This repository provide you python scripts for decoding images of the EMNIST from the official binary format files. The EMNIST contains not only digits images but also alphabet ones.
$ sh download_binary.sh
$ unzip gzip.zip
$ cd gzip/
$ gzip -d emnist-*.gzThe EMNIST dataset consists of some datasets.
Choose desired one from the following list.
EMNIST Balanced: 131,600 characters. 47 balanced classes.EMNIST ByClass: 814,255 characters. 62 unbalanced classes.EMNIST ByMerge: 814,255 characters. 47 unbalanced classes.EMNIST Digits: 280,000 characters. 10 balanced classes.EMNIST Letters: 145,600 characters. 26 balanced classes.
Here I'll show an example in the case of EMNIST Balanced (whatever you choose, the following process is exactly the same).
First, comment out the other lines than the part of EMNIST Balanced from decode.py as follows.
#!/bin/sh
# EMNIST Balanced: 131,600 characters. 47 balanced classes.
python decode.py gzip/emnist-balanced-train-images-idx3-ubyte gzip/emnist-balanced-train-labels-idx1-ubyte gzip/emnist-balanced-mapping.txt ./emnist_balanced/train
python decode.py gzip/emnist-balanced-test-images-idx3-ubyte gzip/emnist-balanced-test-labels-idx1-ubyte gzip/emnist-balanced-mapping.txt ./emnist_balanced/test
# EMNIST ByClass: 814,255 characters. 62 unbalanced classes.
#python decode.py gzip/emnist-byclass-train-images-idx3-ubyte gzip/emnist-byclass-train-labels-idx1-ubyte gzip/emnist-byclass-mapping.txt ./emnist_byclass/train
#python decode.py gzip/emnist-byclass-test-images-idx3-ubyte gzip/emnist-byclass-test-labels-idx1-ubyte gzip/emnist-byclass-mapping.txt ./emnist_byclass/test
# EMNIST ByMerge: 814,255 characters. 47 unbalanced classes.
#python decode.py gzip/emnist-bymerge-train-images-idx3-ubyte gzip/emnist-bymerge-train-labels-idx1-ubyte gzip/emnist-bymerge-mapping.txt ./emnist_bymerge/train
#python decode.py gzip/emnist-bymerge-test-images-idx3-ubyte gzip/emnist-bymerge-test-labels-idx1-ubyte gzip/emnist-bymerge-mapping.txt ./emnist_bymerge/test
# EMNIST Digits: 280,000 characters. 10 balanced classes.
#python decode.py gzip/emnist-digits-train-images-idx3-ubyte gzip/emnist-digits-train-labels-idx1-ubyte gzip/emnist-digits-mapping.txt ./emnist_digits/train
#python decode.py gzip/emnist-digits-test-images-idx3-ubyte gzip/emnist-digits-test-labels-idx1-ubyte gzip/emnist-digits-mapping.txt ./emnist_digits/test
# EMNIST Letters: 145,600 characters. 26 balanced classes.
#python decode.py gzip/emnist-letters-train-images-idx3-ubyte gzip/emnist-letters-train-labels-idx1-ubyte gzip/emnist-letters-mapping.txt ./emnist_letters/train
#python decode.py gzip/emnist-letters-test-images-idx3-ubyte gzip/emnist-letters-test-labels-idx1-ubyte gzip/emnist-letters-mapping.txt ./emnist_letters/testSecond, run decode.sh in your command line.
$ sh decode.shAfter running decode.sh, A directory emnist_balanced is created in the current directory.
The training images are saved in emnist_balanced/train, while the validation images are in emnist_balanced/test.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent