TextVQA eval split — val or test?

Hi, thanks for the great work. Could you briefly clarify whether the TextVQA results in the paper are measured on the validation set or the official test set (submitted to EvalAI)? This would help a lot for fair comparison. Thanks!

Check the paper's TextVQA results and any repository documentation that describes the evaluation split. Done means confirming whether the reported results use the validation set or the official EvalAI-submitted test set.

Written by the indexing model from the issue text.