Draw-and-Recognize-MNIST

This project implements a Simple Feed Forward Neural Network and a Convolutional Neural Network (CNN) for handwritten digit recognition using the MNIST dataset. The project includes a terminal-based interface that allows users to draw digits in real time using their mouse. Once a digit is drawn, it is processed and classified by the trained model.

---

Requirements

For Training the Model

• Python 3.8+
• Libraries:
  • PyTorch
  • torchvision
  • tqdm
  • numpy
  • matplotlib
  • jupyter

For Running the Interface

• g++ or clang (C++11 or later)
• LibTorch (PyTorch C++ API)
• ncurses library
• cmake

---

Getting Started

1. Clone the Repository

   git clone https://github.com/Xzanam/draw-and-recognize-mnist.git
   cd draw-and-recognize-mnist

2. Install the required Python packages:

Training the Model

The training code is provided in mnist.ipynb. Follow these steps:

1. Open the notebook:

   jupyter notebook mnist.ipynb

2. Run the cells to train the model on the MNIST dataset. mnist.ipynb consists of training code for both Feed Forward Neural Network and Convolutional Neural Network. The notebook itself is self-explanatory. After training, the model is exported as 'model_cnn.pt' using PyTorch's JIT.

Compiling the C++ interface

To compile the C++ interface, follow these steps:

1. Create a build directory: Open your terminal and navigate to the project folder. Create a separate directory for the build files to keep things organized:

   mkdir build
   cd build

2. Run CMake: Use CMake to generate the necessary build files. Make sure you are inside the build directory:

   cmake ..

3. Compile the code: Now, compile the code using make:

   make

After these steps, an executable named character_recog should be created in the build directory, ready to run with your trained model.

Running the Interface

Launch the application with:

./character_recog <path_to_the_trained_model>

In my case

./character_recog ../mnist_CNN.pt

Instructions

• Draw: Use your mouse to draw a digit directly in the terminal.
• Predict: Press p to classify the drawn digit.
• Clear: Press c to erase the current drawing and start over.
• Quit: Press q to exit the application.

Demo

digit_recog.mp4

Troubleshooting

Common Errors

1. libtorch.so: No such file or directory

   • This error usually indicates that LibTorch is not installed or the paths provided during compilation are incorrect.
   • Solution: Verify that LibTorch is installed.

2. Mouse not working

   • If the mouse input is not functioning as expected, it might be due to a missing or incompatible ncurses installation.
   • Solution: Ensure that ncurses is properly installed on your system and that your terminal supports mouse input.