Thesis Defense
Friday, June 10
Title: Labeling by Example
Siddhartha Karthik Copesetty
1:00pm Friday, June 10, 2016
Room 326 Teague Building
Abstract
Sketch
Recognition is recognition of hand drawn diagrams. Recognizing sketches
instantaneously, is necessary to build beautiful interfaces with real
time feedback. There are various
techniques to quickly recognize sketches into ten or even twenty
classes. But, what if we have 100,000 sketches and want to classify them
into 3000 different classes? Using the existing techniques, it takes
forever and ever to accurately classify an incoming
sketch into one of these 3000 classes. For example, a class of hundred
sketches takes two hours to get classified into one of the 3000 classes.
This is very very slow, takes significant computation overhead and is
not practical. So, to make things faster,
we propose to have multiple stages of recognition. In the initial
stage, the sketch is recognized starting from the outer level, moving
level by level into the center of the sketch. This recognition is done
by matching it against a set of sketch domain descriptions,
resulting in a list of classes that the sketch could possible be, along
with the accuracy and precision for each. For the ones with accuracy
less than a threshold value, they go through a second stage of
recognition. In this stage, feature values are calculated
and evaluated against our model to accurately classify the sketch.
Thus, the time taken to classify such huge datasets of sketches
decreases significantly with increase in accuracy and precision.
Biography
Siddhartha
Karthik Copesetty is a master’s student in the Sketch Recognition Lab.
He completed his undergraduate degree in Computer Science at National
Institute of Technology,
Tiruchirappalli, India. He was a software engineering intern at Yahoo!,
Sunnyvale last summer.
No comments:
Post a Comment