Number Reconstructor

This program uses the BMP Parser I wrote a few months ago to reconstruct training data from the MNIST Dataset to a BMP image.

The dataset I'm starting with consists of 785 columns, the first column is the actual value of the number and the last 784 is each pixel in a 28x28 grid.

So esentially we start with a CSV of a bunch of pixels and we construct it into an actual image!

What is the MNIST dataset?

The MNIST (Modified National Institute of Standards and Technology database) is a database of handwritten digits that is mostly used for image processing and machine learning.

Why reconstruct them?

I'm experimenting with writing machine learning models in C to target embedded devices (mostly microcontrollers) with the end goal of training and running an AI model to detect a hand written number completely on a microcontroller!

Usage

There is either a demo program which just generates 4.bmp or line-number which allows you to specify a certain row in the testing data to generate.

line-number is mostly used for me to validate the accuracy of my model and help with tweaking my weights

First unzip the mnist-in-csv.zip to a folder called mnist-in-csv and then make and run the program

make
./build/demo

And then you will get something like this!

You can also run line-number like this:

spencer@spencer-thinkpad:~/Documents/code/C/number-reconstructor$ ./build/line-number 
Usage: ./build/line-number <row_number>
spencer@spencer-thinkpad:~/Documents/code/C/number-reconstructor$ ./build/line-number 1
Row number must be 2 or more (row 1 is header)
spencer@spencer-thinkpad:~/Documents/code/C/number-reconstructor$ ./build/line-number 40
Generated BMP for row 40 with number 2
spencer@spencer-thinkpad:~/Documents/code/C/number-reconstructor$ 

and here's our result!