Conference item
Cloud-Stereo: a dataset and benchmark for reconstructing atmospheric clouds from stereo images
- Abstract:
-
Obtaining accurate measurements of clouds is a critical problem in atmospheric physics, as accurate modeling of cloud properties allows us to better understand and predict climate change. Stereo camera networks have shown promise in obtaining such measurements, being able to reconstruct detailed cloud fields over multi-km2 domains. However, previous studies on cloud stereo depth estimation have been limited to using traditional (non-learned) matching techniques, due to the absence of suitable training datasets for this challenging domain. In this work, we present a novel dataset (CloudStereo) specifically tailored for cloud depth estimation. The Cloud-Stereo dataset includes: 1) a synthetic dataset for training, comprising 3000 stereo pairs and simulated dense LiDAR depth data, and 2) a high-accuracy real-world dataset consisting of ≈ 120k frames acquired from a stereo camera and Doppler Aerosol LiDAR for testing. Using our dataset we benchmark existing learning and non-learning based stereo depth estimation approaches, and demonstrate that fine-tuning on our dataset can lead to significant accuracy improvement for learned methods. We believe this dataset will enable the training of future, more accurate, methods for cloud field reconstruction, enhancing a unique measurement capability for developing and evaluating atmospheric models. The dataset is available at https://cloud-stereo.jacob-lin.com/.
- Publication status:
- Published
- Peer review status:
- Peer reviewed
Actions
Access Document
- Files:
-
-
(Preview, Version of record, pdf, 17.8MB, Terms of use)
-
- Publication website:
- https://bmvc2025.bmva.org/proceedings/835/
Authors
- Funder identifier:
- https://ror.org/02b5d8509
- Grant:
- NE/X012255/1
- Funder identifier:
- https://ror.org/0439y7842
- Grant:
- 2922572
- Publisher:
- British Machine Vision Association
- Host title:
- Proceedings of the 36th British Machine Vision Conference (BMVC 2025)
- Article number:
- 835
- Publication date:
- 2025-11-24
- Acceptance date:
- 2025-07-18
- Event title:
- 36th British Machine Vision Conference (BMVC 2025)
- Event location:
- Sheffield, UK
- Event website:
- https://bmvc2025.bmva.org/
- Event start date:
- 2025-11-24
- Event end date:
- 2025-11-27
- Language:
-
English
- Pubs id:
-
2287106
- Local pid:
-
pubs:2287106
- Deposit date:
-
2025-09-09
- ARK identifier:
Terms of use
- Copyright holder:
- Lin et al.
- Copyright date:
- 2025
- Rights statement:
- © 2025. The copyright of this document resides with its authors. It may be distributed unchanged freely in print or electronic forms.
- Notes:
- The author accepted manuscript (AAM) of this paper has been made available under the University of Oxford's Open Access Publications Policy, and a CC BY public copyright licence has been applied.
- Licence:
- CC Attribution (CC BY)
If you are the owner of this record, you can report an update to it here: Report update to this record