The Hyper-Object Challenge: Inferring Beyond The Visible In Low-Cost Hyperspectral Imaging

This paper presents the ICASSP 2026 Hyper-Object Challenge for reconstructing high-resolution hyperspectral reflectance images from low-cost sensing inputs. The challenge targets a diverse set of everyday objects with highly variable material properties and spans both the visible and near-infrared spectrum (400–1000 nm). Two complementary tracks are defined: spectral reconstruction from mosaic measurements and joint spatial–spectral super-resolution from low-resolution RGB images. A curated hyperspectral dataset and a standardized Spectral–Spatial–Color (SSC) evaluation metric are provided. Strong participation and benchmark results reveal current capabilities and open challenges in practical, low-cost hyperspectral imaging.