https://github.com/manohara-ai/hand_written_digit_recognizer
Building a CNN to identify hand written digits
https://github.com/manohara-ai/hand_written_digit_recognizer
cnn machine-learning multiclass-image-classification tinyvgg
Last synced: 10 days ago
JSON representation
Building a CNN to identify hand written digits
- Host: GitHub
- URL: https://github.com/manohara-ai/hand_written_digit_recognizer
- Owner: Manohara-Ai
- License: mit
- Created: 2024-08-17T09:00:57.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-12-08T10:37:38.000Z (5 months ago)
- Last Synced: 2025-03-31T02:41:44.570Z (about 2 months ago)
- Topics: cnn, machine-learning, multiclass-image-classification, tinyvgg
- Language: Python
- Homepage:
- Size: 101 KB
- Stars: 4
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Handwritten Digit Recognition using CNN
This project implements a Convolutional Neural Network (CNN) similar to the TinyVGG architecture to recognize handwritten digits. The model was trained on the MNIST dataset and tested on custom handwritten digits.
## Model Overview
The CNN architecture is inspired by TinyVGG and consists of multiple convolutional layers followed by ReLU activation, max-pooling, and fully connected layers. The model achieved **98.12% accuracy (on random seed and accuracy can be better/ worse on other seeds)** on both the MNIST test data and custom handwritten digits created using GIMP.
## Dataset
- **Training Data:** MNIST dataset of handwritten digits (60,000 training images)
- **Testing Data:** MNIST test set (10,000 images) and custom images created in GIMP
## Results
The model achieved:
- **98.12% accuracy (on random seed and accuracy can be better/ worse on other seeds)** on custom handwritten digits
## Requirements
To run this project, you will need the following dependencies:
- Python 3.x
- PyTorch
- OpenCV
- NumPy
- Matplotlib
NOTE: The images in MNIST dataset, on which the model are greyscale and 28x28 pixels in size. So, ensure to test on images of same size and color channels.
## Contributor
B M Manohara @Manohara-Ai