CIFAR100-LT

该数据集名为CIFAR100，它是一个以对象为中心的平衡分类数据集，包含了属于100个不同类别的微小图像。其长尾版本是通过根据指数函数减少每个类别的训练样本而创建的。在实验中，我们使用了ResNet-32网络架构。该数据集的长尾版本在10到200的范围内变化，以模拟不同程度的失衡情况。所执行的任务是分类。

The dataset is named CIFAR-100, which is an object-centric balanced classification dataset containing tiny images belonging to 100 distinct categories. Its long-tailed variant is created by reducing the number of training samples per class according to an exponential function. In the experiments, the ResNet-32 network architecture was employed. The long-tailed variant of this dataset adjusts the number of training samples per class within the range of 10 to 200 to simulate varying degrees of class imbalance. The task conducted is image classification.

搜集汇总

背景与挑战

背景概述

CIFAR100-LT是CIFAR-100的长尾版本，用于研究类别不平衡问题，支持使用LDAM-DRW方法进行训练。