Faculty Recruiting Support CICS

Machine Learning and Friends Lunch (Online)

21 Oct
Thursday, 10/21/2021 12:00pm to 1:00pm
Virtual via Zoom
Machine Learning and Friends Lunch
Speaker: Ji Hou

Title: "Color and Geometry Learning in 3D Scene Understanding"

Abstract: 3D Scene Understanding aims to better understand our 3D environments/surroundings. Tasks, such as 3D detection, semantic segmentation and instance segmentation, enable a variety of important computer vision applications, such as autonomous robotics and VR/AR. Primarily, geometry signals are widely used in scene understanding tasks, such as TSDF or point cloud. The backbones like PointNet/PointNet++ and SparseConvNet are invented to extract network representations/features from voxels and points. To better use RGB-D Data for scene understanding tasks, such as ScanNet, where both RGB images and Depth are provided, our research tries to find better ways of jointly learning from color and geometry in scene understanding. In this talk, I will present our work on how to use color features in specific downstream tasks, such as instance segmentation and completion, as well as for better representation learning, such as leveraging 3D priors for 2D scene understanding tasks.

Bio: Ji Hou is currently a Ph.D. Candidate in Technical University of Munich with Prof. Matthias Niessner. He was previously interning at Facebook AI Research (FAIR). His research focuses on 3D Scene Understanding in terms of specific tasks, such as 3D detection or instance segmentation as well as 3D Transfer Learning, including data-efficient learning, and 3D priors for 2D tasks. He has published his research works on top-tier computer vision and machine learning conferences, such as CVPR/ICCV/NeuraIPS and serves as reviewers of top-tier journals, such as TPAMI and IJCV.

To obtain the Zoom link for this event, please see the event announcements from MLFL on the college email lists or contact Kalpesh Krishna.