Georgia Institute of Technology
Corroborative labeling and localization for acquired data: a case study with fine-grained vehicle classification
Abstract
dc:description.abstractDeep learning models get trained on the data set provided during training. However, these models get stale and reduce accuracy as the data set updates. With the current increase in the amount of unstructured data, such as images and videos, it is essential for deep learning models to retrain to adapt to new data coming in. Retraining models is crucial in various domains such as medical science, social media, or traffic monitoring system. In this thesis, we present CLAD-GT(Corroboration and Localization of Acquired Data), a system developed to perform two tasks, (i)reduce the cost of retraining models with newly acquired data and (ii) integrate the knowledge of multiple training distributions. We then evaluate CLAD-GT on the vehicle classification task for the traffic management system, specifically traffic management tasks, such as object attributes labeling, like, make, model, and color. We also evaluate CLAD-GT on vehicle re-identification with incremental training data. We further conclude with CLAD-GT’s limitations and discuss further improvements of CLAD-GT approach.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Computer Science
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Vaidya, Sanjyot
- Advisor dc:contributor.advisor
-
- Pu, Calton
- Committee members dc:contributor.committeemember
-
- Liu, Ling
- Xu, Jim
Subjects
dc:subject × 3Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/72017