Updated ReadMe
Browse files
README.md
CHANGED
|
@@ -1,3 +1,68 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
datasets:
|
| 4 |
+
- ylecun/mnist
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
- sw
|
| 8 |
+
base_model:
|
| 9 |
+
- ChufanSuki/LeNet5
|
| 10 |
+
---
|
| 11 |
+
# lenet-5_architecture_model
|
| 12 |
+
# Handwritten Digit Recognition
|
| 13 |
+
|
| 14 |
+
This project implements the LeNet-5 neural network architecture to recognize handwritten digits using the MNIST dataset.
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
## Table of Contents
|
| 18 |
+
|
| 19 |
+
- [Introduction](#introduction)
|
| 20 |
+
- [Prerequisites](#prerequisites)
|
| 21 |
+
- [Installation](#installation)
|
| 22 |
+
- [Model Architecture](#model-architecture)
|
| 23 |
+
- [Results](#results)
|
| 24 |
+
- [Contributing](#contributing)
|
| 25 |
+
|
| 26 |
+
## Introduction
|
| 27 |
+
|
| 28 |
+
LeNet-5 is a classic convolutional neural network (CNN) architecture designed by Yann LeCun,[learn more](https://en.wikipedia.org/wiki/LeNet), primarily for handwritten digit classification. This project uses [TensorFlow](https://www.tensorflow.org/guide/keras/functional_api) and Keras to build and train the LeNet-5 model on the MNIST dataset. The dataset is already included in this project.
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
## Prerequisites
|
| 32 |
+
|
| 33 |
+
Make sure you have the following installed:
|
| 34 |
+
|
| 35 |
+
- [Python_3.6+] (https://www.python.org/downloads/)
|
| 36 |
+
- [TensorFlow_2.x] (https://www.tensorflow.org/install)
|
| 37 |
+
- [NumPy] (https://numpy.org/install/)
|
| 38 |
+
- [Matplotlib](https://matplotlib.org/stable/install/index.html)
|
| 39 |
+
- [Pandas](https://pandas.pydata.org/docs/getting_started/install.html) (optional)
|
| 40 |
+
|
| 41 |
+
## Installation
|
| 42 |
+
|
| 43 |
+
Clone this repository:
|
| 44 |
+
|
| 45 |
+
```bash
|
| 46 |
+
git clone https://github.com/jumarubea/lenet-5_architecture_model.git
|
| 47 |
+
cd lenet-5-digit-recognition
|
| 48 |
+
```
|
| 49 |
+
## Model Architecture
|
| 50 |
+
LeNet-5 consists of the following layers:
|
| 51 |
+
|
| 52 |
+
- Convolutional Layer: 6 filters of size 5x5, activation function: tanh
|
| 53 |
+
- Average Pooling Layer: pool size 2x2
|
| 54 |
+
- Convolutional Layer: 16 filters of size 5x5, activation function: tanh
|
| 55 |
+
- Average Pooling Layer: pool size 2x2
|
| 56 |
+
- Convolutional Layer: 120 filters of size 5x5, activation function: tanh
|
| 57 |
+
- Flatten Layer
|
| 58 |
+
- Dense Layer: 84 units, activation function: tanh
|
| 59 |
+
- Output Layer: 10 units, activation function: softmax
|
| 60 |
+
|
| 61 |
+
Note: for the purpose of accuracy measure, i implement `relu` activation instead of `tanh`
|
| 62 |
+
except for the 84 dense layer.
|
| 63 |
+
|
| 64 |
+
## Results
|
| 65 |
+
The trained LeNet-5 model achieves a test accuracy of approximately 98% on the MNIST dataset.
|
| 66 |
+
|
| 67 |
+
## Contributing
|
| 68 |
+
If you want to contribute to this project, please fork the repository and submit a pull request.
|