Oscar
Oscar copied to clipboard
why dose the pretrain corpus have 100w coco_qa pairs?
This is inconsistent with the number reported in the paper
VQA2.0 dataset use COCO iamges
VQA2.0 dataset use COCO iamges
I know, but VQA2.0 dataset only have 66w qa pairs (train+val). Where did the extra 40w qa pairs come from?