Skip navigation
Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp01sq87bx475
Title: Weakly Supervised Object Detection: Hierarchical Learning with CornerNet
Authors: Li, Matthew
Advisors: Deng, Jia
Department: Computer Science
Certificate Program: Center for Statistics and Machine Learning
Class Year: 2019
Abstract: We propose a new network model and training strategy to improve object detection accuracy for classes with few or no localized annotated training samples. Our approach extracts ideas from related works and applies them to CornerNet, a state-of-the-art object detection system. Specifically, we modify the original CornerNet architecture to facilitate hierarchical learning for over 9000 object classes. This allows us to train jointly on the COCO detection dataset as well as the ImageNet classification dataset, in a semi-supervised fashion. First, we demonstrate that the hierarchical learning and related modifications do not significantly degrade the performance over the original COCO validation set. Finally, we validate the performance over weakly supervised classes using the ImageNet detection task, for which 156 of 200 classes do not have object detection samples in our training set.
URI: http://arks.princeton.edu/ark:/88435/dsp01sq87bx475
Type of Material: Princeton University Senior Theses
Language: en
Appears in Collections:Computer Science, 1988-2019

Files in This Item:
File Description SizeFormat 
LI-MATTHEW-THESIS.pdf2.18 MBAdobe PDF    Request a copy


Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.