Oktoberfest Food Dataset

我们发布了一个现实的、多样化的和具有挑战性的 用于图像上对象检测的数据集。数据 记录在德国的一个啤酒帐篷中，并记录了15种不同类别的食品和饮料 项。我们用手创建了2,500多个对象a符号，用于捕获的1,110图像 收银台上方的摄像机。我们进一步 提供剩余的600gb (unla beled) 数据，其中包含几天的素材。此外，我们还提供训练有素的模型作为基准。可能的应用包括自动化 结帐系统可以显着加快速度 加快进程。We present a realistic, diverse, and challenging dataset for object detection on images. The data was collected in a German beer tent, covering 15 distinct categories of food and beverage items. We manually annotated over 2,500 object instances across 1,110 images captured by a camera mounted above the checkout counter. Additionally, we provide the remaining 600 GB of unlabeled data, which consists of footage gathered over several days. Furthermore, we offer pre-trained models as benchmark baselines. Potential applications include automated checkout systems, which can substantially accelerate the checkout process.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于目标检测的现实多样化数据集，包含15类食品和饮料，基于德国啤酒帐篷中拍摄的1,110张图像，并提供了2,500多个手动标注。此外，它还包含600GB未标注数据和预训练模型，适用于自动化结账系统等应用场景。