Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks

The goal of this project was to create end-to-end Deep Convolutional Neural Networks applied on the Street View House Numbers dataset to classify multi-digit numbers.

All you need to run the code is Python 3.x and Tensorflow.

Note that I trained the models on GPUs to make the training faster so running in local may take some time when training the model.

File Description

• DigitSequence: a folder with Python files to download the SVHN dataset and train the final model. Just type python train.py in this repository using your terminal to train the model.
• MNIST_multi_digit_classification.ipynb: a notebook where a Deep CNN is trained on multi-digit numbers from MNIST dataset (toy exemple)
• SVHN_multi_digit_classification.ipynb: a notebook where a Deep CNN is trained on multi-digit numbers from SVHN dataset
• DigitStructFile.py: a Python file containing helper functions to deal with .mat files

Resources

• Street View House Number (SVHN) Dataset
• Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks