FoodX-251

由于类别众多，不同食物之间的视觉相似性高，以及缺乏用于训练最先进的深度模型的数据集，食物分类是一个具有挑战性的问题。解决此问题将需要计算机视觉模型以及用于评估这些模型的数据集的进步。在本文中，我们将重点放在第二个方面，并介绍FoodX-251，这是一个从网络收集的251个细粒度食品类别的数据集，其中包含158k个图像。我们使用118k图像作为训练集，并为40k图像提供可用于验证和测试的人工验证标签。在这项工作中，我们概述了创建此数据集的过程，并提供了具有深度学习模型的相关基线。FoodX251数据集已在细粒度的视觉分类研讨会 (CVPR 2019的FGVC6) 中用于组织iFood-2019挑战1，可供下载。Food classification is a challenging problem due to the large number of categories, high visual similarity between different food items, and the lack of datasets suitable for training state-of-the-art deep learning models. Addressing this issue requires advancements in both computer vision models and datasets for evaluating these models. In this paper, we focus on the latter aspect and introduce FoodX-251, a dataset of 251 fine-grained food categories collected from the web, containing 158,000 images. We use 118,000 images as the training set, and provide manually verified labels for 40,000 images which can be used for validation and testing. In this work, we outline the process of creating this dataset and present relevant baselines using deep learning models. The FoodX-251 dataset was used to organize the iFood-2019 Challenge 1 at the Fine-Grained Visual Categorization Workshop (FGVC6 of CVPR 2019), and is available for download.

搜集汇总

背景与挑战

背景概述

FoodX-251是一个细粒度食品图像分类数据集，包含251个类别和约158k张图像，其中118k用于训练，40k带有验证标签。该数据集由Google等机构于2019年发布，旨在解决食物分类中类别多、视觉相似性高的挑战，并已用于iFood-2019挑战，为深度学习模型提供基线评估。