In order to train the joint contrastive representation learning module, we constructe a large Text Annotated Distortion, Appearance and Content (TADAC) image database.