MNIST Classification

Introduction

This repository contains code for training a fully Convolutional Network (FCN) model to classify handwritten digits using the MNIST dataset. This experiment adopts the SGD optimizer and cross-entropy loss function to train the network for 10 epochs, achieving a final accuracy of 99.6%.

Installation

Clone this repository:

git clone https://github.com/zlfffan/mnist.git

Install dependencies

conda update conda
conda create -n env_name python=x.x
pip install matplotlib

Usage

1. Run train.py to train the FCN model.
2. Run predict.py to view the model's prediction results.
3. Run the following command to view the training process:

tensorboard --logdir=run