Office-Caltech10数据集中包含Amazon, Caltech, DSLR, Webca...

我将表述调整为更连贯、学术化的风格，并尽量采用机器学习/计算机视觉顶会论文中常见的措辞。

:::writing{variant="document" id="48317" title="中文润色版"} Office-Caltech10 数据集包含 Amazon、Caltech、DSLR 和 Webcam 四个领域。尽管这四个领域共享相同的类别空间，但由于图像采集方式和成像条件不同，其视觉分布存在显著差异。具体而言，Amazon 域的图像主要来自 Amazon 商品页面，通常具有目标居中、光照均匀以及背景干净等特点；Caltech 域中的图像更接近自然场景，其背景、拍摄视角和目标尺度具有更大的变化；DSLR 域由高质量 DSLR 相机拍摄的办公物品图像组成，通常采用近距离拍摄，因此具有较高的清晰度和图像分辨率；相比之下，Webcam 域使用普通网络摄像头采集，其图像分辨率相对较低，并伴随更明显的模糊、光照变化和成像噪声。

另一常用的跨域数据集为 Digits，其中包含 MNIST、SYN、SVHN 和 USPS 四个领域。MNIST 由经典的手写数字图像组成，通常具有较为干净且简单的背景；SYN 由人工合成的数字图像构成，而非来源于真实拍摄或人工书写；SVHN 来源于真实街景中的门牌数字，因此具有更加复杂的背景、颜色分布和光照条件；USPS 则由美国邮政信封上扫描得到的手写数字组成，其成像方式和图像风格与其他数字域存在明显差异。 :::

:::writing{variant="document" id="72641" title="Top-Conference-Style English Version"} The Office-Caltech10 dataset consists of four domains, namely Amazon, Caltech, DSLR, and Webcam. Although these domains share the same label space, they exhibit substantial visual distribution shifts due to differences in image sources and acquisition conditions. Specifically, images in the Amazon domain are collected from Amazon product pages, where objects are typically centered with relatively uniform illumination and clean backgrounds. The Caltech domain is more representative of natural imagery, exhibiting greater variations in background, viewpoint, and object scale. The DSLR domain contains office-object images captured using high-quality DSLR cameras, typically at close range, resulting in high image resolution and sharp visual details. In contrast, the Webcam domain is acquired using standard webcams and therefore generally exhibits lower resolution, stronger blur, and more pronounced variations in illumination and imaging noise.

Another widely used cross-domain benchmark is Digits, which comprises four domains: MNIST, SYN, SVHN, and USPS. MNIST contains canonical handwritten digit images with relatively clean and simple backgrounds. SYN consists of synthetically generated digit images rather than digits obtained from real-world photographs or human handwriting. SVHN is collected from house numbers in natural street-view scenes and thus exhibits substantially more complex backgrounds, color distributions, and illumination conditions. USPS, in contrast, consists of handwritten digits scanned from U.S. postal envelopes, resulting in a distinct acquisition process and visual style compared with the other digit domains. :::

如果用于论文的 Dataset / Experimental Setup 部分，还可以进一步压缩成更典型的 CVPR/ICCV/NeurIPS 风格，使语言更简洁、客观。