CroPe: Cross-Modal Semantic Compensation Adaptation for All Adverse Scene Understanding | Read Paper on Bytez