metadata

license: apache-2.0
task_categories:
  - image-to-text
tags:
  - chess
  - HCS
  - handwriting
  - scoresheets
  - forms
size_categories:
  - 10K<n<100K
dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype: string
  splits:
    - name: train
      num_bytes: 517848025.198
      num_examples: 13731
  download_size: 642552652
  dataset_size: 517848025.198
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Disclaimer

This is not data that I created. It originally came from the Paper Digitization of Handwritten Chess Scoresheets with a BiLSTM Network

You can also find the dataset here and here

Datasets

There are 2 versions of this dataset

processed_hcs Dataset where you are right now
unprocessed_hcs where the whole scoresheet can be seen here

Desciption

The Handwritten Chess Scoresheet Datase (HCS) contains a set of single and double paged chess scoresheet images with ground truth labels. The processed images are a list of the extracted move boxes with the corresponding labels. The labels are strings of the chess moves shown in the move boxes.

Usage

In Python script

import datasets
unprocessed_hcs = datasets.load_dataset("Benjaminkost/unprocessed_hcs")

# Show first image
first_image = unprocessed_hcs["train"][0]["image"]
first_label = unprocessed_hcs["train"][0]["labels"]

first_image.show()
print(first_label)