Office-31 Dataset, also known as Office Dataset, is a mainstream benchmark dataset in visual transfer learning. This dataset contains 31 types of target objects commonly found in office environments, such as laptops, filing cabinets, keyboards, etc., with a total of 4,652 images. These images mainly come from Amazon (online e-commerce pictures), Webcam (low-resolution pictures taken by webcams), and DSLR (high-resolution pictures taken by single-lens reflex cameras). The dataset includes:

• Amazon: 2,817 images, 90 images per class on average, with a single background
• Webcam: 795 images, images show noticeable noise, color, and white balance artifacts
• DSLR: 498 images, 5 objects per class, each object photographed 3 times on average from different viewpoints Example image: