COCO-Text Dataset

Estimated reading: 3 minutes

Visualization of the COCO-Text Train Dataset in the Deep Lake UI

COCO-Text dataset

What is COCO-Text Dataset?

The COCO-Text (Common Objects in Context – Text) Dataset objective is to solve scene text detection and recognition using the largest scene text dataset. The dataset was created using real-scene imagery. The dataset is structured around three tasks End-To-End Recognition, Cropped Word Recognition, and Text Localization.

The dataset was created using images of complex everyday scenes and contains various other images which were collected by not keeping in mind text tasks, this has resulted in presence of the vast and diverse amount of text instances in these images.

The dataset was annotated with a location in terms of the bounding box, fine-grained classification into machine-printed text and handwritten text, classification into legible and illegible text, the script of the text, and transcriptions of legible text. Currently, the dataset contains a total of 173,589 labeled text regions in over 63,686 images.

Download COCO-Text Dataset in Python

Instead of downloading the COCO-Text dataset in Python, you can effortlessly load it in Python via our Deep Lake open-source with just one line of code.

Load COCO-Text Train Subset Dataset in Python

				
					import deeplake
ds = deeplake.load("hub://activeloop/coco-text-train")

Load COCO-Text Test Subset Dataset in Python

				
					import deeplake
ds = deeplake.load("hub://activeloop/coco-text-test")

COCO-Text Dataset Structure

COCO-Text Data Fields

images: tensor containing images of the dataset.
masks: tensor containing masks of the respective image
boxes: tensor containing bounding boxes of the respective image
languages: tensor containing language of the text
legibilities: tensor containing legibility of the text
classes: tensor containing a class of the text
utf8_strings: tensor containing the text of the image
areas: tensor containing the area of the text region
images_meta: tensor containing image metadata

How to use COCO-Text Dataset with PyTorch and TensorFlow in Python

Train a model on the COCO-Text dataset with PyTorch in Python

Let’s use Deep Lake built-in PyTorch one-line dataloader to connect the data to the compute:

				
					dataloader = ds.pytorch(num_workers=0, batch_size=4, shuffle=False)

Train a model on the COCO-Text dataset with TensorFlow in Python

				
					dataloader = ds.tensorflow()

Additional Information about COCO-Text Dataset

COCO-Text Dataset Description

Homepage: https://rrc.cvc.uab.es/?ch=5&com=introduction
Repository: https://github.com/bgshih/coco-text
Paper: Veit, A., Matera, T., Neumann, L., Matas, J., & Belongie, S. (2016). Coco-text: Dataset and benchmark for text detection and recognition in natural images. arXiv preprint arXiv:1601.07140.

COCO-Text Dataset Curators

Veit, A., Matera, T., Neumann, L., Matas, J., & Belongie

COCO-Text Dataset Licensing Information

The annotations in this dataset belong to the SE(3) Computer Vision Group at Cornell Tech and are licensed under a Creative Commons Attribution 4.0 License.

COCO-Text Dataset Citation Information

				
					@article{veit2016coco,
  title={Coco-text: Dataset and benchmark for text detection and recognition in natural images},
  author={Veit, Andreas and Matera, Tomas and Neumann, Lukas and Matas, Jiri and Belongie, Serge},
  journal={arXiv preprint arXiv:1601.07140},
  year={2016}
}