ILSVRC2012

该竞赛的目标是使用大型手工标记的ImageNet数据集的子集 (10,000,000描述10,000对象类别的标记图像) 作为训练来估计用于检索和自动注释目的的照片的内容。测试图像将在没有初始注释的情况下显示-没有分割或标签-算法将必须生成标签，以指定图像中存在哪些对象。新的测试图像将被收集和标记，特别是针对本次比赛，并且不是先前发布的ImageNet数据集的一部分。总体目标是识别图像中存在的主要对象。今年，我们还有一个指定物体位置的检测任务。The objective of this competition is to utilize a subset of the large, manually annotated ImageNet dataset—comprising 10,000,000 labeled images describing 10,000 object categories—as training data to estimate the content of photographs for retrieval and automatic annotation purposes. Test images will be provided without initial annotations, i.e., no segmentations or labels, and algorithms will be required to generate labels specifying which objects are present in the images. The new test images are collected and annotated specifically for this competition and are not part of any previously released ImageNet datasets. The overall goal is to identify the primary objects present in the images. This year, a detection task that requires specifying the locations of the objects will also be included.

搜集汇总

背景与挑战

背景概述

ILSVRC2012是一个大规模图像分类数据集，源自ImageNet项目，包含约1000个类别和大量标记图像，用于训练和评估图像分类与检测算法。该数据集由斯坦福大学和普林斯顿大学于2012年发布，是计算机视觉领域的重要基准，广泛应用于深度学习和模型性能测试。