Scribe is a command line utility for converting the binary data files of the MNIST data set (see: http://yann.lecun.com/exdb/mnist/) to a collection of BMP images.
To compile, run cargo build --release. You will find the executable in the directory ./target/release.
Then, at the same level as the src directory, create the following directories:
./out/0
./out/1
./out/2
./out/3
./out/4
./out/5
./out/6
./out/7
./out/8
./out/9
The script requires 4 parameters:
- Two paths relative to the current working directory
- The location of the image data file
- The location of the corresponding labels file
- A 1-based index describing which image to start with
- How many images to read
For example, if your directory structure looks like this:
./mnist
t10k-images-idx3-ubyte
t10k-labels-idx1-ubyte
train-images-idx3-ubyte
train-labels-idx1-ubyte
You could invoke the script like this to dump all 10,000 testing images:
./target/release/scribe ./mnist/t10k-images-idx3-ubyte ./mnist/t10k-labels-idx1-ubyte 1 10000
As another example, you could just dump 3,000 images, starting from image number 6,000:
./target/release/scribe ./mnist/t10k-images-idx3-ubyte ./mnist/t10k-labels-idx1-ubyte 6000 3000
The script will output images in a directory called out. Images are organized by the type into
separate folders. That is, all the zeros go into ./out/0, all the ones into ./out/1, etc.
Each image is named in the following pattern: d{type}-{id}.bmp so the name: d5-0040.bmp would
indicate this image is a 5 (denoted by d5) and it is the 41st 5 out of all 5s read from the dataset.
The images are 0 indexed, which is why 0040 is the 41st image.
You can also print a help message by invoking the script with a single option: --help, like so:
scribe --help.
The tool does not have very robust error reporting and is not configurable aside from the options described here, but it is functional!