Microsoft COCO 2017 Object Detection Dataset

This is the full 2017 COCO object detection dataset (train and valid), which is a subset of the most recent 2020 COCO object detection dataset.

COCO is a large-scale object detection, segmentation, and captioning dataset of many object types easily recognizable by a 4-year-old. The data is initially collected and published by Microsoft. The [original source of the data is here](https://cocodataset.org/) and the [paper introducing the COCO dataset is here](https://arxiv.org/pdf/1405.0312.pdf).

此为2017年COCO目标检测数据集的完整版，包括训练集和验证集，它是最新2020年COCO目标检测数据集的一个子集。COCO是一个大规模的目标检测、分割和描述数据集，其中包含多种易于四岁儿童识别的对象类型。该数据集最初由微软收集并发布。数据的原始来源见[此处](https://cocodataset.org/)，介绍COCO数据集的论文见[此处](https://arxiv.org/pdf/1405.0312.pdf)。

搜集汇总

背景与挑战

背景概述

Microsoft COCO 2017目标检测数据集包含121,408张图像，用于目标检测任务，是2017年COCO数据集的完整版本。该数据集由Microsoft发布，涵盖多种易于识别的对象类型，适用于大规模目标检测、分割和标注研究。