Convolutional Neural Network on Street View House Numbers Classification
Procedures to execute the program:
Execute the data_preprocessing.ipynb to download, preprocessing downloaded images and generate a preprocessed data file named as SVHN_data.pickle.
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.
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.
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.