Met

Met 数据集是艺术品领域中实例级识别 (ILR) 的大规模数据集。它依赖于纽约大都会艺术博物馆 (The Met) 的开放获取收藏来形成训练集，其中包含来自超过 224k 类的约 400k 图像，具有世界级地理覆盖范围和按时间顺序约会的艺术品回到旧石器时代。每个博物馆展品都对应一件独特的艺术品，并定义了自己的类别。训练集呈现长尾分布，超过一半的类由单个图像表示，使其成为少样本学习的特例。

The Met Dataset is a large-scale dataset for instance-level recognition (ILR) in the art domain. It draws its training set from the open-access collections of The Metropolitan Museum of Art (The Met), which contains approximately 400,000 images spanning over 224,000 classes. The dataset features world-class geographic coverage, with artworks chronologically dated back to the Paleolithic Era. Each museum exhibit corresponds to a unique artwork and defines its own class. The training set follows a long-tailed distribution, where more than half of the classes are represented by only a single image, rendering it a special case for few-shot learning.

搜集汇总

背景与挑战

背景概述

Met数据集是一个大规模的艺术品实例级识别数据集，包含约400k图像覆盖224k类别，具有世界级地理覆盖和广泛的时间跨度，适合少样本学习研究。