Skip to content

Handwritten digits, a bit like the MNIST dataset.

License

Notifications You must be signed in to change notification settings

Godlovecui/numbers

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

63 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Numbers

This repository tries to create a repository of handwritten digits, much like the MNIST database of handwritten digits. In Switzerland, the handwritten digites sometimes look a bit different, which is why we undertake this effort.

There are two data sets in this repository. They are described below.

Warning: If you're using git on Windows to clone this repository, it make take a very long time because there are so many (tiny) files in it!

Set 1: manual collection

Our goal was 10,000 handwritten digits and we have met that goal! If you want to help us reach 20,000 handwritten digits, check out the tools directory.

The directory name of every contribution adheres to the following naming scheme:

  1. four digits to identify the person
  2. an underscore character
  3. two letter country code (ISO 3166 Alpha-2 codes) or XX if unknown (e.g. Switzerland is CH)
  4. age, rounded to the nearest decade (e.g. 35 to 44 years is 4) or X if unknown
  5. sex (M for man, F for woman, X for unknown, O for other)

Set 2: automatic collection

As part of a commercial project a neural network was trained on the numbers cut from a very large collection of documents. There are so many digits in this set that the data quality is lower than in the manual collection. You can find this set in the UNCATEGORIZED directory. A bit over 800,000 digits!

  • digits may be miscategorized or malformed (e.g. 554)
  • digits are both handwritten and printed (e.g. 552)
  • the distribution of digits is not uniform
  • no information is available about the authors

I'll be happy to take pull requests which fix miscategorized or malformed digits.

As for the distribution of numbers:

digit  files
  0   323945
  1    94075
  2    71820
  3    55240
  4    46143
  5    96376
  6    39868
  7    34836
  8    37124
  9    34571

About

Handwritten digits, a bit like the MNIST dataset.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published