Skip to content

Latest commit

 

History

History
44 lines (28 loc) · 1.73 KB

README.md

File metadata and controls

44 lines (28 loc) · 1.73 KB
Raw

MNIST Handwritten Digits as JSON

All data is from http://yann.lecun.com/exdb/mnist/.

There are 70,000 examples in total in the datasets:

  • 60,000 training examples in mnist_handwritten_train.json
  • 10,000 test examples in mnist_handwritten_test.json

The data is in a JSON format. Each JSON file contains an array of objects, where each object has two fields:

  • image: a 784-element array of numbers [0,255], representing pixel values for each 28x28 handwritten image. 0 means background (white) and 255 means foreground (black).
  • label: the label for this character

There are two JSON files; one is training data and the other is test data. These two sets were provided from the original MNIST dataset.

File sizes

2.8M  mnist_handwritten_test.json.gz
17M   mnist_handwritten_train.json.gz

I strongly suggest you download the gzipped files, as the non-gzipped files are very large

Download / Decompress

To download the files, something like this should work:

curl -LO https://github.com/lorenmh/mnist_handwritten_json/raw/master/mnist_handwritten_test.json.gz
curl -LO https://github.com/lorenmh/mnist_handwritten_json/raw/master/mnist_handwritten_train.json.gz

To decompress:

gunzip *.gz

If you clone you will need git-lfs.

To run

You'll need the process.sh and convert_to_json.py files in some directory. The process.sh script will create a directory data/ in the same directory as the script, and will call convert_to_json.py to convert the data from the IDX format to JSON.

I just removed the non-gzipped JSON files as Github has been notifying me that I am going over my data limits, my guess is it's the 420MB uncompressed JSON file I had in here :)