Hello:wave:, I am a research intern at KIST, working on computer vision and machine learning.

My research goal is to design machine to learn and understand the world like human.
To achieve this goal, I do research on visual recognition, particularly

  1. Data- and Label-efficient Learning - [2], [3]
  2. Learning from Noisy Labels
  3. Video Understanding
  4. Perceptual Intelligence for Robotics



[3] Object Discovery via Contrastive Learning for Weakly Supervised Object Detection.
Jinhwan Seo, Wonho Bae, Danica J. Sutherland, Junhyug Noh*, Daijin Kim*
ECCV 2022 - paper | project page | code | bibtex

TL;DR: How can we find as many instances as possible in weakly supervised learning where we don’t know how many instances in an image? We introduced a novel approach to find ignored pseudo groundtruths via object discovery module guided by contrastive learning for weakly supervised object detection.

[2] Revisiting Class Activation Mapping for Learning from Imperfect Data.
Wonho Bae*, Junhyug Noh*, Jinhwan Seo, Gunhee Kim
CVPRW 2020 - paper | bibtex

TL;DR: Why does the output of weakly supervised object localization only highlight the discriminative part of the object? We investigated the phenomenon of part domination in weakly supervised learning and introduced effective ways to address it.

[1] Better to Follow, Follow to be Better: Towards Precise Supervision of Feature Super-Resolution for Small Object Detection.
Junhyug Noh, Wonho Bae, Wonhee Lee, Jinhwan Seo, Gunhee Kim
ICCV 2019 - paper | bibtex

TL;DR: It is difficult to recognize small objects due to the lack of feature information or distortion by pooling layer. We proposed feature super-resolution approach by providing a better supervision for SR model.