Food2K

Food2K 是一个大型食品识别数据集，包含 2000 个类别和超过 100 万张图像。与现有的食物识别数据集相比，Food2K 在类别和图像上都绕过了它们一个数量级，从而为开发用于食物视觉表示学习的高级模型建立了一个新的具有挑战性的基准。可以进一步探索 Food2K 以使更多与食物相关的任务受益，包括新兴的和更复杂的任务（例如，对食物的营养理解），并且可以预期在 Food2K 上训练的模型作为骨干来提高更多与食物相关的任务的性能。Food2K is a large-scale food recognition dataset that contains 2000 categories and over 1 million images. Compared with existing food recognition datasets, Food2K outnumbers them by an order of magnitude in both categories and image volume, thus establishing a new challenging benchmark for developing advanced models for food visual representation learning. Food2K can be further explored to benefit a wider range of food-related tasks, including emerging and more complex ones such as nutritional understanding of food. Moreover, models trained on Food2K are expected to serve as backbones to improve the performance of more food-related downstream tasks.

搜集汇总

背景与挑战

背景概述

Food2K是一个大型食品识别数据集，包含2000个类别和超过100万张图像，在类别和图像数量上均显著超越现有数据集，为食物视觉表示学习建立了新的挑战性基准。该数据集由中国科学院计算技术研究所和美团集团于2021年发布，旨在支持高级模型开发，并可应用于更复杂的食物相关任务，如营养理解，以提升性能。