimental analysis on Hyundai vehicle models (with Hyundai Motor Company)
- Analyzed a sentiment of Hyundai vehicle models from comments on web communities.

Pohang University of Science and Technology, Pohang, Korea (2019.09 - 2026.02)
- Integrated M.S./Ph.D. in Convergence IT Engineering
- Advisor: Prof. Minsu Cho
Pohang University of Science and Technology, Pohang, Korea (2013.03 - 2019.02)
- B.S. in Convergence IT Engineering
- Advisor: Prof. Sangwoo Kim