Efficient Vision-Language Pre-training by Cluster Masking

Benchmark Model Rank Results
image-classification-on-imagenetOurs-RGB0.3–Top 1 Accuracy: 63.1%
zero-shot-cross-modal-retrieval-on-coco-2014Ours-Embedding–Image-to-text R@1: 23.77
zero-shot-cross-modal-retrieval-on-flickr30kOurs-Embedding–Text-to-image R@1: 57.0