Efficient Visual Learning for Scene Understanding

Author:
Description:
Significant advancements in scene understanding have been driven by deep neural networks. These learning-based frameworks enhance performance through extensive training datasets and a large number of trainable parameters. However, they are less scalable and require substantial computational and financial resources. This dissertation investigates two aspects of efficient visual learning for scene understanding: label-efficient learning and parameter-efficient learning. To reduce label supervision in instance-level scene understanding tasks, we develop a series of semi-supervised learning frameworks. These frameworks improve the label efficiency under various detector architectures and unconstrained data settings. To reduce parameter usage in multi-task training, we re-evaluate parameter-efficient methods from NLP for scene understanding and then propose a more parameter-efficient method for vision architectures. These advancements demonstrate the practicality and adaptability of efficient learning frameworks in diverse, resource-constrained environments. ; Ph.D.
Publisher:
Georgia Institute of Technology
Contributors:
Kira, Zsolt ; Hoffman, Judy ; Heck, Larry ; Davenport, Mark ; Yang, Diyi ; Electrical and Computer Engineering
Year of Publication:
2024-01-10T18:47:36Z
Document Type:
Text ; Dissertation ; [Doctoral and postdoctoral thesis]
Language:
en_US
Subjects:
Label-efficient Learning ; Parameter-efficient Learning ; Semi-supervised Learning ; Scene Understanding ; Object Detection ; Semantic Segmentation ; Out-of-distribution Detection
DDC:
004 Data processing & computer science (computed)
Relations:
URL:
https://hdl.handle.net/1853/73133
Content Provider:
Georgia Institute of Technology: SMARTech - Scholarly Materials and Research at Georgia Tech  Flag of United States of America