Date＆Time：13:30 – 17:30, December 28, 2021 Location: Prince Ballroom
Intro: Excellent Doctoral forum is a forum for CSIG Excellent Doctoral Dissertation Award Recipients to present their recent work, to receive feedback from senior researchers in the field, and to share their experiences with current fellow PhD students. It is an opportunity for young researchers to discuss their first results, future plans, challenges and opportunities towards developing their research career.
Organizer: Ruiping Wang (Chinese Academy of Sciences) and Xiang Bai (Huazhong Univesity of Science and Technology)
Research on Hyperspectral and Multispectral Image Fusion
Adversarial Robustness in Computer Vision: Attack, Defense, and Beyond
Image Segmentation: Techniques and Challenges
Towards robust and efficient person search
Synthesizing Graphic Layouts with Vector-Wireframe Adversarial Networks
Title: Research on Hyperspectral and Multispectral Image Fusion
Speaker: Renwei Dian，Hunan University
Abstract: Hyperspectral image (HSI) contains abundant spectral information, which enables accurate identification of materials. Therefore, they find broad application in earth observation, military detection, environmental monitoring, geological exploration, and precision agriculture. However, there is the trade-off between the spatial resolution and the spectral resolution for the imaging sensor, that is, HSI with high spectral resolution suffers from low spatial resolution, which limits the application of HSI. The imaging sensor can acquire the multispectral image (MSI) of much higher spatial resolution. Hence, fusing low-resolution HSI with high-resolution MSI of the same scene has become an effective way to obtain high-resolution HSI (fused HSI). This report focuses on the aforementioned main problems and challenges in HSI-MSI fusion, and introduces the newly proposed tensor based and CNN based fusion methods.
Biography: Dr. Renwei Dian is currently an associate researcher in the College of Electrical and Information Engineering, Hunan University. He received the B.S. degree from Wuhan University of Science and Technology, Wuhan, China, in 2015, and the Ph.D. degree from Hunan University, Changsha, China, in 2020, respectively. From November 2017 to November 2018, he is a visiting Ph.D. student with University of Lisbon, Lisbon, Portugal, supported by the China Scholarship Council. He has published over ten papers on top conferences and top journals, such as IEEE TIP, IEEE CVPR, IEEE TNNLS, and etc. He was a finalist for the Best Student Paper Award at the International Geoscience and Remote Sensing Symposium (IGARSS) 2018. He was awarded the Fellowship of China National Postdoctoral Program for Innovative Talents in 2020 and Excellent Doctoral Dissertation by China Society of Image and Graphics in 2020.
Title: Adversarial Robustness in Computer Vision: Attack, Defense, and Beyond
Speaker: Hong Liu，National Institute of Informatics
Abstract: This talk provides an overview of our recent progress on adversarial robustness in computer vision. It consists of three main parts, such as adversarial attacks, adversarial defenses, and their applications. For the adversarial attack, this talk will present the analysis of the visual low-frequency property and introduce a universal attack and a black-box attack. In terms of adversarial defense, this talk will introduce two anti-models, one is a NAS-based defense based on anti-bandits, and the other is about anti-perturbation inference. In addition to attacks and defenses, this talk will conclude with an analysis of the sensitivity of adversarial samples in visual retrieval tasks.
Biography: Dr. Hong Liu is currently a JSPS Fellowship researcher at the National Institute of Informatics, Japan. Before that, he got his Ph.D. degree in Computer Science at Xiamen University. His research interests include large-scale image retrieval and deep adversarial learning. He has published about 20+ papers in top journals and conferences such as TPAMI, IJCV, TIP, CVPR, ICCV, ECCV, ICLR, etc. He was awarded the Outstanding Doctoral Dissertation Award of the CSIG, JSPS International Fellowship, and Top 100 Chinese New Stars in Artificial Intelligence by Baidu Scholar.
Title: Image Segmentation: Techniques and Challenges
Speaker: Pingping Zhang，Dalian University of Technology
Abstract: Image segmentation is one of the basic problems in computer vision and artificial intelligence field. In recent years, deep learning based image segmentation methods have achieved great success in autonomous driving, intelligent interaction, multimedia analysis and other fields. However, it is still difficult to obtain high precision, high efficiency and high resolution prediction results. In addition, the complementary effect of multi-modal data also provides a new paradigm for the study of image segmentation. This talk will first review the mainstream image segmentation methods, then analyze existing problems and challenges. On these basis, this talk will give new image segmentation methods and discuss the application of relevant methods in the field of autonomous driving and 3D vision, to provide new ideas for future research.
Biography: Dr. Pingping Zhang is currently an Associate Professor with the School of Artificial Intelligence, Dalian University of Technology. He is also a member of IIAU-Lab (Intelligent Image Analysis and Understanding Lab). Before that, he was a Ph.D candidate supervised by Prof. Huchuan Lu and Prof. Hongyu Wang. His research interests include deep learning, computer vision and multimedia analysis. He serves as the PC member or reviewer for top journals and conferences, such as IEEE TPAMI,TIP,TITS、TMM,TCSVT,PR,CVPR,ICCV,ECCV,NeurIPS,AAAI,IJCAI. He has published 50+ peer reviewed papers (CCF-A/B with 2000+ citations) in artificial intelligence and computer vision. From 2017 to 2018, he has been with ACRV Lab as a visiting researcher supervised by Prof. Chunhua Shen. He won the Young Talents of Dalian High-level Talents, Xinghai Young Talent Training Program of Dalian University of Technology, and CSIG Outstanding Doctoral Thesis for year 2020.
Title: Towards robust and efficient person search
Speaker: Yichao Yan，Shanghai Jiao Tong University
Abstract: Person search, which aims to localize and identify a target person from a gallery of realistic, uncropped scene images, has recently emerged as a practical task with real-world applications. To tackle this task, we need to address two fundamental tasks in computer vision, i.e., pedestrian detection and person re-identification (re-id). Both detection and re-id are very challenging tasks and have received tremendous attention in the past decade. This report includes a brief review of the development of person search frameworks, i.e., one-step vs. two-step models. It also includes several of our recent works towards robust and efficient person search. For example, we build a context graph to improve robustness, while constructing an anchor-free model to yield high efficiency. Furthermore, we introduce the weakly supervised and unsupervised setting for person search, which shows great potential in reducing manual annotations.
Biography: Dr. Yichao Yan is an assistant professor at the AI Institute of Shanghai Jiao Tong University. Before that, he was a research scientist at the Inception Institute of Artificial Intelligence (IIAI). He received his Ph.D. degree from Shanghai Jiao Tong University in 2019. His research interests include computer vision and machine learning, e.g., person re-identification, image/video generation, action recognition, and neural rendering. He has published 20+ peer-reviewed papers in top venues in AI and has filed 7 patents. In 2020, he won the Excellent Doctoral Dissertation Award from the China Society of Image and Graphics.
Title: Synthesizing Graphic Layouts with Vector-Wireframe Adversarial Networks
Speaker: Jianan Li， Beijing Institute of Technology
Abstract: Layout is important for graphic design and scene generation. We propose a novel Generative Adversarial Network, called LayoutGAN, that synthesizes layouts by modeling geometric relations of different types of 2D elements. The generator of LayoutGAN takes as input a set of randomly-placed 2D graphic elements, represented by vectors and uses self-attention modules to refine their labels and geometric parameters jointly to produce a realistic layout. Accurate alignment is critical for good layouts. We thus propose a novel differentiable wireframe rendering layer that maps the generated layout to a wireframe image, upon which a CNN-based discriminator is used to optimize the layouts in image space. We validate the effectiveness of LayoutGAN in various experiments including MNIST digit generation, document layout generation, clipart abstract scene generation, and tangram graphic design. Besides, we further introduce an attribute-conditioned LayoutGAN with user data by incorporating elements’ content-based attributes as conditions, which demonstrates successful practical applications for automatic advertisement design.
Biography: Dr. Jianan Li is currently an Assistant Professor at School of Optoelectronics, Beijing Institute of Technology, where he received his Ph.D. degree in 2019. From 2019 to 2020, he worked as a research fellow at National University of Singapore, where he once studied as a two-year joint training Ph.D. student. From 2017 to 2018, he worked as an intern at Adobe Research. He has published nearly 30 papers in top journals and conferences. His current research interests mainly include computer vision, computational imaging and real-time video processing. He won the Wang Daheng Optics Award for year 2018 and CSIG Outstanding Doctoral Thesis 2020.