Handwriting-to-Text Recognition Model (C Implementation)

This repository contains the C implementation of a Convolutional Neural Network (CNN) model trained on the MNIST dataset for recognizing handwritten digits.

Installation

Clone the repository:

git clone https://github.com/yourusername/handwriting-to-text-recognition-c.git

Compile the code:
```
 make
```

Usage

Run the training file:
```
./main
```
Run the test file:
```
./test
```

Model Architecture

The model architecture is implemented in model_saver.c and consists of functions for initializing the model, saving the model and loading the model.

Dataset

The dataset used is included in /data consisting of 60,000 training and 10,000 testing dataset in idx3-ubyte format for images and id1-ubyte format for labels. The file MNIST_data_loader.c consists of funtion to load, unload and get data images and labels. +

Training

The training data is loaded from the MNIST dataset and the model is trained using stochastic gradient descent. After the training, the model parameter is saved in /reults/model.txt that is stored and can be loaded using funtions stored in model_saver.c.

Evaluation

The model's performance is evaluated on the test dataset to measure its accuracy in recognizing handwritten digits.

Results

Inference time: Inference time of C is far less than that observed by running that with the python code which is about (1/10) of that. The training reult is stored in /results folder. After training, the model achieved an accuracy of 99% on the training set and 98% on the vali-dation set. On the test set, the model achieved an accuracy of 97%, demonstrating its effectiveness in recognizing handwritten digits.

Train Accuracy

Test Accuracy

Train Loss

Test Loss

madhavkataria1010/ics_majorproject