RILS: Masked Visual Reconstruction in Language Semantic Space | Read Paper on Bytez