Sudoku Dataset

Sudoku Dataset Overview

Dataset Content

• Image Source: Sudoku images captured from newspapers using smartphone cameras.
• Number of Images: 200 total.
• Dataset Split: 160 training images and 40 testing images.

Dataset Versions

• V2: 200 images (160 for training, 40 for testing).
• mixed: Same images as V2, but each puzzle is manually solved.
• V1: Older version with 160 images, no longer recommended.

Download Options

• V2:
  • Training set: v2_training.tar.bz2
  • Test set: v2_test.tar.bz2
• mixed:
  • Training set: v2_mixed_training.tar.bz2
  • Test set: v2_mixed_test.tar.bz2

Citation

• Reference:
  • Wicht, Baptiste; Hennebert, Jean, "Camera‑based Sudoku recognition with deep belief network" Soft Computing and Pattern Recognition (SoCPaR), 2014 6th International Conference, vol., no., pp.83‑88, 11‑14 Aug. 2014
  • Wicht, Baptiste, and Jean Hennebert, "Mixed handwritten and printed digit recognition in Sudoku with Convolutional Deep Belief Network." Document Analysis and Recognition (ICDAR), 2015 13th International Conference on. IEEE, 2015.

Dataset Format

• File Structure: each imageX.jpg has a corresponding imageX.dat containing metadata.
• Metadata: includes phone brand/model, image format, and Sudoku puzzle description.

Contact

• Author: Baptiste Wicht
• Email: baptiste.wicht@gmail.com