Reproduce “SpecUltra: A Unified Foundation Model for Heterogeneous Spectral Images” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weili Kong, Baisen Liu, Xiaojun Bi, Jiaming Pei
SpecUltra represents wavelength and spatial position explicitly in spectral image tokens, using 3D convolution and wavelength-aware positional encoding. It pretrains with masked image modeling and adapts HSI-trained input filters to MSI through learnable combinations of pretrained kernel slices. On SegMunich, SpecUltra-B reaches 50.90% mean intersection over union, versus SatMAE’s 50.21%. This supports transfer on one sparse-band MSI task, not universal transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.