A framework pretrained on authentic binary occlusion masks uses guided sampling and intersection-based partitioning to train diffusion models on incomplete physical observations without zero-query regions.
Generative data assimilation for surface ocean state estimation from multi-modal satellite observations.Journal of Advances in Modeling Earth Systems, 17(8):e2025MS005063, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Observation-Aligned Mask Priors for Learning Physical Dynamics from Authentic Occlusions
A framework pretrained on authentic binary occlusion masks uses guided sampling and intersection-based partitioning to train diffusion models on incomplete physical observations without zero-query regions.