Light field saliency map estimation for 3D objects with the usage of light field cameras
Loading...
Date
2023-03-08
Authors
Ντόγκας, Κωνσταντίνος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Light field technology has been rapidly advancing in the later years for many computer
vision tasks such as depth estimation and salient object extraction. One main factor
to this advance, is the commercialization of light field cameras which have been made
cheaper and improved. Saliency map extraction on images has also been crucial in
many computer vision problems in order to pay attention to the most detailed aspects
of an image or a video. In this thesis, we explore the possibility of salient map
extraction with the usage of light field cameras. This task is not easy due to the lack of
solid light field dataset from which we can exploit the rich information provided to us
by light field images.
In this thesis we will suggest a methodology in order to create a synthetic dataset of
light field images which can be later used for saliency map estimation. Furthermore,
we will train a well-known neural network used for depth estimation, called EpiNET,
in order to achieve the saliency map extraction. We achieve to perform the map
extraction with a 10-20% error and prove that this task is possible and we can open
the way for more research to be done on this subject.
Description
Keywords
Neural networks, Salient map extraction, Computer vision, Light field