The TRUE Benchmark is introduced in the paper "A Status Check on Current Vision-Language Models in Text Recognition and Understanding".
There are 4 splits:
full: The complete dataset for the TRUE Benchmark, consisting of our newly collected data.
hard: A challenging subset of the TRUE Benchmark.
textvqa_edited: An edited subset of images sourced from TextVQA
docvqa_edited: An edited subset of images sourced from DocVQA
More details are available on our project Homepage
