Aaron Inbum Park

I am a master's student at the University of Michigan, advised by Prof. Andrew Owens.

I obtained my bachelor's degree in Electrical and Computer Engineering from Seoul National University, where I was fortunate to work under Prof. Jungdam Won, Prof. Se Young Chun, and Prof. Young Min Kim.

My research interests lie in the area of computer vision, especially leveraging generative models for image, shape, or motion reconstruction.

Email  /  CV  /  Google Scholar  /  LinkedIn

profile photo
Factorized Diffusion: Perceptual Illusions by Noise Decomposition
Daniel Geng*, Inbum Park*, Andrew Owens
arXiv, 2024  
paper / arXiv / project

We generate various perceptual illusions by controlling components of an image using off-the-shelf diffusion models. Our method can produce hybrid images and relates to prior work in compositional generation and inverse problems.

Visual Anagrams: Generating Multi-View Optical Illusions with Diffusion Models
Daniel Geng, Inbum Park, Andrew Owens
Conference on Computer Vision and Pattern Recognition (CVPR), 2024 (Oral)  
paper / arXiv / project

A simple, zero-shot method for generating multi-view optical illusions, which are images that change their appearance or identity upon a transformation.

On the Robustness of Normalizing Flows for Inverse Problems in Imaging
Seongmin Hong, Inbum Park, Se Young Chun
International Conference on Computer Vision (ICCV), 2023  
paper / arXiv / project

A study on erroneous images occasionally generated from flow-based models for inverse problems in imaging.

Text2PointCloud: Text-Driven Stylization for Sparse PointCloud
Inwoo Hwang, Hyeonwoo Kim, Donggeun Lim, Inbum Park, Young Min Kim
Eurographics (Short Papers), 2023  
paper / video

A framework that stylizes and upsamples an uncolored, sparse pointcloud given a text description to render a high-quality 3D output.

Probabilistic Implicit Scene Completion
Dongsu Zhang, Changwoon Choi, Inbum Park, Young Min Kim
International Conference on Learning Representations (ICLR), 2022 (Spotlight)  
paper / arXiv / code

A probabilistic approach to shape completion and scene reconstruction using 3D implicit representations.


I am immersed in the choreography scene and also love learning languages, including French and Italian.

Source code credit to Dr. Jon Barron