Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View | Read Paper on Bytez