* This dataset supports online use.Click here to jump.

The SVHN dataset is a real-world image dataset for developing machine learning and object recognition algorithms, with minimal requirements for data preprocessing and formatting. It can be seen that it has similar characteristics to MNIST (for example, the images are small cropped digits), but contains orders of magnitude more labeled data (over 600,000 digit images) and comes from a more difficult, unsolved real-world problem (recognizing digits and numbers in natural scene images).

This dataset was released by Stanford University in 2011. The data are all collected from house numbers in Google Street View images. The related paper is "Reading Digits in Natural Images with Unsupervised Feature Learning".