Towards longitudinal mapping of transportation infrastructure with spatio-temporal generative modeling
Abstract/Contents
- Abstract
- This work proposes a super-resolution pipeline to generate high resolution satellite imagery for a preferred point in time. We use a spatio-temporal generative model to perform super-resolution of low spatial resolution (LR) satellite imagery from the preferred time, conditioning on high resolution imagery (HR) from another time to generate HR satellite imagery for the preferred time. To train the spatio-temporal generative model, we curated the largest-ever super-resolution dataset of satellite imagery at sub-meter resolution, consisting of over 96, 000 image triplets. We present a proof-of-concept of this approach by evaluating the performance of our generated imagery on a sidewalk segmentation task. Our results demonstrate that our pipeline generates high spatial resolution images that are structurally similar to the original high resolution satellite imagery while accurately reproducing the style and content of the input low resolution imagery. Furthermore, our generated imagery outperforms LR imagery in the sidewalk segmentation task, providing compelling evidence for the utility and effectiveness of our approach. Since LR satellite imagery is available longitudinally at large geographical scale, researchers can use our pipeline to generate HR satellite imagery for their preferred time and location. Future work could leverage generated longitudinally available HR satellite imagery to longitudinally map transportation networks to inform and evaluate transportation planning.
Description
Type of resource | text |
---|---|
Date created | June 8, 2023 |
Publication date | June 10, 2023 |
Creators/Contributors
Author | Singla, Ayush | |
---|---|---|
Thesis advisor | Mitchell, John |
![]() |
Thesis advisor | Ng, Andrew |
Subjects
Subject | Super-Resolution |
---|---|
Subject | Remote-sensing images |
Subject | Transportation > Planning |
Subject | Generative Adversarial Networks (GANs) |
Subject | Climate change mitigation |
Subject | Artificial intelligence > Industrial applications |
Subject | Machine learning > Industrial applications |
Genre | Text |
Genre | Thesis |
Bibliographic information
Access conditions
- Use and reproduction
- User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 4.0 International license (CC BY-NC).
Preferred citation
- Preferred citation
- Singla, A. (2023). Towards longitudinal mapping of transportation infrastructure with spatio-temporal generative modeling. Stanford Digital Repository. Available at https://purl.stanford.edu/py677ft1204. https://doi.org/10.25740/py677ft1204.
Collection
Undergraduate Theses, School of Engineering
View other items in this collection in SearchWorksContact information
- Contact
- ayushsingla@stanford.edu
Also listed in
Loading usage metrics...