Ten Architectures, One Error: Shared Failure Modes in Hyperspectral Classification under Spatially Disjoint Evaluation
A leakage-free evaluation protocol linking spatial separation to the model's receptive field is introduced, which shows that Macro-F1 drops by 0.147 on average and model rankings change by as many as five places, and reveals that all ten architectures misclassify largely the same pixels.
Ehsan Faghih, Fatemeh Ashrafi, Marguerite Moore et al.
· 0 citations