Learning-based Geometric Image Matching with Modern Deep Learning Techniques

PhD Thesis Proposal Defence


Title: "Learning-based Geometric Image Matching with Modern Deep Learning 
Techniques"

by

Mr. Zixin LUO


Abstract:

Geometric image matching targets to establish reliable sparse correspondences 
that fit a static scene model across images under different perspective or 
lighting conditions, which serves as an essential basis for a broad range of 
computer vision tasks, including panorama stitching, visual localization, 
Structure-from-Motion (SfM), Simultaneous Localization and Mapping (SLAM), 
Augmented Reality (AR) and 3D reconstruction. During the past decade, 
hand-crafted local features and engineered geometric matchers have been widely 
used as the de-facto standard, upon which many popular applications are 
developed or already in commercial use in real scenarios. With the emerging of 
deep learning, a great amount of effort has been recently spent on integrating 
the image matching pipeline into modern neural network architectures in a 
differentiable manner. In this survey, we will first review the recent 
achievements on learning-based image matching techniques, and then elaborate 
the methods we have proposed that give rise to state-of-the-art results on 
several important benchmarking datasets. More specifically, we decompose the 
learning-based image matching into four sub-problems, including 1) a keypoint 
detector and 2) a keypoint descriptor for local feature extraction. Next, 3) an 
image retrieval system that shortlists the matching candidates from a large 
image collection and finally, 4) a feature matcher that solves the geometry 
model. To facilitate the above research, we further present a data generation 
pipeline that offers accurate and rich geometric learning labels automatically 
from off-the-shelf 3D reconstructions. Through extensive evaluations, we 
demonstrate the superiority of the integration of learning-based image matching 
methods in real applications, and show great potential for future improvements 
in this area.


Date:			Wednesday, 11 December 2019

Time:                  	2:00pm - 4:00pm

Venue:                  Room 2132B
                         (lift 19)

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


**** ALL are Welcome ****