Learning  3D Scene Analogies with Neural Contextual Scene Maps
Junho Kim, Gwangtak Bae, Eunsun Lee, Young Min Kim
arXiv  preprint, 2025.
[arXiv] [Code] [Paper] [Video] [Project Page]

We propose a new task of finding 3D scene analogies, which are dense maps connecting regions sharing similar scene contexts.

Fully Geometric Panoramic Localization
Junho Kim, Jiwon Jeong, Young Min Kim
CVPR 2024  (Nectar Track Spotlight Presentation at 3DV 2025)
[arXiv] [Code] [Paper] [Video] [Project Page]

By leveraging lines and their intersections, we can perform panoramic localization without using any visual features, in a fully geometric manner.

LDL: Line Distance Functions for Panoramic Localization
Junho Kim, Changwoon Choi, Hojun Jang, Young Min Kim
ICCV 2023
[arXiv] [Code] [Paper] [Video] [Project Page]

We introduce a simple pose search method that operates on 2D, 3D line segments based on distance functions, which attains competitive performance with a very short runtime.

Calibrating  Panoramic Depth Estimation for Practical Localization and Mapping
Junho Kim, Eunsun Lee, Young Min Kim
ICCV 2023
[arXiv] [Code] [Paper] [Video] [Project Page]

With a simple test-time adaptation scheme, we can 'calibrate' panoramic depth estimation algorithms to make more robust predictions and further improve on downstream tasks in localization and mapping.

Privacy-Preserving Visual  Localization with Event Cameras
Junho Kim, Young Min Kim,  Yicheng Wu, Ramzi Zahreddine, Weston Anthony Welge, Gurunandan Krishnan, Sizhuo Ma, Jian Wang 
arXiv  preprint, 2022.
[arXiv] [Code] [Paper] [Video] [Project Page]

We propose a event-based visual localization method that effectively leverages the strengths of event cameras while offering privacy protection for alleviating user concerns in client-server localization scenarios. 

Ev-NeRF:  Event Based Neural Radiance Field
Inwoo Hwang, Junho Kim, and Young Min Kim
WACV 2023
[arXiv] [Paper] [Video] [Project Page]

Neural radiance fields (NeRF) for event cameras can be learned without using any intensity image supervision, and enables robust high-dynamic range imaging amidst large amounts of sensor noise.

CPO: Change Robust Panorama to Point  Cloud Localization
Junho Kim, Hojun Jang, Changwoon Choi, and Young Min Kim
ECCV 2022
[arXiv] [Code] [Paper] [Video] [Project Page]

Constructing score maps in 2D, 3D that reflect regional color consistencies enable robust localization amidst scene changes.

MoDA:  Map style transfer for self-supervised Domain Adaptation of embodied agents
Eunsun Lee, Junho Kim, Sangwon Park, and Young Min Kim
ECCV 2022
[arXiv] [Paper] [Video] [Project Page]

By applying style transfer on the grid map domain, effective domain adaptation is possible for environments containing visual or dynamic noise.

Ev-TTA: Test-Time Adaptation for Event-Based Object Recognition
Junho Kim, Inwoo Hwang, and Young Min Kim
CVPR 2022
[arXiv] [Code] [Paper] [Video] [Project Page]

A simple test-time adaptation objective can help event-based classifiers to make robust predictions in extreme conditions such as large motion or low lighting. 

Self-Supervised Domain Adaptation for Visual Navigation with Global Map Consistency
Eunsun Lee, Junho Kim, and Young Min Kim
WACV 2022 
[arXiv] [Paper] [Video] [Project Page]

Imposing global map consistency by performing round-trips can enhance the navigation performance of embodied agents amidst various dynamics corruptions.

SGoLAM: Simultaneous Goal Localization and Mapping for Multi-Object Goal Navigation
Junho Kim, Eunsun Lee, Mingi Lee, Dongsu Zhang, and Young Min Kim
arXiv preprint, 2021. (2nd place in the MultiON challenge @ Embodied AI workshop in CVPR 2021) 
[arXiv] [Code] [Video]

By performing goal localization and mapping simultaneously with a simple color matching scheme, embodied agents can effectively perform multi-object goal navigation.

PICCOLO: Point Cloud-Centric Omnidirectional Localization
Junho Kim, Changwoon Choi, Hojun Jang, and Young Min Kim
ICCV 2021 
[arXiv] [Code] [Paper] [Video] [Project Page]

By minimizing a novel loss function that penalizes color discrepancies in 2D and 3D, effective localization can be performed using panorama images without learning.

N-ImageNet: Towards Robust, Fine-Grained Object Recognition with Event Cameras
Junho Kim, Jaehyeok Bae, Gangin Park, Dongsu Zhang, and Young Min Kim
ICCV 2021 
[arXiv] [Code] [Paper] [Video] [Project Page]

We propose a large scale dataset for event-based object recognition that enables systematic quantification of classification robustness amidst extreme lighting or motion.