Fetching the paper…

ImageBERT: Cross-modal Pre-training with Large-scale Weak-supervised Image-Text Data · Around