Skip to content

Latest commit

 

History

History
35 lines (25 loc) · 1017 Bytes

textocr.md

File metadata and controls

35 lines (25 loc) · 1017 Bytes
Raw

TextOCR Dataset

Data Downloading

The TextOCR dataset Official Website | Download Link

After downloading the images and annotations, unzip the files, after which the directory structure should be like as follows (ignoring the archive files):

TextOCR
  |--- train_val_images
  |    |--- <image_name>.jpg
  |    |--- <image_name>.jpg
  |    |--- ...
  |--- TextOCR_0.1_train.json
  |--- TextOCR_0.1_val.json

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 textocr --task det \
    --image_dir path/to/TextOCR/train_val_images/ \
    --label_dir path/to/TextOCR/TextOCR_0.1_train.json \
    --output_path path/to/TextOCR/det_gt.txt

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

Back to dataset converters