[4eyes] MASTERS PRESENTATION: MAY 23 @ 9AM, Qiaodong Cui HFH 1132

[Pradeep Sen]

In case you guys are interested, CQ's doing his Masters project
presentation RIGHT NOW
in HFH 1132.

Best,

-Pradeep

---
Pradeep Sen
Associate Professor
UCSB MIRAGE Lab
Dept. of Electrical & Computer Engineering
University of California, Santa Barbara
Santa Barbara, CA 93106-9560

---------- Forwarded message ----------
From: Lesley Fredrickson <lesley at mat.ucsb.edu>
Date: Mon, May 21, 2018 at 3:23 PM
Subject: THIS WEEK: MASTERS PRESENTATION: MAY 23 @ 9AM, Qiaodong Cui
To: faculty at lists.mat.ucsb.edu, staff at mat.ucsb.edu, students at mat.ucsb.edu


*Please reserve this time and date for Qiaodong Cui's Master's
Presentation:*

*Date*:  Wednesday, May 23th
*Time*: 9:00 AM
*Location*: CS conference room, HFH 1132.

*Committee*:
Theodore Kim (Co-Chair)
Pradeep Sen (Co-Chair)
Yon Visell

*Title*:  GraphMatch: Efficient Large-Scale Graph Construction for
Structure from Motion

*Abstract:*

Structure from motion (SfM) is a powerful tool to achieve 3D
reconstructions from a large collection of photos. A crucial step of
structure from motion is to construct a matching graph from the photo
collections. In a matching graph, images are nodes and edges represent
matching image pairs that share visual content. Despite many recent
progress, constructing a good matching graph is still a bottleneck for most
SfM algorithms, since most images do not match.

This project presents GraphMatch, an approximate yet efficient method for
building the matching graph for large-scale SfM pipelines. First, we show
by using a score computed from Fisher vectors of any two images, a better
prior can be obtained to indicate image matching than the conventional
vocabulary tree prior. Next, by using the graph distance between vertices
in the underlying matching graph as a second prior, we show matching graph
can be greatly densified. Finally, inspired by the PatchMatch algorithm, by
combining these two priors into an iterative “sample-and-propagate” scheme,
GraphMatch finds the most image pairs as compared to competing, approximate
methods while at the same time being the most efficient.

------------------

-- 




-- 
___________________________
Lesley A. Fredrickson
Graduate Program Advisor
Media Arts & Technology Program
Phelps 3313
893-5439
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cs.ucsb.edu/pipermail/ilab-users/attachments/20180523/02b2650a/attachment.html>