AI RESEARCH

Perspective-Equivariant Fine-tuning for Multispectral Demosaicing without Ground Truth

arXiv CS.CV

ArXi:2603.01332v2 Announce Type: replace Multispectral saicing is crucial to reconstruct full-resolution spectral images from snapshot mosaiced measurements, enabling real-time imaging from neurosurgery to autonomous driving. Classical methods are blurry, while supervised learning requires costly ground truth (GT) obtained from slow line-scanning systems. We propose Perspective-Equivariant Fine-tuning for saicing (PEFD), a framework that learns multispectral saicing from mosaiced measurements alone.