Office-31 (Office Dataset)

Office 数据集包含三个域中的 31 个对象类别：Amazon、DSLR 和 Webcam。数据集中的 31 个类别由办公环境中常见的对象组成，例如键盘、文件柜和笔记本电脑。 Amazon 域平均每个类包含 90 张图像，总共包含 2817 张图像。由于这些图像是从在线商家网站捕获的，因此它们是在干净的背景下以统一的比例捕获的。 DSLR域包含498张低噪声高分辨率图像（4288×2848）。每个类别有 5 个对象。每个对象平均从不同的角度拍摄 3 次。对于网络摄像头，795 张低分辨率 (640×480) 图像表现出明显的噪点和色彩以及白平衡伪影。

The Office Dataset contains 31 object categories across three domains: Amazon, DSLR, and Webcam. The 31 categories consist of common objects found in office environments, such as keyboards, filing cabinets, and laptops. The Amazon domain has an average of 90 images per category, with a total of 2817 images. As these images were captured from online merchant websites, they are shot against clean backgrounds with uniform scales. The DSLR domain includes 498 high-resolution (4288×2848) images with low noise. Each category contains 5 objects, and each object is photographed an average of 3 times from different angles. For the Webcam domain, the 795 low-resolution (640×480) images exhibit significant noise, as well as color and white balance artifacts.

搜集汇总

背景与挑战

背景概述

Office-31数据集是一个用于迁移学习和领域适应的图像分类数据集，包含31个办公常见对象类别，分为Amazon、DSLR和Webcam三个域，总计约4110张图像。Amazon域图像背景干净且比例统一，DSLR域为高分辨率低噪声图像，Webcam域为低分辨率图像带有噪点和色彩伪影，适用于跨域视觉模型研究。