ICDAR_2019_ArT Dataset

ICDAR 2019 ArT is an arbitrary shape text dataset. The dataset includes 10,176 images, of which 5603 images are training sets and 4573 images are test sets.

The images in this dataset come from the TotalText, CTW1500 and LSVT datasets. The parts containing Chinese text come from Baidu Street View and are annotated in a crowdsourcing manner.