Convolutional Neural Networks Applied to House Numbers Digit Classification

TL;DR

This paper demonstrates that convolutional neural networks can effectively classify house number digits, achieving state-of-the-art accuracy by learning multi-stage features and employing advanced pooling methods.

Contribution

The paper introduces an augmented ConvNet architecture with multi-stage feature learning and Lp pooling, setting a new accuracy benchmark on the SVHN dataset.

Findings

Achieved 94.85% accuracy on SVHN dataset

Demonstrated benefits of multi-stage features and Lp pooling

Provided source code and tutorial for replication

Abstract

We classify digits of real-world house numbers using convolutional neural networks (ConvNets). ConvNets are hierarchical feature learning neural networks whose structure is biologically inspired. Unlike many popular vision approaches that are hand-designed, ConvNets can automatically learn a unique set of features optimized for a given task. We augmented the traditional ConvNet architecture by learning multi-stage features and by using Lp pooling and establish a new state-of-the-art of 94.85% accuracy on the SVHN dataset (45.2% error improvement). Furthermore, we analyze the benefits of different pooling methods and multi-stage features in ConvNets. The source code and a tutorial are available at eblearn.sf.net.