This repository contains notebooks, assets and a pre-trained model for a Convolutional Neural Network intended to predict digits from 28x28 images.
This CNN was used in the PREN project @ HSLU.
The repository consists of 3 parts:
- A jupyter notebook to augment and multiply the test images in the
images/folder - A jupyter notebook to train a CNN using the augmented test images
- A pre-trained model in
models/
The CNN classifies the numbers 1-9 and 0 for not-a-number.
See the jupyter notebooks for more details about the image augmentation and model architecture, training and evaluation.
Graph:
Accuracy during training:
This project is published under MIT.
A Timo Furrer project.
- 🎉 -