Overview
SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting. It can be seen as similar in flavor to MNIST (e.g. the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.
- 10 classes, 1 for each digit. Digit
'1'has label 1,'9'has label 9 and'0'has label 0. - 73257 digits for training, 26032 digits for testing, and 531131 additional, somewhat less difficult samples, to use as extra training data.
- Comes in two formats:
- Original images with character level bounding boxes.
- MNIST-like 32-by-32 images centered around a single character (many of the images do contain some distractors at the sides).
Data Format
Format 1: Full Numbers
These are the original, variable-resolution, color house-number images with character level
bounding boxes, as shown in the examples images above. (The blue bounding boxes here are just
for illustration purposes. The bounding box information are stored in digitStruct.mat instead
of drawn directly on the images in the dataset.) Each tar.gz file contains the orignal images
in png format, together with a digitStruct.mat file, which can be loaded using Matlab. The
digitStruct.mat file contains a struct called digitStruct with the same length as the number
of original images. Each element in digitStruct has the following fields: name which is
a string containing the filename of the corresponding image. bbox which is a struct array
that contains the position, size and label of each digit bounding box in the image. Eg: digitStruct(300).bbox(2).height
gives height of the 2nd digit bounding box in the 300th image.
Format 2: Cropped Digits
Character level ground truth
in an MNIST-like format. All digits have been resized to a fixed resolution of 32-by-32 pixels.
The original character bounding boxes are extended in the appropriate dimension to become
square windows, so that resizing them to 32-by-32 pixels does not introduce aspect ratio distortions.
Nevertheless this preprocessing introduces some distracting digits to the sides of the
digit of interest. Loading the .mat files creates 2 variables: X which is a 4-D matrix
containing the images, and y which is a vector of class labels. To access the images, X(:,:,:,i)
gives the i-th 32-by-32 RGB image, with class label y(i).
Citation
Please use the following citation when referencing the dataset:
@article{netzer2011reading,
title={Reading digits in natural images with unsupervised feature learning},
author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Bo
and Ng, Andrew Y},
year={2011}
}