ImageNet-ReaL

此数据集包含ILSVRC-2012 (ImageNet) 验证图像，其中添加了一组新的 “重新评估” (真实) 标签，来自 “我们是否使用ImageNet” 论文，请参阅https://arxiv.org/abs/2006.07159使用增强协议收集这些标签，从而产生多标签和更准确的注释。
重要说明: 大约3500个示例不包含标签，在计算准确性时应将其从平均中排除。

This dataset contains the validation images of ILSVRC-2012 (ImageNet), to which a new set of "re-evaluated" (ground-truth) labels has been added. These labels originate from the paper "Do We Use ImageNet?", and were collected via an enhanced annotation protocol detailed in https://arxiv.org/abs/2006.07159, yielding multi-label and more accurate annotations.

Important note: Approximately 3500 examples do not have associated labels, and these should be excluded from the average calculation when computing accuracy.

搜集汇总

背景与挑战

背景概述

ImageNet-ReaL是基于ImageNet验证图像的数据集，通过重新评估添加了多标签和更准确的真实注释，旨在提升图像分类模型的评估准确性。数据集包含约3500个无标签示例，需在计算时排除，适用于计算机视觉研究。