SCANS Data Set - A Reference Data Set for Handwritten Text Detection

# SCANS Data Set

A scientific paper with handwritten annotations as reference data set for handwritten text detection.

The paper is

Sharon Fogel, Hadar Averbuch-Elor, Sarel Cohen, Shai Mazor, Roee Litman; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti
on (CVPR), 2020, pp. 4324-4333

The annotations were made by two people with different pens. The text is randomly taken from "The Fellowship Of The Ring" by JRR Tolkien.

The documents in folder `images` were scanned with a Canon Pixima TR4550 with 300DPI. The `labels-` folders contain the labels for paragraph, line, and word level in following JSON format:

{
"image_file": "paper-0001.png", # The corresponding image in folder `images`
"shape": [ # The size of the image
3495,
2473
],
"properties": [ # A list of all line bounding boxes (for paragraphs, lines, or words)
{
"type": "HWL", # HWA: handwritten paragraph, HWL: handwritten line, HWW: handwritten word
"id": "18",
"points": [ # The coordinates of the bounding boxes
[
1243.87096093748,
1515.21674083916
],
[
2149.01588303892,
1515.21674083916
],
...
]
},
...
]
}