Jeongsoo Park
jespark at umich dot edu

I am a first-year CSE PhD student at the University of Michigan, advised by Prof. Andrew Owens. I have a general interest in efficient computer vision, image processing, and machine learning.

I graduated with a master's degree in ECE from the University of Michigan, during which I was advised by Prof. Justin Johnson.

I received a BS degree in Electronic and Electrical Engineering from Sungkyunkwan University, where I worked with Prof. Jong Hwan Ko.

LinkedIn  /  Google Scholar  /  Twitter  /  Github

profile photo

I have been working on designing efficient computer vision system. My recent work, RGB no more, focuses on training Vision Transformers directly from the encoded features of a JPEG image to accelerate the entire pipeline. In my previous work, Auto-Tiler, I aim to design an efficient compression scheme for over-the-network model inference.

RGB no more: Minimally-decoded JPEG Vision Transformers
Jeongsoo Park, Justin Johnson
CVPR 2023
Project page / Code / Paper

We train ViTs directly from JPEG encoded features and accelerate train and eval by up to 39.2% and 17.9%.

Auto-Tiler: Variable-Dimension Autoencoder with Tiling for Compressing Intermediate Feature Space of Deep Neural Networks for Internet of Things
Jeongsoo Park, Jungrae Kim, Jong Hwan Ko
Sensors, 2021

We use autoencoders and show up to 67% higher accuracy and 81% reduced latency during network-constrained inference versus the image or video codec-based compression.

This website uses the html template by Jon Barron.