Convolutional Neural Network on Street View House Numbers Classification

Procedures to execute the program:

Step 1:

Execute the data_preprocessing.ipynb to download, preprocessing downloaded images and generate a preprocessed data file named as SVHN_data.pickle.

Step 2:

Execute the data_analysis.ipynb to extract images from SVHN_data.pickle and generate two files named SVHN_weight.ckpt, SVHN_weight.ckpt.meta separately.

Step 3:

Execute the prediction_test.ipynb to read in folder named ‘prediction’’s images and read in the file SVHN_weight.ckpt generated in Step 2 then predict prediction images’ results.

Demo

Note:

Prediction folder contains 17 pictures cropped by author himself from Google Street View in Manhattan, New York. All the 17 images have been resized to 32 x 32.

Environment:

TensorFlow 0.11

Python 2.7

Jupyter Notebook