ICDAR 2019-LSVT (Large-scale Street View Text) is a large-scale scene text dataset. The dataset includes 450,000 images with text, all taken from real streets (such as storefronts, landmarks, etc.), of which 400,000 images are partially annotated as training data; 50,000 images (30,000 training images + 20,000 test images) are fully annotated.