README.md

COCO-Text-Patch

COCO-Text-Patch is the first text verification data set created to encourage researchers to use machine learning techniques for text verification which will in turn enhance the whole end-to-end text detection and recognition process.
The data set is derived from COCO-Text and contains approximately 354 thousands images labeled as text and non-text.

Using the data set

If you are going to use this data set please cite the paper

@Inbook{Ibrahim2016,
     author="Ibrahim, Ahmed and Abbott, A. Lynn and Hussein, Mohamed E.",
     title="An Image Dataset of Text Patches in Everyday Scenes",
     bookTitle="Advances in Visual Computing: 12th International Symposium, ISVC 2016, Las Vegas, NV, USA, December 12-14, 2016, Proceedings, Part II",
     year="2016",
     publisher="Springer International Publishing",
     address="Cham",
     pages="291--300",
     isbn="978-3-319-50832-0",
     doi="10.1007/978-3-319-50832-0_28",
     url="http://dx.doi.org/10.1007/978-3-319-50832-0_28"
}

Creating the data set

to create the data set you have to:-

-In the COCOAPI folder, unzip the COCO_Text.json.gz file

-Download the COCO 2014 training data set from http://msvocds.blob.core.windows.net/coco2014/train2014.zip

-Make sure you have Python 2.7 installed with numpy, skimage, math, Image, HD5 packages

-Clone the Git repository

-From Python shell run createdataset.py

supplied models

the folder "models" contains model description files ready to be used in caffe to train the models descibed in our paper "Text Patches Data Set for Text verification"

the folder contains model desciptions as well as suggested solver files.

you need to have Caffe installed from http://caffe.berkeleyvision.org/

for any inquiries please contact me on nady at vt.edu