Seongchan Kim
One observed past, many possible futures. Move the cursor to steer them.

Seongchan Kim김성찬

Integrated M.S.–Ph.D. student at CVLAB, Kim Jaechul Graduate School of AI, KAIST, advised by Prof. Seungryong Kim.

I study how intelligent systems perceive dynamic scenes and act within them. I started with multimodal video understanding and generation, and now work on vision-language-action models for autonomous driving and robot manipulation.

Next, I want to understand world models and action generation: how well perception, context modeling, and action generation inside a world model can capture the uncertainty of future dynamics.

Publications* equal contribution

Under review

ICLR 2027, under review

Exploring What to Preserve for Reasoning and Action in Structured Pruning of Driving VLAs

Seongchan Kim, Soowon Son, Sangoh Kim, Hyunah Ko, Eunbeen Hong, Hyunkoo Lee, Sanghyun Lee, WonJun Moon, Kihong Kim, Dongyeon Son, Woongchan Nam, Seungryong Kim

ICLR 2027, under review

PLAYGROUND: Progressive Layout Generation with Render-Grounded Planning

Jaeho Lee, Junhwan Heo, Junghyun Park, WonJun Moon, Eunju Yang, Seungho Jang, Seongchan Kim, Seungryong Kim

ICLR 2027, under review

DreamX: Video Diffusion Model is a Universal Latent Action Learner

Jisoo Kim, Chaehyun Kim, Junsu Kim, Jiho Park, Seongchan Kim, Eunbeen Hong, Soowon Son, Fu-En Yang, Seungryong Kim

CVPR 2027, under review

Factorized MLLM Embeddings for Composed Image Retrieval

WonJun Moon, Seongchan Kim, Dahun Kim, Seungryong Kim

Conference

ICLR 2026

MATRIX: Mask-Track Alignment for Interaction-Aware Video Generation

Siyoon Jin, Seongchan Kim, Dahyun Chung, Jaeho Lee, Hyunwook Choi, Jisu Nam, Jiyoung Kim, Seungryong Kim

CVPR 2026

InterRVOS: Interaction-Aware Referring Video Object Segmentation

Woojeong Jin, Seongchan Kim, Jaeho Lee, Seungryong Kim

AAAI 2025

MUG-VOS: Multi-Granularity Video Object Segmentation

Sangbeom Lim*, Seongchan Kim*, Seungjun An*, Seokju Cho, Paul Hongsuck Seo, Seungryong Kim

Workshop

ODI, PTA and VLM4RWD Workshops at NeurIPS 2026

One Loss Is Not Enough: Cross-Objective Structured Pruning for Reasoning Vision–Language–Action Models

Seongchan Kim, Soowon Son, Sangoh Kim, Hyunah Ko, Eunbeen Hong, Hyunkoo Lee, Sanghyun Lee, WonJun Moon, Kihong Kim, Dongyeon Son, Woongchan Nam, Seungryong Kim

Preliminary version of the ICLR 2027 submission on driving VLA pruning.

AdaptFM Workshop at ICML 2026

Referring Video Object Segmentation via Language-Aligned Track Selection

Seongchan Kim*, Woojeong Jin*, Sangbeom Lim, Heeji Yoon, Hyunwook Choi, Seungryong Kim

Wild3D Workshop at ICCV 2023

Self-Evolving Neural Radiance Fields

Jaewoo Jung*, Jisang Han*, Jiwon Kang*, Seongchan Kim, Min-Seop Kwak, Seungryong Kim

Research experience

May 2026 – now

Efficient autonomous driving models

LG Innotek, industry–academia project at CVLAB, KAIST

Proposed an analysis-driven structured pruning method for reasoning driving VLAs (Alpamayo) that cuts memory and inference latency while preserving performance (ICLR 2027, under review).

Jan – Mar 2026

Research intern

Maum AI, WORV Research Team

Researched vision-language-action models for robot manipulation that take long-context inputs, using input token compression for the underlying multimodal LLM.

Dec 2024 – Nov 2025

Visual inspection with segmentation and captioning models

Samsung Electro-Mechanics, industry–academia project at CVLAB, KAIST

Combined segmentation and captioning models for visual inspection, and developed an interaction-aware referring video object segmentation method grounded in natural language (InterRVOS, CVPR 2026).

Background

Education

  • KAIST, Integrated M.S.–Ph.D., Kim Jaechul Graduate School of AIMar 2025 – now. Advisor: Prof. Seungryong Kim
  • Korea University, B.S. in Computer Science and EngineeringMar 2020 – Aug 2024

Honors

  • Future Industry Graduate ScholarshipHyundai Motor Chung Mong-Koo Foundation, Mar 2025 – now
  • Future Industry Undergraduate ScholarshipHyundai Motor Chung Mong-Koo Foundation, Sep 2023 – Aug 2024

Service

  • Conference reviewerICLR 2027, AAAI 2026 and 2027

Contact