Skip to content

CTW Dataset

Data Downloading

The CTW images dataset Official Website | Download Link

Note: Please fill in a form to download this dataset.

The images are in 26 batches, i.e. 26 different .tar archived files of the format images-trainval/ctw-trainval*.tar. All 26 batches need to be downloaded.

The CTW annotations (in JSON Lines format i.e. .jsonl) can be downloaded from this download link. The annotations archived file is named ctw-annotations.tar.gz.

After downloading the zipped images, unzip the batches and collect all the images into a single folder e.g. train_val/. After downloading the zipped annotations, unzip them. Finally, the directory structure should look like this (ignoring the archive files):

CTW
  |--- train_val
  |    |--- 0000172.jpg
  |    |--- 0000174.jpg
  |    |--- ...
  |--- train.jsonl
  |--- val.jsonl
  |--- test_cls.jsonl
  |--- info.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 ctw --task det \
    --image_dir path/to/CTW/train_val/ \
    --label_dir path/to/CTW/train.jsonl \
    --output_path path/to/CTW/det_gt.txt

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

Note that the label_dir flag may be altered to prepare the validation data.

Back to dataset converters