Byung-Ki Kwon

Early Failure Detection and Intervention in Video Diffusion Models

Kwon Byung-Ki, Sohwi Lim, Nam Hyeon-Woo, Moon Ye-Bin, Tae-Hyun Oh

Inspect intermediate video previews to detect failures early and intervene only when needed.

Zero-Shot Rankability: Revealing Latent Ordinal Structure in Multimodal Large Language Models via Language

Nam Hyeon-Woo, Moon Ye-Bin, Sohwi Lim, Kwon Byung-Ki, Tae-Hyun Oh

Use language to reveal ordinal structure in multimodal model embeddings without labeled examples.

HDR-NSFF: High Dynamic Range Neural Scene Flow Fields

Shin Dong-Yeon, Kim Jun-Seong, Kwon Byung-Ki, Tae-Hyun Oh

Reconstruct dynamic HDR scenes from monocular videos with alternating exposures.

Earlier version: Dynamic HDR Radiance Fields via Neural Scene Flow (ICCV 2025 Wild3D Workshop).

Revisiting Deep Video Motion Magnification for Real-time Applications

Ha Hyunwoo, Oh Hyun-Bin, Kim Jun-Seong, Kwon Byung-Ki, Kim Sung-Bin, Ji-Yun Kim, Sung-Ho Bae, Tae-Hyun Oh

Study efficient network architectures for real-time video motion magnification.

JointDiT: Enhancing RGB-Depth Joint Modeling with Diffusion Transformers

Kwon Byung-Ki, Qi Dai, Lee Hyoseok, Chong Luo, Tae-Hyun Oh

One joint model for RGB–depth generation, depth estimation, and depth-conditioned image synthesis.

Work performed at Microsoft Research Asia.

Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior

Lee Hyoseok, Kyeong Seon Kim, Kwon Byung-Ki, Tae-Hyun Oh

Align a diffusion depth prior with sparse measurements at test time for generalizable depth completion.

Learning-based Axial Video Motion Magnification

Kwon Byung-Ki, Oh Hyun-Bin, Kim Jun-Seong, Hyunwoo Ha, Tae-Hyun Oh

Make subtle motions easier to interpret by magnifying movement along a user-specified direction.

Uni-DVPS: Unified Model for Depth-Aware Video Panoptic Segmentation

Kim Ji-Yeon, Oh Hyun-Bin, Kwon Byung-Ki, Dahun Kim, Yongjin Kwon, Tae-Hyun Oh

Unify video panoptic segmentation, depth estimation, and tracking in one model.

The Devil is in the Details: Simple Remedies for Image-to-LiDAR Representation Learning

Jo Won-jun, Kwon Byung-Ki, Kim Ji-Yeon, Hawook Jeong, Kyungdon Joo, Tae-Hyun Oh

Revisit spatial and temporal design choices in image-to-LiDAR representation learning.

The Devil in the Details: Simple and Effective Optical Flow Synthetic Data Generation

Kwon Byung-Ki, Kim Sung-Bin, Tae-Hyun Oh

Identify the simple data-generation choices that help optical flow models generalize.

DFlow: Learning to Synthesize Better Optical Flow Datasets via a Differentiable Pipeline

Kwon Byung-Ki, Nam Hyeon-Woo, Kim Ji-Yun, Tae-Hyun Oh

Learn to synthesize optical flow training data with a differentiable generation pipeline.

Samsung HumanTech Paper Award · Bronze Prize