COCO-Text

by No License

COCO-Text

A Large-Scale Scene Text Dataset, Based on MSCOCO. COCO-Text V2.0 contains 63,686 images with 239,506 annotated text instances. Segmentation mask is annotated for every word, allowing fine-level detection. Three attributes are labeled for every word: machine-printed vs. handwritten, legible vs. illgible, and English vs. non-English.

Dataset Attributes