Face API置信度分数 [英] Face API Confidence scores

问题描述

我正在尝试查找有关Azure如何得出置信度得分以及0.5(相同)的统计概率不正确的更多信息.

I am trying to find more information on how Azure arrives at the confidence score and the statistical probabilities of a 0.5 (identical) being incorrect.

从我阅读的内容看，这取决于开发人员调整此级别以满足您自己的要求，但是让别人对这些分数发表看法或统计数据绝对很有趣.

From what I have read it sounds like its down to the developers to adjust this level to meet your own requirements but it would definitely be interesting to get someone elses view or stats on these scores.

我曾尝试过使用双胞胎，而同卵双胞胎的得分为0.6，这是我所期望的，也有关于Doppelgangers的报道，我看到这些得分在0.5-0.6之间，在Azure的眼中是如此.

I have tried it with twins and got 0.6 for identical twins which is what I expected, there was also a report about Doppelgangers and I have seen 0.5-0.6 scores for these, so identical in the eyes of Azure.

我设法获得0.5-0.6分的一些看似随机的人，从统计学上讲，我找到同一个人的图像的可能性是什么.

How about some seemingly random people that I have managed to get 0.5-0.6 scores for, what statistically is the likelyhood that I have found images of the same person.

如果有人发现了可以使用此服务的AI内部工作的任何地方，那么我非常感谢被指出的消息来源.

If anyone has found anywhere that goes into the inner workings of the AI for this service then I would really appreciate being pointed towards the sources.

谢谢

推荐答案

有关Face API的问题，请使用以下论坛:

For questions on Face API, pls use these forums:

此致，
Jaya

Regards,
Jaya

这篇关于Face API置信度分数的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！