Meta Omnium

Meta Omnium是由爱丁堡大学和北京邮电大学等机构联合创建的多任务元学习基准数据集，涵盖了识别、关键点定位、语义分割和回归等多个视觉任务。该数据集包含约16万条数据，涉及21个任务和4个视觉领域，旨在评估模型在多样化任务上的泛化能力。创建过程中，研究者们精心设计了数据集的结构，确保其既具有挑战性又便于广泛研究使用。Meta Omnium的应用领域广泛，主要用于解决元学习模型在跨任务和跨领域知识转移的问题，推动元学习研究的进一步发展。

Meta Omnium is a multi-task meta-learning benchmark dataset jointly developed by institutions including the University of Edinburgh and Beijing University of Posts and Telecommunications. It covers a variety of visual tasks such as recognition, keypoint localization, semantic segmentation and regression. The dataset contains approximately 160,000 data samples, spanning 21 tasks and 4 visual domains, and is designed to evaluate the generalization ability of models across diverse tasks. During its development, researchers meticulously designed the dataset structure to ensure it is both challenging and accessible for widespread research use. Meta Omnium has broad application scenarios, primarily used to address the issue of cross-task and cross-domain knowledge transfer for meta-learning models, and to further advance the development of meta-learning research.