Semantic Segmentation from 2D to 3D: A Survey

PhD Qualifying Examination


Title: "Semantic Segmentation from 2D to 3D: A Survey"

by

Mr. Mingmin ZHEN


Abstract:

Nowadays, large-scale 3D reconstruction of urban city has achieved great 
success. With plentiful reconstructed 3D urban city models, one of the 
most urgent demands but also a big challenge is to understand a 3D 
cityscape, specifically, to parse urban scenes into semantic categories. 
To meet this requirement, we give a review for semantic segmentation on 
both 2D images and 3D scenes. Semantic segmenation is to assign 
corresponding label to each pixel of 2D image or each point (triangle) of 
3D point cloud (mesh), which helps to understand 2D or 3D scenes.

In this survey, techniques on deep learning methods for 2D image semantic 
segmentation is introduced in detail, mainly including fully convolutional 
network (FCN) and its variants. In addition, some post-processing methods 
based on conditional random field (CRF) are also presented.

Then we review several classical methods for semantic segmentation on 3D 
scenes. Finally, we conclude some limitations and open problems in current 
state-of-the-arts, and disscuss possible future research directions.


Date:			Tuesday, 20 August 2019

Time:                  	4:00pm - 6:00pm

Venue:                  Room 5501
                         Lifts 25/26

Committee Members:	Prof. Long Quan (Supervisor)
 			Dr. Pedro Sander (Chairperson)
 			Dr. Qifeng Chen
 			Prof. Chiew-Lan Tai


**** ALL are Welcome ****