OCRBench Text Recognition Benchmark Dataset

OCRBench is a text recognition benchmark dataset released by Huazhong University of Science and Technology and Microsoft Research. This dataset is an evaluation benchmark for multimodal large-scale optical character recognition (OCR). The relevant paper results are:OCRBench: On the Hidden Mystery of OCR in Large Multimodal Models", which aims to evaluate the OCR capabilities of large multimodal models (LMMs) in different text-related tasks.

The dataset contains 1000 manually screened and corrected question-answer pairs from five representative text-related tasks: text recognition, scene text centering, document orientation, key information extraction (KIE), and handwritten mathematical expressions (HMER).

The data includes:

• Text recognition 300 images (including regular, irregular, artistic and other text types).

• Scene Text Centric Visual Question Answering 200 Questions.

• Document-guided visual question answering 200 questions.

• 200 questions for key information extraction.

• Handwritten mathematical expression recognition 100 images from the HME100k dataset.