# Evaluate

* Rank Accuracy：

  将gallery中的图片，按照与probe图片的相似度排序，如果在第X个之前就命中，则rankX命中数+1, rankX命中数/probe图片总数则为rankX acc，rankX有时又名为topK，当gallery中存在junk image时，命中junk image时跳过。
* mAP:

  信息检索中的mAP：

  对probe中的每个图片，计算一个AP，取平均为mAP，AP为P的均值，P的含义为，probe对应的目标中，第k个命中时，前k个结果中的命中率。

  举例：假设有两个主题，主题1有4个相关网页，主题2有5个相关网页。某系统对于主题1检索出4个相关网页，其rank分别为1, 2, 4, 7；对于主题2检索出3个相关网页，其rank分别为1,3,5。对于主题1，平均准确率为(1/1+2/2+3/4+4/7)/4=0.83。对于主题2，平均准确率为(1/1+2/3+3/5+0+0)/5=0.45。则MAP= (0.83+0.45)/2=0.64。

  在market1501的评估代码里有另一种计算mAP的方法（可见于[Liang Zheng的综述](https://arxiv.org/pdf/1610.02984v1.pdf)），是求Precision-Recall曲线的面积。这是用[Pascal VOC 2007(The PASCAL Visual Object Classes (VOC) Challenge)的算法](https://link.springer.com/content/pdf/10.1007%2Fs11263-009-0275-4.pdf)，计算的是Precision-Recal曲线下的面积，与信息检索领域计算每次命中时准确率的平均值不同


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://blog.cweihang.io/ml/papers/reid/evaluate.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.