Ph.D student in CSE at Seoul National University
location_on Seoul, KoreaHello! I am a Ph.D student in CSE at Seoul National University, and studying under the supervision of Prof. Gunhee Kim.
My research interests are in the field of 3D Place Recognition and Visual Localization, especially about (i) understanding the scene from images and point clouds, (ii) dealing with 2D-3D cross-modalities, and (iii) utilizing high-level semantic information for place recognition.
EP2P-Loc: End-to-End 3D Point to 2D Pixel Localization for Large-Scale Visual Localization
Minjung Kim, Junseo Koo, Gunhee Kim
ICCV 2023
[pdf]
Indoor/Outdoor Transition Recognition Based on Door Detection
Seohyun Jeon, Minjung Kim, Seunghwan Park, Jaeyoung Lee
UR 2022
[pdf]
Drop-Bottleneck: Learning Discrete Compressed Representation for Noise-Robust Exploration
Jaekyeom Kim, Minjung Kim, Dongyeon Woo, Gunhee Kim
ICLR 2021
[OpenReview]
[arxiv]
[pdf]
[talk]
[code]
Logo Detection and Recognition Algorithm using YOLO-v3 Model
Minjung Kim, Sungen Kim, Gunhee Kim
CICS 2020
Memorization Precedes Generation: Learning Unsupervised GANs with Memory Networks
Youngjin Kim, Minjung Kim, Gunhee Kim
ICLR 2018
[OpenReview]
[arxiv]
[pdf]
[code]
Machine Learning for Determining Duplicate Question
Minjung Kim, Yeongjoon Park, Hyungsuk Lim, Jihoon Yang
KCS 2017
Sketch based Face Image Generation with Text Mapping
Minjung Kim, Hyungsuk Lim, Yeongjoon Park, Yeseul Joo, MyoungWan Koo
KCS 2017
DeepGuider 2019.02 - Current [code]
The DeepGuider Project is a national government-funded research project focused on developing a navigation guidance system for robots to navigate urban environments without pre-mapping.
I contribute to finding clues to locate autonomous robots by detecting and recognizing points of interests (POIs) in images of a scene, including text, landmarks, and doors for indoor-outdoor transition, while also developing robust training methods for environmental changes.
PRIDE: 3D Place Recognition In Dynamic Environment 2022.03 - Current [code]
This work proposes a new dataset called PRIDE, which includes dynamic objects such as cars and pedestrians, for 3D place recognition in dynamic environments that are more realistic and challenging than current benchmark datasets.
The proposed PRIDE-Net architecture with a new loss function focuses on extracting discriminative global descriptors and capturing global context using spatial information, while being robust to dynamic environments.
Experiments on the PRIDE dataset and existing benchmarks show that our proposed method outperforms previous methods and that each proposed module effectively improves performance.
The code will be released after acceptance.
FCAT: Fully Convolutional Network with Self-Attention for Point Cloud based Place Recognition 2020.12 - 2022.02 [code]
We construct a novel network named FCAT (Fully Convolutional network with a self-ATtention unit) that can generate a discriminative and context-aware global descriptor for place recognition from the 3D point cloud.
It features with a novel sparse fully convolutional network architecture with sparse tensors for extracting informative local geometric features computed in a single pass.
It also involves a self-attention module for 3D point cloud to encode local context information between local descriptors.
The code will be released after acceptance.
Bayesian Deep Learning course 2018.02 - 2018.07 [lecture]
To understand deep learning papers, we explain the basic concepts of probability and Bayesian, and introduce papers related to Bayesian neural networks.
This lecture can be taken through edwith of Naver Connect.
Sketch based Face Image Generation with Text Mapping 2017.09 - 2018.02 [code]
A typical sketch might have been uncomfortable when a person or program was used to map a person’s features in detail.
This process is limited not only because it is very complex and requires technicians, but also because it creates a feeling of incompatibility with real people.
This program, named Metamon, makes a picture of a person’s face by entering the image of the border sketch of the person’s face and the text information that shows the characteristics of the face.
Arduino & Raspberry Pi & Internet of Things (IoT) Tutorial 2016.12 - 2017.03 [project]
I create tutorial pages with Youtube videos and code for beginners in Arduino kit and Raspberry Pi development.
I introduce the concept of the Internet of Things (IoT) and work on a mini-project using ThingSpeak.
Sogang Navigation and Introduction (SNI) 2015.03 - 2015.07 [code]
We develop a navigation system that introduces the internal facilities of each building and displays the shortest route and time from building to building using the Floyd-Washall algorithm.
To build data for the development, we measured the time taken by walking directly on each path.
Reviewer of International Conferences
IEEE/CVF International Conference on Computer Vision (ICCV) 2023
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2023
Asian Conference on Computer Vision (ACCV) 2022
International Conference on Learning Representations (ICLR) 2022, 2023
Neural Information Processing Systems (NeurIPS) 2021, 2022, 2023
Technical Coaching
2022-3 SK hynix ML Engineer Technical Coaching