Skip to content

Dataset · ufldl-stanford

Street view house numbers recognition

ufldl-stanford/svhn

Over 600 thousand real-world digit images for classification and object detection.

What's inside

A classic Stanford dataset consisting of over 600 thousand labeled digit images cropped from real-world street view house number photographs. Format-wise it resembles MNIST, but presents a much more challenging real-world task with noise and distortions. The data concerns the English segment and does not require complex preprocessing.

What you can do with it

  • Train image classification and basic OCR models
  • Train digit and text detectors in natural street environments
  • Test convolutional neural network robustness against real visual noise
Tasks
Image classification, Object detection
Languages
English
Size
100K–1M rows
Downloads this month
29.7K
Likes
16

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("ufldl-stanford/svhn")