Street View House Numbers

Project built on two major steps which are coded in those files:

• 1-preprocess.ipynb: preprocess SVHN data.
• 2-recognition.ipynb: builds a convolutional neural network to recognize numbers from SVHN dataset

Important

1-preprocess.ipynb automatically downloads and extracts all required data(however, it takes more than an hour), so there is no need to do manually. However, if you want, you can do this on your own. In this case put all downloaded data in project directory(where this README file is located). Here are all the data which should be exist in project folder:

• Training set(train.tar.gz)
• Testing set(test.tar.gz)
• Extra set(extra.tar.gz)

Once again, this data could be downloaded automatically during script execution or put to project directory manually. If you already have these archives there is no need to do it again.

If you don't want to pass through all steps with data preprocessing(they take quite a lot time) at all you are able to go directly to step 2(file 2-recognition.ipynb) but first you need to put a SVHN.picke file into project directory. SVHN.pickle file could be downloaded here(Dropbox link) and should be moved to project directory where this README file is located.

Dataset

This project uses the The Street View House Numbers (SVHN) Dataset.

SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting. It can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.

Install

This project requires Python 2.7 and the following Python libraries installed:

• NumPy
• TensorFlow
• matplotlib
• scikit-learn
• SciPy library
• Six

You will also need to have iPython Notebook installed to run and execute code

Code

Code is provided in the following notebook files:

• 1-preprocess.ipynb
• 2-recognition.ipynb

Execute

In a terminal or command window, navigate to the top-level project directory P5-Capstone/ (that contains this README) and execute following commands:

jupyter notebook 1-preprocess.ipynb
jupyter notebook 2-recognition.ipynb

This will open the iPython Notebook and open file in your browser.