Skip to content

Latest commit

 

History

History
38 lines (27 loc) · 1.22 KB

sroie.md

File metadata and controls

38 lines (27 loc) · 1.22 KB
Raw

SROIE Dataset

Data Downloading

The SROIE dataset Official Website | Download Link

Note: Please register an account to download this dataset.

This dataset is divided into 3 tasks: (1) Text Localisation, (2) OCR, and (3) Key Information Extraction. For now, we consider and download the updated dataset only for Task 1.

After downloading and unzipping the dataset as above and renaming the extracted folder as appropriate e.g. train, the directory structure should be like as follows (ignoring the archive files):

SROIE
  |--- train
  |    |--- <image_name>.jpg
  |    |--- <image_name>.txt
  |    |--- <image_name>.jpg
  |    |--- <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 sroie --task det \
    --image_dir path/to/SROIE/train/ \
    --label_dir path/to/SROIE/train \
    --output_path path/to/SROIE/det_gt.txt

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

Back to dataset converters