Woobin Im

About Me

I'm a Ph.D. candidate at KAIST, South Korea. My research is focused on solving computer vision problems using machine learning. I am especially interested in motion understanding, so I have been working on video-related topics.

How to deal with the lack of labeled datasets is where I have expertise. I have mainly studied optical flow estimation with deep learning in an unsupervised or semi-supervised manner. Recently, I presented two optical flow papers (UnsupSimFlow and FlowSupervisor) at top-tier vision conferences; the two papers are related to unsupervised learning and semi-supervised learning, which are crucial in optical flow learning. I want to research similar topics related to motion understanding and label-efficient learning in the future.

Not only on the topics mentioned above, I'm knowledgeable about general computer vision tasks and robotics. I have implemented my code using Python, C++ with OpenCV, Tensorflow (1 and 2), and Pytorch while participating in many vision-related projects: face recognition, 3D teleportation, and multi-view line matching.

Award

• Naver Ph.D Fellowship Award, 2022.
• Finalist at Qualcomm Innovation Fellowship Korea (QIFK), 2020.

Education

• KAIST, Ph.D., Computer Science2018-current
  • Advisor: Professor Sung-Eui Yoon
• KAIST, M.S., Computer Science2016-2018
  • Advisor: Professor Hyun Seung Yang
• Yonsei University, B.S., Computer Science2012-2016

Publications

• Diffusion Probabilistic Models for Scene-Scale 3D Categorical Data
  Jumin Lee Woobin Im Sebin Lee Sung-Eui Yoon
  arXiv preprint
  [arxiv] [github]
• Scenario Generation by Action Scene-Graph Prediction
  Woobin Im Woo Jae Kim Sung-Eui Yoon
  Korea Software Congress (KSC)
  [web] [paper]
• Semi-Supervised Learning of Optical Flow by Flow Supervisor
  Woobin Im Sebin Lee Sung-Eui Yoon
  European Conference on Computer Vision (ECCV)
  [web] [arxiv] [github]
• In-N-Out: Towards Good Initialization for Inpainting and Outpainting
  Changho Jo Woobin Im Sung-Eui Yoon
  British Machine Vision Conference (BMVC)
  [web] [arxiv] [github]
• Self-Supervised Visual Odometry via Frame Interpolation
  Sebin Lee Woobin Im Sung-Eui Yoon
  Korea Robotics Society Annual Conference (KRoC)
  [paper]
• Unsupervised Learning of Optical Flow with Deep Feature Similarity
  Woobin Im Tae-Kyun Kim Sung-Eui Yoon
  European Conference on Computer Vision (ECCV)
  [web] [paper] [github]
• Combined Center Dispersion Loss Function for Deep Facial Expression Recognition
  Abhilasha Nanda Woobin Im Key-Sun Choi Hyun Seung Yang
  Pattern Recognition Letters
  [paper]
• Two-stream Spatiotemporal Feature for Video QA Task
  Chiwan Song Woobin Im Sung-Eui Yoon
  https://arxiv.org/abs/1907.05006
  [arxiv]
• Acoustic Material Estimation with Convolutional Neural Network
  Doheon Lee Inkyu An Woobin Im Sung-Eui Yoon
  Korea Robotics Society Annual Conference (KRoC)
  [paper]
• An Application of Convolutional-LSTM Network and Video QA
  Chiwan Song Woobin Im Sung-Eui Yoon
  Korea Computer Congress (KCC)
  [paper]
• Scale-Varying Triplet Ranking with Classification Loss for Facial Age Estimation
  Woobin Im Sungeun Hong Sung-Eui Yoon Hyun S. Yang
  Asian Conference on Computer Vision (ACCV)
  [web] [paper] [github]
• CBVMR: Content-Based Video-Music Retrieval Using Soft Intra-Modal Structure Constraint
  Sungeun Hong Woobin Im Hyun S. Yang
  Proceedings of the ACM international conference on Multimedia Retrieval (ICMR)
  [paper] [video]
• D3: Recognizing dynamic scenes with deep dual descriptor based on key frames and key segments
  Sungeun Hong Jongbin Ryu Woobin Im Hyun S. Yang
  Neurocomputing
  [paper]
• SSPP-DAN: Deep Domain Adaptation Network for Face Recognition with Single Sample Per Person
  Sungeun Hong Woobin Im Jongbin Ryu Hyun S. Yang
  International Conference on Image Processing (ICIP'17), IEEE
  Oral [paper]
• Convolutional Texture Networks based on Histogram Pooling
  Jongbin Ryu Sungeun Hong Woobin Im Hyun S. Yang
• Image-text multi-modal representation learning by adversarial backpropagation
  Gwangbeen Park Woobin Im
  arXiv preprint arXiv:1612.08354
• Deep CNN-based Person Identification using Facial and Clothing Features
  Sungeun Hong Woobin Im Junwoo Park Hyun S. Yang
  Jun 2016, Summer General Conference '16, IEEK

Patents

• Using Triplet-based Loss for Training Ordinal Classification Deep Models [detail]
• Partial Face Based Person Identification Across Poses [detail]

Projects

• Rendering for teleportaion in AR devices
  • Mar 2018 - Dec 2020, funded by National Research Foundation (NRF)
• Lab website renovation, Dec 2019 (sgvr.kaist.ac.kr)
• Age, gender, and expression recognition using face images
  • Dec 2016 - Feb 2018, funded by Korea Advanced Institute of Science and Technology
• Multi-view Face Recognition based on Deep Learning
  • May 2016 - May 2017, funded by Electronics and Telecommunications Research Institute (ETRI)

Experiences

• Teaching Assistants
  • GSAG-KAIST Research and Education Program, 1/2019-2/2019
  • CS206: Data Structure (Spring 2019), 3/2019-6/2019
  • CS688: Web-Scale Image Retrieval (Fall 2018), 9/2018-12/2018
  • CS101: Introduction to Programming, 9/2016-12/2017, 9/2018-12/2018
• Undergraduate Research Assistant
  • DB Lab, Yonsei University, 1/2014-2/2014