Closed alice-cool closed 2 years ago
Dear @alice-cool,
Thank you for your interest in our paper 😃
The score difference is explained by the fact that the two tables are not using the same GQA split. The scores in our Table 2 have been computed on the GQA validation split, while scores in Table 3 of LXMERT have been computed on the GQA testdev split.
Dear scholar, I am glad to read your paper. But I have questions about the result you showed in table 2 in the paper.
I think the result is too high about 60%, maybe from some datasets
In Lxmert, I found the BUTD method only gets at most 52%