Observation-Aligned Mask Priors for Learning Physical Dynamics from Authentic Occlusions

2026-05-16

AI Summary

This work addresses the challenge of incomplete observations caused by real-world occlusions, whose missing patterns are complex and non-random, rendering conventional masking strategies ineffective. The authors propose a learning framework that leverages a prior derived from real observation masks: a pretrained Bayesian flow network models occlusion topology, while a globally normalized cross-entropy loss guides the generation of sample-specific masks. An intersection-driven context-query partitioning mechanism is innovatively designed to guarantee that every valid observed dimension has a positive probability of being queried, thereby preventing local generative collapse. Evaluated on three oceanic datasets featuring realistic satellite occlusions (up to 256×256 resolution), the method significantly outperforms strong diffusion-based baselines in both MSE and PSNR, marking the first successful integration of mask priors learned directly from real occlusions to enhance physical dynamics reconstruction.

Abstract

Learning physical dynamics directly from incomplete observations is challenging because authentic occlusions are structured, sample-dependent, and often missing not at random, whereas existing methods typically rely on heuristic masking rules or predefined mask distributions. We propose Observation-Aligned Mask Priors, a framework that learns the distribution of authentic observation masks and uses it to construct context-query partitions for training from incomplete data. Specifically, we pretrain a Bayesian Flow Network (BFN) on binary observation masks to capture real occlusion topologies, then guide BFN sampling with a globally normalized cross-entropy objective to generate sample-specific masks aligned with each sparse observation. The intersection between the guided mask and the observed mask defines the context, and the remaining observed entries become query targets for a diffusion-based reconstruction model. We show that this intersection-based partitioning gives every valid observed dimension a strictly positive probability of being queried, preventing zero-query dead zones and local generative collapse. Experiments on three real-world oceanographic datasets with authentic satellite occlusions, across resolutions up to 256$\times$256, show consistent improvements over strong diffusion baselines in MSE and PSNR. These results demonstrate that learning mask priors from authentic occlusions is an effective alternative to heuristic masking for learning from incomplete physical observations without access to fully observed fields.

Problem

• physical dynamics
• authentic occlusions
• incomplete observations
• mask priors
• missing not at random

Innovation

• Observation-Aligned Mask Priors
• Bayesian Flow Network
• authentic occlusions
• diffusion-based reconstruction
• context-query partitioning