The Street View House Numbers (SVHN) Dataset - Revision history

Dimos at 10:20, 17 January 2012

2012-01-17T10:20:15Z

Dimos: /* Metadata and Ground Truth Data */

2012-01-04T15:58:12Z

‎Metadata and Ground Truth Data

Dimos at 15:27, 4 January 2012

2012-01-04T15:27:04Z

Dimos: Created page with 'Datasets -> Datasets List -> Current Page {| style="width: 100%" |- | align="right" | {| |- | '''Created: '''2012-01-05 |- | {{Last updated}} |} |} =Contact Author=…'

2012-01-04T15:25:48Z

Created page with 'Datasets -> Datasets List -> Current Page {| style="width: 100%" |- | align="right" | {| |- | '''Created: '''2012-01-05 |- | {{Last updated}} |} |} =Contact Author=…'

New page

[[Datasets]] -> [[Datasets List]] -> Current Page

{| style="width: 100%"
|-
| align="right" |

{|
|-
| '''Created: '''2012-01-05
|-
| {{Last updated}}
|}

|}

=Contact Author=
For questions regarding the dataset, please contact streetviewhousenumbers@gmail.com

=Current Version=
[[Image:svhn_examples.png|340px|thumb|right| Example images from the SVHN dataset. Full numbers format: original variable resolution images with character level bounding boxes.]]
[[Image:svhn_32x32eg.png|340px|thumb|right| Example images from the SVHN dataset. Cropped digits format: character level images in a MNIST-like format, fixed resolution of 32x32 pixels.]]

1.0 (also available from the [http://ufldl.stanford.edu/housenumbers/ SVHN Web site])

=Keywords=
OCR, Natural Scene, Scene Text, Numbers, Scene Text Recognition

=Description=
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.

==Overview==
* 10 classes, 1 for each digit. Digit '1' has label 1, '9' has label 9 and '0' has label 10.
* 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:
** 1. Original images with character level bounding boxes.
** 2. MNIST-like 32-by-32 images centered around a single character (many of the images do contain some distractors at the sides).

=Metadata and Ground Truth Data=

==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 in the example image 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).

[[#References|[1]]], the technical report [[#References|[2]]] or the [http://www.iapr-tc11.org/dataset/NEOCR/neocr_metadata_doc.pdf metadata documentation] for further details on the metadata.

=Related Tasks=
* [[Text Recognition in Natural Scenes]]

=References=
Please cite the following reference in papers using this dataset:
# Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, Andrew Y. Ng Reading Digits in Natural Images with Unsupervised Feature Learning NIPS Workshop on Deep Learning and Unsupervised Feature Learning 201 [http://www.iapr-tc11.org/dataset/SVHN/nips2011_housenumbers.pdf (PDF)]

=Submitted Files=
==Disclaimer==
This dataset is provided for non-commercial use only.

By downloading and using the dataset you agree to acknowledge it's source and cite the above papers in related publications. Please link to the authors' URL for this dataset as http://ufldl.stanford.edu/housenumbers.

==Version 1.0==
===Format 1: Full Numbers===
* [http://www.iapr-tc11.org/dataset/SVHN/train.tar.gz Training set (73257 digits)] (*** MB)
* [http://www.iapr-tc11.org/dataset/SVHN/test.tar.gz Test set (26032 digits)] (*** KB)
* [http://www.iapr-tc11.org/dataset/SVHN/extra.tar.gz Extra training set (531131 additional, somewhat less difficult samples)] (*** MB)

===Format 2: Cropped Digits===
* [http://www.iapr-tc11.org/dataset/SVHN/train_32x32.mat Training set (73257 digits)] (*** MB)
* [http://www.iapr-tc11.org/dataset/SVHN/test_32x32.mat Test set (26032 digits)] (*** KB)
* [http://www.iapr-tc11.org/dataset/SVHN/extra_32x32.mat Extra training set (531131 additional, somewhat less difficult samples)] (*** MB)

----
This page is editable only by [[IAPR-TC11:Reading_Systems#TC11_Officers|TC11 Officers ]].

@@ Line 61: / Line 61: @@
 ==Version 1.0==
 ===Format 1: Full Numbers===
-* [http://www.iapr-tc11.org/dataset/SVHN/train.tar.gz Training set (73257 digits)] (*** MB)
+* [http://www.iapr-tc11.org/dataset/SVHN/train.tar.gz Training set (73257 digits)] (395 MB)
-* [http://www.iapr-tc11.org/dataset/SVHN/test.tar.gz Test set (26032 digits)] (*** KB)
+* [http://www.iapr-tc11.org/dataset/SVHN/test.tar.gz Test set (26032 digits)] (270 MB)
-* [http://www.iapr-tc11.org/dataset/SVHN/extra.tar.gz Extra training set (531131 additional, somewhat less difficult samples)] (*** MB)
+* [http://www.iapr-tc11.org/dataset/SVHN/extra.tar.gz Extra training set (531131 additional, somewhat less difficult samples)] (1.9 GB)
 ===Format 2: Cropped Digits===
-* [http://www.iapr-tc11.org/dataset/SVHN/train_32x32.mat Training set (73257 digits)] (*** MB)
+* [http://www.iapr-tc11.org/dataset/SVHN/train_32x32.mat Training set (73257 digits)] (177 MB)
-* [http://www.iapr-tc11.org/dataset/SVHN/test_32x32.mat Test set (26032 digits)] (*** KB)
+* [http://www.iapr-tc11.org/dataset/SVHN/test_32x32.mat Test set (26032 digits)] (62 MB)
-* [http://www.iapr-tc11.org/dataset/SVHN/extra_32x32.mat Extra training set (531131 additional, somewhat less difficult samples)] (*** MB)
+* [http://www.iapr-tc11.org/dataset/SVHN/extra_32x32.mat Extra training set (531131 additional, somewhat less difficult samples)] (1.3 GB)
 ----
 This page is editable only by [[IAPR-TC11:Reading_Systems#TC11_Officers|TC11 Officers ]].

@@ Line 35: / Line 35: @@
 ** 2. MNIST-like 32-by-32 images centered around a single character (many of the images do contain some distractors at the sides).
-=Metadata and Ground Truth Data=
+=Technical Details=
 ==Format 1: Full Numbers==

@@ Line 29: / Line 29: @@
 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 10.
 * 73257 digits for training, 26032 digits for testing, and 531131 additional, somewhat less difficult samples, to use as extra training data
@@ Line 49: / Line 48: @@
 To access the images, X(:,:,:,i) gives the i-th 32-by-32 RGB image, with class label y(i).
 =References=
@@ Line 76: / Line 69: @@
 * [http://www.iapr-tc11.org/dataset/SVHN/test_32x32.mat Test set (26032 digits)] (*** KB)
 * [http://www.iapr-tc11.org/dataset/SVHN/extra_32x32.mat Extra training set (531131 additional, somewhat less difficult samples)] (*** MB)
 ----
 This page is editable only by [[IAPR-TC11:Reading_Systems#TC11_Officers|TC11 Officers ]].