Georgia Institute of Technology
Domain Generalization for Semantic Segmentation of Aerial Imagery Datasets
Abstract
dc:description.abstractSince curating large datasets with dense annotations from UAVs under diverse weather, daytime and layout conditions in the real-world is restrictive due to several prohibitive constraints (sampling images from same locations under different conditions, annotation cost, etc.), we propose using SKYSCAPES to enhance the visual recognition capabilities of UAV systems. Specifically, SKYSCAPES enables training visual recognition models for aerial images that can be used to systematically study (i) whether models trained on synthetic data can generalize to real-world settings, (ii) can synthetic data serve as additional (augmented) training data for real-world settings and (ii) do trained models generalize across different factors of variation such as weather, daytime, layouts. We plan to publicly release both the annotated dataset and our code to curate \csim images and train models.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Deepanshi, Deepanshi
- Advisors dc:contributor.advisor
-
- Hoffman, Judy
- Muthukumar, Vidya
- Committee members dc:contributor.committeemember
-
- Anderson, David
- Davenport, Mark
Subjects
dc:subject × 2Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/75147