/Multi-number-recognition-street-view

CNN model to detect street view imagery

Primary LanguageJupyter Notebook

Multi-number-recognition-street-view

paper link | report link

Final notebook: "SVHN_Classifier Random Cropping Latest.ipynb"

Please download the .ipynb file titled "SVHN_Classifier Random Cropping Latest.ipynb" and import it into Google Colab and run the notebook. The reproduction of the network on SVHN dataset has been tested and the results have been confirmed as correct.

For the code used, we have taken reference from these 2 links:

  1. Street View House Numbers Recognition Using ConvNets
  2. SVHNClassifier

Some examples of images from street view imagery data:

CNN architecture

Results

We achieved an accuracy of over 95% at a coverage of 93%, which was comparable to the results the authors obtained