We randomly crop word images from 1000 document page images to create the Akshara-Assamese dataset. The pages are taken from multiple books and scanned using a flatbed scanner. The pages are scanned in 300 DPI. We manually annotate ground truth transcriptions of cropped word images. This dataset consists of 99,775 word images, and their corresponding ground truth transcriptions. We divide this dataset into Training, Validation, and Test Sets consisting of 79,697, 9,932, and 10,146 word images and their corresponding ground truth transcriptions. There are 20,133 unique Assamese words in the training set.

Training Set:

train.zip contains folder named “images” with 79,697 word level images, “train_gt.txt” file containing image name and ground truth transcription separated by “Tab space”, and “vocabulary.txt” contains list of 20,133 unique words in the Training set.

Validation Set:

val.zip contains folder named “images” with 9,932 word level images, and “val_gt.txt” containing image name and ground truth text separated by “Tab space”.

Test Set:

test.zip contains folder named “images” with 10,146 word level images, and “test_gt.txt” containing image name and ground truth text separated by “Tab space”.