OpenSR · Spectrally faithful super-resolution
LR vs SR Detections
OpenSR uses a physics-constrained latent diffusion model to sharpen Sentinel-2 imagery while preserving
spectral signatures that drive environmental analytics. The maps below show how super-resolution
changes the detection of flood water (Valencia 2024) and burn scars (Palisades 2025) versus the
original low-resolution inputs. Explore the source at
github.com/ESAOpenSR/spectral_usecases.
Key takeaway
Spectrally constrained SR reduced false negatives/positives in water and burn-severity indices,
showing that fidelity to spectra matters more than visual sharpness for operational RS pipelines.
What to explore
Toggle SR and LR overlays, inspect spectral threshold outputs, and compare how boundaries tighten
around the red AOI outline in each scenario.