Genius-Society/svhn

SVHN数据集是一个面向机器学习和对象识别算法的实际图像数据集，具有较低的数据预处理和格式要求。与MNIST类似，SVHN包含小尺寸裁剪数字的图像，但在标记数据方面，SVHN比MNIST大一个数量级，包含超过60万张数字图像。与MNIST不同，SVHN处理的是一个更具挑战性和未解决的实时世界问题——在自然场景图像中识别数字和数字字符串。该数据集来源于Google街景图像中捕获的门牌号码。

The Street View House Numbers (SVHN) dataset is a real-world image dataset developed and designed for machine learning and object recognition algorithms, characterized by low data preprocessing and formatting requirements. Similar to MNIST, SVHN contains images of small cropped numbers, but in terms of labeled data, SVHN is an order of magnitude larger than MNIST, comprising over 600,000 digital images. Unlike MNIST, SVHN deals with a much more challenging and unsolved real-world problem - recognizing numbers and strings of numbers in natural scene images. The dataset is derived from house numbers captured in Google Street View images.