Skip to content

Latest commit

 

History

History
37 lines (26 loc) · 1.14 KB

cocotext.md

File metadata and controls

37 lines (26 loc) · 1.14 KB

COCO-Text Dataset

Data Downloading

Official Website

The COCO-Text images dataset and the annotations (in JSON format) annotations v1.4 JSON can be downloaded from this link.

Note: Please register an account to download this dataset.

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

COCO-Text
  |--- train_images
  |    |--- COCO_train2014_000000000036.jpg
  |    |--- COCO_train2014_000000000064.jpg
  |    |--- ...
  |--- COCO_Text.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 cocotext --task det \
    --image_dir path/to/COCO-Text/train_images/ \
    --label_dir path/to/COCO-Text/COCO_Text.json \
    --output_path path/to/COCO-Text/det_gt.txt

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

Back to dataset converters