Transfer Learning by Optical Flow

Distribution Rights granted to UMBC by the author.
Access limited to the UMBC community. Item may possibly be obtained via Interlibrary Loan thorugh a local library, pending author/copyright holder's permission.
This item is likely protected under Title 17 of the U.S. Copyright Law. Unless on a Creative Commons license, for uses protected by Copyright Law, contact the copyright holder or the author.

Abstract

One of the biggest challenges in deep learning field of research is the need for having a large amount of annotated data. Self-supervised learning is one of the methods to tackle this issue. In this method, a new task is designed to learn features without having annotated data available. These learned features can be transferred to a new task by fine-tuning the pre-trained model. This process of transferring the learned features on one task trained on unannotated data to a new task is called transfer learning. The objective of this study is to learn low level features through novel self-supervised task, with the hypotheses being that the learned features from self-supervised task would improve object classification in the supervised learning. In addition, there is a significant reduction in the complexity of the overall model when primarily representation is learned in a deep network and the resulting knowledge is transferred to the second task. Compared to some of the existing self-supervised methods, transfer learning method described in this study is shown to have achieved superior results in terms of accuracy on object classification on PASCAL VOC 2007 dataset.

Transfer Learning by Optical Flow

Files

Links to Files

Permanent Link

Collections

Author/Creator

Author/Creator ORCID

Date

Type of Work

Department

Program

Citation of Original Publication

Rights

Subjects

Abstract