Benchmark RadarHyperImageNet
HyperImageNet is a dataset of 26,084 airborne hyperspectral image patches with 224 spectral bands and 138 fine-grained land-cover categories, providing raw imagery, pixel-level semantic labels, and object-level instance masks for semantic and instance segmentation, along with an open-environment evaluation protocol.
- Released
- 2026-07-23
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing hyperspectral benchmarks lack fine-grained categories and instance-level annotations; HyperImageNet enables evaluation of models on high-spatial-resolution imagery with strict spatial separation for open-environment generalization.
Motivation
We present HyperImageNet, a large-scale benchmark for fine-grained hyperspectral land-cover understanding.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.