Voxel51/sku110k_test

SKU-110K数据集是一个大规模的对象检测基准数据集，专为密集零售场景设计。该数据集包含了来自全球数千家超市的11,762张零售货架图像，涵盖了美国、欧洲和东亚等不同地理位置。数据集含有超过1.73百万个边界框注释，平均每张图像大约有147个对象。所有图像都被调整到一百万像素的分辨率。该数据集旨在解决对象检测研究中的一个关键空白：缺乏用于密集场景的大规模数据集。现有的数据集如COCO和Pascal VOC虽然包含对象检测注释，但通常场景相对稀疏，对象之间分离良好。而现实世界的零售环境呈现出的独特挑战包括极端的密度、严重的遮挡、尺寸变化和对象的方向性。

The SKU-110K dataset is a large-scale object detection benchmark designed for densely packed retail scenes. It consists of 11,762 images of retail shelves from thousands of supermarkets worldwide, covering diverse geographic locations including the United States, Europe, and East Asia. The dataset contains over 1.73 million bounding box annotations, with an average of approximately 147 objects per image. All images have been resized to a resolution of one million pixels. The dataset was created to address a critical gap in object detection research: the lack of large-scale datasets for densely packed scenes. Existing datasets like COCO and Pascal VOC contain object detection annotations but typically feature relatively sparse scenes with well-separated objects. Real-world retail environments present unique challenges including extreme density, heavy occlusion, scale variation, and object orientation patterns.