mat files creates 2 variables: X which is a 4-D matrix containing the images, and y which is a vector of class labels. Nevertheless this preprocessing introduces some distracting digits to the sides of the digit of interest.

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. All digits have been resized to a fixed resolution of 32-by-32 pixels. Format 2: Cropped Digits: train_32x32.mat, test_32x32.mat, extra_32x32.mat (Note: for non-commercial use only)Ĭharacter level ground truth in an MNIST-like format.Eg: digitStruct(300).bbox(2).height gives height of the 2nd digit bounding box in the 300th image. Each element in digitStruct has the following fields: name which is a string containing the filename of the corresponding image.ībox which is a struct array that contains the position, size and label of each digit bounding box in the image. The digitStruct.mat file contains a struct called digitStruct with the same length as the number of original images. The bounding box information are stored in digitStruct.mat instead of drawn directly on the images in the dataset.)Įach tar.gz file contains the orignal images in png format, together with a digitStruct.mat file, which can be loaded using Matlab. (The blue bounding boxes here are just for illustration purposes. These are the original, variable-resolution, color house-number images with character level bounding boxes, as shown in the examples images above. Format 1: Full Numbers:, , (Note: for non-commercial use only).MNIST-like 32-by-32 images centered around a single character (many of the images do contain some distractors at the sides). Original images with character level bounding boxes.Ģ. Digit '1' has label 1, '9' has label 9 and '0' has label 10.ħ3257 digits for training, 26032 digits for testing, and 531131 additional, somewhat less difficult samples, to use as extra training dataġ. SVHN is obtained from house numbers in Google Street View images.ġ0 classes, 1 for each digit. It can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits),īut incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes fromĪ significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting.

Download street view