5000-Images-of-Turkish-Natural-Scene-OCR-Data

Description

5,000 Turkish natural scenarios OCR data include a variety of natural scenarios and multiple shooting angles. For annotation, quadrilateral or polygon bounding box annotation and transcription for the texts were annotated in the data. This data can be used for tasks such as the Turkish language OCR.

For more details, please refer to the link: https://www.nexdata.ai/datasets/ocr/1697?source=Github

Specifications

Data size

5,000 images, 40,793 bounding boxes

Collecting environment

including slogan, receipt, poster, warning sign, road sign, book, cover, station sign and signboard

Data diversity

including a variety of natural scenes, multiple shooting angles

Device

cellphone

Photographic angle

looking up angle, looking down angle, eye-level angle

Data format

the image format is common format such as.jpg, the annotation file format is .json

Annotation content

line-level quadrilateral bounding box annotation and transcription for the texts；polygon bounding box annotation and transcription for the texts

Accuracy rate

the error bound of each vertex of quadrilateral or polygon bounding box is within 5 pixels, which is a

Licensing Information

Commercial License