Notice

Attended relation deep network on facial dynamics has been accepted as regular paper in IEEE Transactions on Information Forensics & Security.

The title is "Attended Relation Feature Representation of Facial Dynamics for Facial Authentication". The paper contribution is to propose new attended relation deep network to represent the relation feature on facial dynamics. In this paper, the relation feature representation on facial dynamics is proved to be useful to highly accurate facial authentication.

This paper has been written by Seong Tae Kim and Yong Man Ro.

Visually interpretable deep network has been accepted as regular paper in Physics in Medicine and Biology.

The title is "Visually interpretable deep network for diagnosis of breast masses on mammograms". The paper contribution is to propose new visual interpretable deep network for doctors to understand why diagnosis deep network predicts a malignancy decision. The visual interpretation based on doctor's medical description, is indeed very needed to give strong confidence in the deep learning based CAD.

This paper has been written by Seong Tae Kim, Jae-Hyeok Lee, and Hakmin Lee, and Yong Man Ro.

Mode variational LSTM robust to unseen modes of variation has been accepted in AAAI 2019 (acceptance rate: 16.2 %).

The paper title is " Mode Variational LSTM Robust to Unseen Modes of Variation: Application to Facial Expression Recognition". The spatio-temporal feature encoding in deep learning is essential for encoding the dynamics in video sequences. Recurrent neural networks, particularly long short-term memory (LSTM) units, have been popular as an efficient tool for encoding spatio-temporal features for moving objects. This paper presents the mode variational LSTM to encode spatiotemporal features robust to unseen modes of variation. The proposed mode variational LSTM has been verified to be useful for real-world spatio-temporal recognition.

This paper has been written by Wissam J. Baddar and Yong Man Ro.

Region-guided adversarial learning for anatomical landmark detection has been accepted in SPIE Medical Imaging 2019.

The title is "Region-guided adversarial learning for anatomical landmark detection in uterus ultrasound image". The paper contribution is to detect anatomical key points guided by anatomical region. New region-guided adversarial learning is proposed and anatomically meaningful landmarks are detected. The anatomical landmark detection applied in ultrasound images shows the state of art performance, which is practically useful to various ultrasound medical imaging system.

This paper has been written by Hongjoo Lee, Hak Gu Kim, Hyenok Park, Dongkuk Shin, and Yong Man Ro.

Visual evidence for interpreting diagnostic decision has been accepted as an oral in SPIE Medical Imaging 2019.

The title is "Visual evidence for interpreting diagnostic decision of deep neural network in computer-aided diagnosis". The paper contribution is to provide visual interpreting evidence for computer aided diagnostic decision. New deep network scheme providing visual interpreting evidence and associated interpretation guided learning algorithm are devised. The visual interpreting evidence results in this paper are very meaningful in the deep learning based CAD research, which can avoid non interpretable decisions being used so far.

This paper has been written by Seong Tae Kim, Jae-Hyeok Lee, and Yong Man Ro.

VRSA Net: VR Sickness Quality Assessment has been accepted in IEEE Trans. on Image Processing.

The title is "VRSA Net: VR Sickness Assessment considering Exceptional Motion for 360-degree VR Video". The paper contribution is to provide a possible algorithm to quantify VR sickness quality which is known an intractable problem. New deep network scheme called "VRSA Net" and associated new learning algorithm are devised. The paper results will be used as a very useful tool in VR research.

This paper has been written by Hak Gu Kim, Heoun-taek Lim, Sangmin Lee and Yong Man Ro.

Minho Park's paper has been accepted to the 25th international MultiMedia Modeling Conference (MMM 2019).

The title is "Photo-realistic Facial Emotion Synthesis using Multi-level Critic Networks with Multi-level Generative Model".

The paper contribution is to propose a new multi-level generative model and associated learning scheme with multi-level critic networks.

The multi-level generative model learned by the proposed multi-level critic networks demonstrates its usefulness in photo-realistic facial emotion synthesis.

This paper has been written by Minho Park, Hak Gu Kim, and Yong Man Ro.

2019년도 전기 박사과정(국비,KAIST장학), 석사과정 (국비,KAIST장학), 산학장학생 (KEPSI, EPSS, LGenius) 등을 모집합니다.

(http://admission.kaist.ac.kr/graduate/)

모집 연구분야:

Deep learning, Machine learning in computer vision and image processing (2D, 3D, VR), Language-visual embedding, Image processing, Medical imaging, Deep learning Quality Assessment

현재 진행중인 연구과제:

Explainable (Interpretable) Deep learning, Deep learning algorithms in computer vision, Recognition/Emotion recognition, 3D/VR quality assessment with deep learning approach, Medical Image analysis with deep learning, Language-visual embedding

최근 연구실 연구결과 - 링크 (LINK)

최근 연구실 석박사과정 딥러닝 관련 해외 학회 발표실적 - 링크 (LINK)

최근 연구실 석박사과정 해외 저널 실적 - 링크 (LINK)

을 참고하세요.

연구실 들어오고자 하는 학생은 노용만 교수님(ymro@kaist.ac.kr) 께 이메일/사전미팅추천합니다.

Jae-Hyeok Lee's paper has been accepted to ECCV 18 workshop (Bioimage computing). The title is " Feature2Mass: Visual Feature Processing in Latent Space for Realistic Labeled Mass Generation ". The paper contribution is to propose a new feature processing methodology using generative model with semantic vectors. The feature processing result done by the proposed method has been proved by demonstrating the generated target (here is mass image) which is quite realistic and useful in computer aided diagnosis (CAD). This paper has been written by Jae-Hyeok Lee, Seong Tae Kim, Hakmin Lee, and Yong Man Ro.

Seong Tae’ paper has been accepted to ECCV 2018. The title is "Facial Dynamics Interpreter Network: What are the Important Relations between Local Dynamics for Facial Trait Estimation?". The paper contribution is to propose novel dynamics Interpreter deep learning which provides the relations between locally moving parts. This paper has been written by Seong Tae Kim and Yong Man Ro.

Kihyun Kim, a master's in IVY lab under the supervision of Prof. Yong Man Ro, received the best paper award of IEEE ICCE-Asia 2018 which was held in June, 2018.

The title is " FSF-C Net: Face Spatial Frequency-Critic Network for Face Super Resolution".

The paper contribution is to proposed FSF-C Net to make realistic high resolution face by from low resolution face image. In the paper, face spatial frequency is preserved and detailed by Spatial Frequency-Critic Network.

Sincerely congratulations!!

Jung Uk and Jungsu’s paper has been accepted to 2018 IEEE International Conference on Image Processing (ICIP2018). The title is "Object Bounding Box Critic Networks for Occlusion-robust Object Detection in Road Scene". The paper contribution is to propose novel OBB critic networks, where novel plug and play style deep network (OBB Critic Net) and associated critic learning algorithm are devised. By the proposed method, the occlusion problem in object detection is much mitigated. This paper has been written by Jung Uk Kim, Jungsu Kwon, Hak Gu Kim and Yong Man Ro.

Sangmin S’ paper has been accepted to 2018 IEEE International Conference on Image Processing (ICIP2018). The title is "Adversarial spatial frequency domain critic learning for age and gender classification ". The paper contribution is to propose new critic learning in frequency domain, where unique features of spatial frequency is adopted in critic network to achieve high classification performance. This paper has been written by Sangmin S. Lee, Hak Gu Kim, Kihyun Kim and Yong Man Ro.

Kihyun Kim’ paper has been accepted to IEEE ICCE-Asia 2018. The title is " FSF-C Net: Face Spatial Frequency-Critic Network for Face Super Resolution". The paper contribution is to proposed FSF-C Net to make realistic high resolution face by from low resolution face image. In the paper, face spatial frequency is preserved and detailed by Spatial Frequency-Critic Network. This paper has been written by Kihyun Kim, Hak Gu Kim and Yong Man Ro.

Deep Learning 3D Visual Assessment has been accepted in IEEE Trans. on Circuits and Systems for Video Technology Note Hak Gu and Hyunwook's paper has been accepted to IEEE Transactions on Circuits and Systems for Video Technology. The title is "Binocular Fusion Net: Deep Learning Visual Comfort Assessment for Stereoscopic 3D". The paper contribution is to propose novel 3D video quality assessment, where new deep network (Binocular Fusion Net) and associated new learning algorithm are devised. The paper results will be very useful in 3D visual quality assessment. This paper has been written by Hak Gu Kim, Hyunwook Jeong, Heoun-taek Lim and Yong Man Ro.