Removing Word-Level Spurious Alignment between Images and Pseudo-Captions in Unsupervised Image Captioning

42498

Knowledge Triples