Hello, I'm Yeji Song

I am a Ph.D student at Seoul National University, under the supervision of Prof. Nojun Kwak.

My primary focus is on video & image generation, aiming to push the boundaries of their applications in real-world scenarios. Specifically, developing generative models that provide more diverse experiences to users is my central goal. My research interests also include a broader computer vision area, with experience spanning diffusion, video rendering, segmentation, and 3D object detection.

CV Scholar Github

Publications

Point-to-Point: Sparse Motion Guidance for Controllable Video Editing

Point-to-Point: Sparse Motion Guidance for Controllable Video Editing

Yeji Song, Jaehyun Lee, Mijin Koo, Nojun Kwak
ArXiv, 2025

Project Page PDF arXiv
Targeted Data Protection for Diffusion Model by Matching Training Trajectory

Targeted Data Protection for Diffusion Model by Matching Training Trajectory

Hojun Lee, Mijin Koo, Yeji Song, Nojun Kwak
AAAI, 2026

We prevented unauthorized T2I customization by guiding the generated image to show the intended target.

Project Page PDF arXiv
ReFlex: Text-Guided Editing of Real Images in Rectified Flow via Mid-Step Feature Extraction and Attention Adaptation

ReFlex: Text-Guided Editing of Real Images in Rectified Flow via Mid-Step Feature Extraction and Attention Adaptation

Jimyeong Kim, Jungwon Park, Yeji Song, Nojun Kwak, Wonjong Rhee
ICCV, 2025 Highlight

We developed a new real-image editing approach for FLUX, leveraging analysis of intermediate representations.ReFlex achieved text alignment improvements of 1.69-7.11% on PIE-Bench and 3.21-16.46% on Wild-TI2I-Real.

Project Page PDF arXiv
Harmonizing Visual and Textual Embeddings for Zero-Shot Text-to-Image Customization

Harmonizing Visual and Textual Embeddings for Zero-Shot Text-to-Image Customization

Yeji Song, Jimyeong Kim * , Wonhark Park * , Wonsik Shin, Wonjong Rhee, Nojun Kwak
AAAI, 2025

We discovered a conflict among contextual embeddings in zero-shot T2I customization.By resolving the conflict, we improved text alignment by 3.82-5.26% and image alignment by 2.11-12.2%.

Project Page PDF arXiv
SAVE: Protagonist Diversification with Structure Agnostic Video Editing
Hover Here!

SAVE: Protagonist Diversification with Structure Agnostic Video Editing

Yeji Song, Wonsik Shin, Junsoo Lee, Jeesoo Kim, Nojun Kwak
ECCV, 2024

We expanded video editing capabilities to enable subject replacement across diverse body structures.We introduced a new motion token, resulting in text alignment improvements of up to 7.66%.

Project Page PDF arXiv
Towards Efficient Neural Scene Graphs by Learning Consistency Fields
Hover Here!

Towards Efficient Neural Scene Graphs by Learning Consistency Fields

Yeji Song, Chaerin Kong, Seoyoung Lee, Nojun Kwak * , Joonseok Lee *
BMVC, 2022

In video scene rendering, we reformulate neural radiance fields to additionally consider consistency fields, enabling more efficient and controllable scene manipulation.

Project Page PDF arXiv
Exploring Causal Mechanisms for Machine Text Detection Methods

Exploring Causal Mechanisms for Machine Text Detection Methods

Kiyoon Yoo, Wonhyuk Ahn, Yeji Song, Nojun Kwak
NAACL Workshop, 2024

We showed the existence of backdoor path that confounds the relationships between text and machine text detection score.

Project Page PDF arXiv
Finding Efficient Pruned Network via Refined Gradients for Pruned Weights

Finding Efficient Pruned Network via Refined Gradients for Pruned Weights

Jangho Kim, Jayeon Yoo * , Yeji Song * , Kiyoon Yoo, Nojun Kwak
ACM MM, 2023

We employed refined gradients to update the pruning weights, enhancing both training stability and the model performance.

Project Page PDF arXiv
Md3d: Mixture-density-based 3d object detection in point clouds

Md3d: Mixture-density-based 3d object detection in point clouds

Jaeseok Choi, Yeji Song, Yerim Kim, Jaeyoung Yoo, Nojun Kwak
IEEE Access, 2022

We utilize the Gaussian Mixture Model (GMM) in the 3D object detection task to predict the distribution of 3D bounding boxes, eliminating the need for laborious, hand-crafted anchor design.

Project Page PDF arXiv
Part-aware data augmentation for 3d object detection in point cloud

Part-aware data augmentation for 3d object detection in point cloud

Jaeseok Choi, Yeji Song, Nojun Kwak
IROS, 2021

We delve into data augmentation in 3D object detection, leveraging sophisticated and rich structural information present in 3D labels.

Project Page PDF arXiv