[CFAR Rising Star Lecture Series]Visual Perception with Little Supervision by Prof Tang Meng
19 Jan 2023 | 10.00am (Singapore Time)
Visual perception has made significant progress in the era of deep learning with big data. Perception models are often trained with big datasets that are fully annotated. However, obtaining full annotation is unscalable, in particular for 3D perception problems.
In this talk, Prof Tang Meng will discuss the techniques and ideas for visual perception with little supervision. First, he will introduce a regularised loss framework for image segmentation with weak supervision. He will also propose several regularisation losses, including pairwise MRF energy, entropy regularisation, and clustering criterion. Next, he will present FroDo; From Detections to 3D Objects, an end-to-end pipeline for multi-objects 3D detection reconstruction given localised RGB frames. Lastly, Prof Meng will also share some of his thoughts towards 3D perception with little supervision.
SPEAKER
Prof Tang Meng is an assistant professor of computer science at the University of California, Merced. He was a research scientist at Meta Reality Labs Research from 2019 to 2022. He obtained his Ph.D. in computer science from the University of Waterloo in 2019. He was also a recipient of the David R. Cheriton Graduate Scholarship from the Cheriton School of Computer Science at the University of Waterloo. Prof Tang’s research interests lie in 2D/3D visual perception, localisation, and scene understanding, with a focus on unsupervised learning, semi-supervised learning, and optimisation methods. His work appears at top-tier conferences and journals including ICCV/CVPR/ECCV/TPAMI/IJCV.