Low accuracy (~50%) with SSL (BYOL/MAE/VICReg) on hyperspectral crop stress data — what am I missing? [R]

The content details a persistent problem with achieving low accuracy (~50%) using self-supervised learning methods like BYOL, MAE, and VICReg for hyperspectral crop stress detection. Despite trying various techniques, performance remains barely better than random for three classes, leading to suspicions about data separability or SSL method suitability.