README
------

This zip contains DIGITS dataset and USPS dataset in
MNIST-compatible format as used in the following paper. If you use
these datasets please cite it.

Alexander K. Seewald, "On the Brittleness of Handwritten Digit
Recognition Models," ISRN Machine Vision, vol. 2012, Article ID
834127, 10 pages, 2012. doi:10.5402/2012/834127.

All data sets contain one .data file and one .label file in MNIST
format (similar to *-labels-idx1-ubyte and *-images-idx3-ubyte from
MNIST). For simplicity we only refer to the filename without
extension.

all-digits-MNIST-gammaC
  All DIGITS digit (train+test), gamma-corrected to fit MNIST data
all-digits-MNIST-gammaC-sh
  gamma-corrected and shifted to center-of-gravity
all-digits-MNIST-gammaC-sh-asp
  gamma-corrected, shifted to c-o-g, and aspect ratio changed to fit MNIST

zip.train-MNIST
  USPS train split, gamma-corrected and shifted to center-of-gravity
zip.test-MNIST
  USPS test split,  gamma-corrected and shifted to center-of-gravity
zip.test-MNIST-l2t
  USPS test split, gamma-corrected, shifted to c-o-g and thinned to
  binary image... worse performance so no training data was generated.

For any questions please contact me, the main author of above paper.

Best,
  Alex
--
Dr. Alexander K. Seewald

Seewald Solutions
www.seewald.at
Tel. +43 7243 21520
Mob. +43 664 1106886
Fax. +43 1 2533033 2764



