coco关键点标注json_使用COCO API评估模型在COCO数据集上的结果

由于最近沉迷写论文导致课外学习不多（不写毕不了业啊！！！），所以也没有新知识来很更新啊，刚好今天使用COCO API来评估自己的结果，就学习和调试过程中遇到的问题进行记录。

目前绝大多数目标检测算法都使用coco数据集进行模型评估。那么如何验证自己的实验结果，或者自己的数据増广策略、训练调参等是否能提升模型的AP？COCO_API提供了便捷的接口，安装使用及评估过程如下

一、COCO API的下载和安装

官方使用说明

COCO API - Common Objects in Context

COCO is a large image dataset designed for object detection, segmentation, person keypoints detection, stuff segmentation, and caption generation. This package provides Matlab, Python, and Lua APIs that assists in loading, parsing, and visualizing the annotations in COCO. Please visit Common Objects in Context for more information on COCO, including for the data, paper, and tutorials. The exact format of the annotations is also described on the COCO website. The Matlab and Python APIs are complete, the Lua API provides only basic functionality.

In addition to this API, please download both the COCO images and annotations in order to run the demos and use the API. Both are available on the project website.
-Please download, unzip, and place the images in: coco/images/
-Please download and place the annotations in: coco/annotations/
For substantially more details on