Prof. Zhenzhong Wei
Beihang University, China
Wei Zhenzhong is a professor and PhD supervisor in the School of Instrumentation Science and Opto-electronics Engineering of Beihang University. He is a distinguished professor of the Chang Jiang Scholars Program and an awardee of the National Science Fund for Distinguished Young Scholars.
Prof. Wei received his doctor's degree from Beihang University in 2003. His research focuses on computer vision, particularly in image processing and pattern recognition, position and orientation measurement and object tracking, etc. He has made several contributions to the field. For example, he proposed the new methods of field flexible calibration and expanded the field calibration system of visual measurement. He has led more than ten national projects and published 36 papers in journals indexed by the Science Citation Index. He also holds 32 patents. He has won two second prizes of the State Technological Invention Award and four provincial or ministry-level prizes.
Prof. Limei Song
Tiangong University, China
Prof. Limei Song received her bachelor's, master's and doctor's degrees from Tianjin University in 1999, 2001 and 2004. She spent a year as a visiting scholar at Tsinghua University from 2016 to 2017. She was selected as the first-level talent of "131" in Tianjin. She is the executive member of Tianjin Artificial Intelligence Council, and the member of Tianjin Robotics Council. She won two second prizes and two third prizes of Tianjin Science and Technology Progress. She won the first prize of the second Tianjin "Haihe Talents" Post-doctoral Innovation Group and the National Bronze Prize of the first National Post-doctoral Innovation and Entrepreneurship Competition. She was awarded the Tianjin 5.1 Labor Medal and the Tianjin "March 8 Red-banner pacesetter" and other honorary titles. Her research interests are image processing, pattern recognition, detection technology and automation devices, artificial intelligence, etc. She presided over two National Natural Science Foundation projects and more than 10 provincial research projects.
Speech Title: 3D Vision-Guided Robot Intelligent Processing Technology and Applications
Abstract: Robots have been widely used in various industries in the national economy. 3D vision inspection systems are the high-precision three-dimensional "eyes" of robots, which can quickly obtain accurate 3D data of target scenes and guide robots to carry out intelligent processing and manufacturing. This report will introduce the application of the team's self-developed monocular and binocular 3D precision vision inspection systems in industries such as end-of-life vehicle dismantling, 3D intelligent polishing of shoe lasts, and casting polishing. The 3D vision guides the heavy-duty robotic arm to automatically complete the positioning and grasping of end-of-life vehicles, improving the accuracy and efficiency of grasping; the data after 3D vision imaging can automatically carry out path and trajectory planning to guide the robot to carry out intelligent polishing of shoe lasts, which not only eliminates the process of programming the robot by human but also automatically adapts to different sizes and dimensions of shoe lasts; 3D vision guides the robot to carry out The intelligent separation algorithm separates the target and area to be polished from the main casting, enabling precise and efficient polishing. 3D vision-guided robotic intelligent processing technology can avoid the impact of highly dangerous, noisy, and polluting work on human life and health, and improve processing efficiency and processing quality in the manufacturing industry.
Prof. Haiyan Li
Yunnan University, China
Haiyan Li ,Ph.D. Professor, Doctoral supervisor, School of Information Science and Engineering, Yunnan University, China. Selected as one of the Yunnan Ten Thousand Talents Plan "Famous Yunling Teachers". Presided over 5 NSFC and provincial projects, and published more than 70 papers, in which more than 60 are SIC or EI indexed. Published 5 textbooks and monographs. Won more than 10 patents and software copyrights, and more than 120 international, national and provincial teaching awards.
Speech Title: Resampling-based Cost Loss Attention Network for Explainable Imbalanced Diabetic Retinopathy Grading
Abstract: Diabetic retinopathy (DR) is considered to be one of the most common diseases that cause blindness currently. However, DR grading methods are still challenged by the presence of imbalanced class distributions, small lesions, low accuracy for less sample classes and poor explainability. To address the issues, a resampling-based cost loss attention network for explainable imbalanced diabetic retinopathy grading is proposed. Firstly, the progressively-balanced resampling strategy is put forward to create a balanced training data by mixing the two sets of samples obtained from instance-based sampling and class-based sampling. Subsequently, a neuron and normalized channel- spatial attention module (Neu-NCSAM) is designed to learn the global features with 3-D weights and apply a weight sparsity penalty to the attention module to suppress irrelevant channels or pixels, thereby capturing detailed small lesion information. Thereafter, a weighted loss function of the Cost-Sensitive (CS) regularization and Gaussian label smoothing loss, called cost loss, is proposed to intelligently penalize the incorrect predictions and thus to improve the grading accuracy for less sample classes. Finally, the Gradient-weighted Class Activation Mapping (Grad-CAM) is performed to acquire the localization map of the questionable lesions in order to visually interpret and understand the effect of our model. Comprehensive experiments are carried out on two public datasets, and the subjective and objective results demonstrate that the proposed network outperforms the state-of-the-art methods, achieving the best DR grading results with 83.46%, 60.44%, 65.18%, 63.69% and 92.26% for Kappa, BACC, MCC, F1 and mAUC, respectively.
Assoc. Prof. Zhen Ye
Chang'an University, China
She received the B.S. degree in Electronic & Information Engineering M.S. and the Ph.D. degree in information & communication engineering from Northwestern Polytechnical University, China, in 2007, 2010 and 2015, respectively. Meanwhile, she spent one year as a co-training Ph. D student from September, 2011 to October, 2012 in Mississippi State University, USA. She is currently an Associate Professor with the School of Electronics and Control Engineering, Chang’an University, Xi’an. Her research interests include remote sensing, pattern recognition and machine learning.
Speech Title: Multi-Scale Spatial-Spectral Feature Extraction Based on Dilated Convolution for Hyperspectral Image Classification
Abstract: Convolutional neural networks have garnered increasing interest for the supervised classiﬁcation of hyperspectral imagery. However, images with a wide variety of spatial land-cover sizes can hinder the feature-extraction ability of traditional convolutional networks. Consequently, many approaches intended to extract multiscale features have emerged; these techniques typically extract features in multiple parallel branches using convolutions of differing kernel sizes with concatenation or addition employed to fuse the features resulting from the various branches. In contrast, the present work explores a multiscale spatial-spectral feature-extraction network that operates in a more granular manner. Speciﬁcally, in the proposed network, a dual-branch structure expands the convolutional receptive ﬁelds, applying dense connection and cascaded strategy for spectral and spatial multi-scale feature extraction, respectively. The experimental results show that the classification performance of our methods outstands that of several state-of-the-art methods, even under small-sample-size (SSS) situation.
Assoc. Prof. Ioannis Ivrissimtzis
Durham University, UK
Ioannis Ivrissimtzis is an Associate Professor at the Department of Computer Science at Durham University. His research has contributed to the areas of subdivision surfaces, surface reconstruction, and digital 3D watermarking and steganalysis. His recent research contributions include work in the area of applied machine learning, tackling problems such as wind turbine early fault diagnostics, blind source separation in GNSS time series, and face anti-spoofing.
Speech Title: Race Bias Analysis in Face Anti-spoofing
Abstract: In recent years, the study of bias in Machine Learning has received considerable research attention. In this talk, we propose the use of a set of statistical methods for the systematic study of race bias and present a case study based on a VQ-VAE face anti-spoofing algorithm. The main characteristics of the case study are: the focus is on analysing bias in bona fide errors, where significant ethical issues lie; the analysis is not restricted to the final binary classification outcomes, but also covers the classifier's scalar responses and the latent space; the threshold determining the classifier’s operating point is considered variable. The results show that race bias does not always come from differences in the mean responses of the various populations. Instead, it can be better understood as the combined effect of several possible statistical characteristics of their distributions: different means; different variances; bimodal behaviour; existence of outliers. Joint work with Latifah Abduh.
Assoc. Prof. Md Baharul Islam
American University of Malta, Malta
Dr. Md Baharul Islam is an Associate Professor in Computer Science at the American University of Malta (AUM), Malta, and an Adjunct Professor of Computer Engineering at Bahcesehir University, Istanbul, Turkey. Before joining the AUM, he was a Postdoctoral Research Fellow at AI and Augmented Vision Lab of the Miller School of Medicine, University of Miami, United States. He had completed his Ph.D. in Computer Science from Multimedia University in Malaysia, and M.Sc. in Digital Media from Nanyang Technological University in Singapore. Dr. Islam has more than 15 years of working experience in teaching and cutting-edge research in image processing and computer vision area. His current research interests lie in 3D stereoscopic media processing, computer vision, and AR/VR-based vision rehabilitation. Dr. Islam secured several (four) gold medals from different international scientific and technological competitions and received (three) best paper awards from different international conferences, workshops, and symposiums. He received the IEEE SPS Research Excellence Award in 2018. He authored/co-authored more than 60 international peer-reviewed research papers, including journal articles, conference proceedings, books, and book chapters. Dr. Islam received TUBITAK 2232 Outstanding Researchers Award and Grant that support funds for up to 5 postgraduate students under his supervision. Dr. Islam is an active IEEE Senior Member since 2018.
Assoc. Prof. Peixian Zhuang
University of Science and Technology Beijing, China
Peixian Zhuang is currently an Associate Professor in the Key Laboratory of Knowledge Automation for Industrial Processes, Ministry of Education, the School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing, China. From 2020 to 2022, he was a Postdoctoral Fellow and an Assistant Research Fellow with the Department of Automation, Tsinghua University, Beijing, China. He received the Ph.D. degree from Xiamen University, Xiamen, China, in 2016. From 2017 to 2020, he was a Lecturer and the Master Supervisor with Nanjing University of Information Science and Technology, Nanjing, China.
His research interests involve sparse representation, Bayesian modeling, deep learning, and calcium signal processing. He has published more than 30 research papers (IEEE TIP, IEEE TRGS, IEEE TCVST, IEEE ICIP, IEEE JOE, EAAI, etc.) in Image Processing and Computer Vision, with more than 800 times of Google citations and two papers of highly cited ESI. He was a recipient of the Outstanding Doctoral Dissertations of Fujian province in 2017. He served as the Guest Editor of Journal of Electronics and Information Technology in 2021, the session chair of IEEE International Conference on Signal and Image Processing in 2019, the Top 25% reviewer of the Association for the Advancement of Artificial Intelligence (AAAI) in 2021, the invited speaker of International Conference on Optics and Image Processing in 2022, etc.
Speech Title: Underwater Image Enhancement With Hyper-Laplacian Reflectance Priors
Abstract: We develop a hyper-laplacian reflectance priors inspired retinex variational model for enhancing single underwater images. The hyper-laplacian reflectance priors are proposed with the L1/2-norm penalty on multi-order gradients of the reflectance, which can exploit sparsity-promoting and complete-comprehensive reflectance to booth boost salient and fine-scale structures and recover authentic color naturalness. Besides, the L2 norm is found to be suitable for accurately estimating the illumination. As a result, we transform a complex underwater image enhancement issue into simple sub-problems, where their optimal solutions can be theoretically analyzed and proved. For addressing the proposed model, we present an alternating minimization algorithm that is efficient on element-wise operations and independent of additional underwater prior knowledges. Final experiments demonstrates the superiority of our method in both subjective results and objective assessments over several conventional and state-of-the-art methods.
Assoc. Prof. Kangjian He
Yunnan University, China
Kangjian He is an Associate Professor at School of Information Science and Engineering, Yunnan University. He was selected as the “Donglu Talent Young Scholar” of Yunnan University. He received the Ph.D. degree from Yunnan University, Kunming, China, in 2019. From 2020 to 2022, he is a Post-Doctoral Research Fellow with the School of Information Science and Engineering, Yunnan University. He presided over two NSFC and provincial projects. He has authored and co-authored over 50 papers in refereed international journals and conferences. His current research interests include multimodal image processing, neural network theory and applications.
Speech Title: Research and Application of Task-driven Multi-modality Information Fusion
Abstract: Multimodal data can provide more detailed and sufficient information than the single modal data source. Multi-modality information fusion can achieve a more accurate and comprehensive description of the target. How to extract and fuse effective information from multi-modality data is crucial for computer vision tasks under the background of big data. Most of the existing fusion schemes are driven by data or models, which pursue high evaluation indicators but ignore the user perception and the support for subsequent high-level tasks. In this talk, we take the application of multi-modality medical imaging in computer-aided diagnosis as an example to introduce our research on task-driven multi-modality information fusion.
Assoc. Prof. Janaka Rajapakse
Tainan National University of the Arts, Taiwan, China
Janaka Rajapakse is an Associate Professor at the Graduate Institute of Animation and Film Art, Tainan National University of the Arts, Taiwan. He is also a visiting scholar at the Department of Media Engineering, The Graduate School of Engineering, Tokyo Polytechnic University of Japan. He was an Assistant Professor in the CG Application Laboratory, Graduate School of Engineering, and special researcher in the Center for Hyper Media Research, Tokyo Polytechnic University, Japan. He received his B.Sc. degree in computer science from the University of Colombo, Sri Lanka. He won the Japanese Government Scholarship for his higher studies in Japan. He received his M.Sc. and Ph.D. degrees from the Japan Advanced Institute of Science and Technology in 2005 and 2008. His research interests include computer animations, virtual reality, augmented reality, haptic interfaces, artificial intelligence, motion capture techniques, computer graphics, 3D printing, interactive media, and Kansei engineering. Prof. Rajapakse is the author or co-author of over eighty academic publications. He is a member of the Society for Art and Science, Motion Capture Society (MCS), Association of AISIA NETWORK BEYOND DESIGN, ASIAGRAPH Association, IEEE, and SIG-Design Creativity.
Speech Title: Developments of Passive Interactions in Virtual and Mixed Realities
Abstract: Recent VR/MR/XR applications have already provided interactions via devices and controllers. To create realistic communication between virtual and physical worlds, we need to develop passive interaction methods and what it means to be in real interaction in a virtual environment. This speech will focus on the differences between "active interaction" and "passive interaction." This speech will also explore how the passive interactions will be made by gesture recognition methods and machine learning in extended realities, as well as the devices, platforms, and development engines they will use to build it and how to implement it. Furthermore, this talk also explores how to evaluate passive interactions in immersive environments and the role of user feedback.
Asst. Prof. Arren Matthew Antioquia
De La Salle University, Philippines
Arren Matthew C. Antioquia received his Master of Science in Computer Science from both De La Salle University (DLSU) and the National Taiwan University of Science and Technology (NTUST), as part of the ladderized BS-MS program and the Dual Masters Program between DLSU and NTUST. He is an Assistant Professor at the Department of Software Technology of De La Salle University in the Philippines. His most recent major publication introduced a computationally efficient way of fusing features from different layers of convolutional neural networks, which resulted in a faster and more accurate general object detection performance compared to state-of-the-art techniques. His research interest includes applying deep learning techniques in computer vision applications, specifically in problems involving image classification and object detection.
Speech Title: Towards Effective Multi-Scale Object Detection
Abstract: Despite recent improvements, the arbitrary sizes of objects still impede the predictive ability of object detectors. Recent solutions combine feature maps of different receptive fields to detect multi-scale objects. However, these methods have large computational costs resulting in slower inference time, which is not practical for real-time applications. Contrarily, fusion methods depending on large networks with many skip connections demand larger memory requirements, prohibiting usage in devices with limited memory. In this talk, we will discuss recent works on improving the performance on multi-scale object detection, together with their advantages, issues, and suggested improvements.