STL-10数据集
原⽂：
The STL-10 datat is an image recognition datat for developing unsupervid feature learning, deep learning, lf-taught learning algorithms. It is inspired by the but with some modifications. In particular, each class has fewer labeled training examples than in CIFAR-10, but a very large t of unlabeled examples is provided to learn image models prior to supervid training. The primary challenge is to make u of the unlabeled data (which comes from a similar but different distribution from the labeled data) to build a uful prior. We also expect that the higher resolution of this datat (96x96) will make it a challenging benchmark for developing more scalable unsupervid learning methods.
Overview
我的大学生活
10 class: airplane, bird, car, cat, deer, dog, hor, monkey, ship, truck.
Images are 96x96 pixels, color.
陌路人聊天
500 training images (10 pre-defined folds), 800 test images per class.
100000 unlabeled images for unsupervid learning. The examples are extracted from a similar but broader
奥运会会徽的含义
distribution of images. For instance, it contains other types of animals (bears, rabbits, etc.) and vehicles (trains, bus, etc.) in addition to the ones in the labeled t.
新东方名师
caterpillar音标Images were acquired from labeled examples on ImageNet.
译：冷静的英文
STL-10数据集是⼀个⽤于开发⽆监督特征学习、深度学习、⾃学学习算法的图像识别数据集。它的灵感来⾃CIFAR-10数据集，但经过了⼀些修改。特别是，每个类的标记训练⽰例⽐CIFAR-10中的要少，但是在监督训练之前，提供了⼀组⾮常⼤的未标记⽰例来学习图像模型。主要的挑战是利⽤未标记的数据（来⾃与标记数据相似但不同的分布）来构建有⽤的先验。我们还预计，该数据集（96x96）的更⾼分辨率将使其成为开发更具可伸缩性的⽆监督学习⽅法的具有挑战性的基准。
概述：
equalizer10级：飞机、鸟、车、猫、⿅、狗、马、猴、船、卡车。
图像为96x96像素，彩⾊。每日一句
500张训练图像（10张预先定义的折叠），每节课800张测试图像。
10万张⽆标签图像⽤于⽆监督学习。这些例⼦是从相似但更⼴泛的图像分布中提取的。例如，除了标签集中的动物外，它还包含其他类型的动物（熊、兔⼦等）和车辆（⽕车、公共汽车等）。sit
图像从ImageNet上的标记样本中获取。mbp
⼤家可以到官⽹地址下载数据集，我⾃⼰也在百度⽹盘分享了⼀份。可关注本⼈公众号，回复“2020100703”获取下载链接。
