Enhancing Compositional Reasoning in CLIP via Reconstruction and Alignment of Text Descriptions | Read Paper on Bytez