We are excited to announce that our next speaker in ITU AI Seminar series is Prof. Jian Zhang.
Topic: Unsupervised Learning for 3D Point Cloud Understanding
Onsite Event: 12 June 2023, 11:00 - 13:00, Istanbul Technical University, Ayazaga Campus, Faculty of Electrical & Electronics Engineering, Ömer Korzay Conference Hall
Abstract
The unsupervised learning paradigm has demonstrated significant success in the 2D image and Natural Language Processing domains. However, a lack of research in the realm of 3D point clouds has attracted many groups’ interests. To address this disparity, we introduce a novel unsupervised approach for point clouds called SoftClu. It utilizes soft clustering to learn transferable point-level features. Our method operates under the assumption that points belonging to a cluster should exhibit proximity in both geometric and feature domains or spaces. Leveraging this idea, we enable self-training through a pseudo-label prediction task, and we frame our approach as an optimal transport problem to achieve a balanced distribution of the point cloud and prevent model collapse. Additionally, we formulate an unsupervised loss that minimizes cross-entropy between pseudo-labels and predicted labels. Remarkably, our method proves highly effective and outperforms current state-of-the-art techniques in tasks such as classification, part segmentation, and semantic segmentation, even without the need for data augmentation during our pre-training.
Short Bio
Dr. Jian Zhang is a Professor in the Faculty of Engineering and IT at UTS working on cutting-edge research in the areas of 2D and 3D- based computer vision, pattern recognition & data analytics, large- scale multimedia and cross-media content analysis. Prof Zhang has published more than 240 peer-reviewed papers (h-index 43, and over 6706 citations on Google Scholar) and holds 11 approved US patents. As Director of the Multimedia Data Analytics Lab in the Global Big Data Technologies Centre at UTS, Prof Zhang built an award-winning, cross-disciplinary UTS research and engineering team, uniquely focusing on computer vision research for the Agri-food sector. Since 2011, working with industry partners, he has led more than 20 research projects worth more than $4.73 million. Prof Zhang was the General Co-Chair and Technical Program Co-Chair of the International Conference on Multimedia and Expo (ICME) in 2012 and 2020 respectively; and the Technical Program Co-Chair and General Co- Chair of the IEEE Conference on Visual Communications and Image Processing in 2014 and 2019 respectively. He was an Associate Editor for the IEEE Transactions on Circuits and Systems for Video Technology (2006 – 2015) and the IEEE Transactions on Multimedia (2017- 2022) respectively. He was a Member of Technical Directions Board, IEEE Signal Processing Society (Jan. 2019 – Dec. 2020).
