Skip to content

Born-Digital Images Dataset

Data Downloading

The Born-Digital dataset Official Website | Download Link

Note: Please register an account to download this dataset.

This dataset is divided into 4 tasks: (1.1) Text Localization, (1.2) Text Segmentation, (1.3) Word Recognition, and (1.4) End To End. For now, we consider and download only the dataset for Task 1.1.

After downloading the images and annotations, unzip the files and rename as appropriate e.g. train_images for the images and train_labels for the ground truths, after which the directory structure should be like as follows (ignoring the archive files): 

Born-Digital
  |--- train_images
  |    |--- <image_name>.jpg
  |    |--- <image_name>.jpg
  |    |--- ...
  |--- train_labels
  |    |--- <image_name>.txt
  |    |--- <image_name>.txt
  |    |--- ...

Data Preparation

For Detection Task

To prepare the data for text detection, you can run the following commands:

python tools/dataset_converters/convert.py \
    --dataset_name borndigital --task det \
    --image_dir path/to/Born-Digital/train_images/ \
    --label_dir path/to/Born-Digital/train_labels \
    --output_path path/to/Born-Digital/det_gt.txt

The generated standard annotation file det_gt.txt will now be placed under the folder Born-Digital/.

Back to dataset converters