情感应用程序参考 [英] Emotion App References

问题描述

情感api的参考是什么？我的意思是，一般来说，"中性"情绪是结果中最重要的部分。例如，当一个人的比率大于一定百分比时，是否存在可以被认定为悲伤（"悲伤"）的比率？或者，如果我删除了'中性'
情绪率，并通过使用剩余的情绪（愤怒，恐惧，悲伤，快乐，惊讶，蔑视，厌恶）进行新的分配，这个结果是否可靠？

What is the reference of emotion api? I mean,generally 'neutral' mood is biggest part of the results.For example, Is there a rate that a person can be identified as sad('sadness') when the rate is bigger than a certain percentage? Or if i removed the 'neutral' mood rate and I made a new rate distribution by using remaining mood(anger,fear,sadness,happiness,surprise,contempt,disgust), would this result be reliable?

谢谢，

推荐答案

在现实世界的应用程序中，大多数面部都是中性的并不少见，幸福第二。大多数人大多数时候都会表现出情绪的中性表达。

It is not uncommon in real world applications that most of the faces are neutral, with happiness second. Most people have neutral expression of emotion most of the time.

我们文档中的官方指南是：

The official guidance in our documentation is:

"在解释结果时从Emotion API中，检测到的情绪应该被解释为具有最高分数的情绪，因为分数被归一化为总和为1。用户可以选择在他们的应用程序中设置更高的置信度阈值，
，具体取决于他们的需求。"

如果中性是最大数字，那么API已经确定面部很可能是中性的。如果你排除中性（如果低于某个阈值，则排除中性），那么你将选择API确定
为第二大的图像。可能是正确的情绪。虽然我们对阈值没有任何特别指导，但如果您选择尝试忽略"中性"，那么结果，您还应该尝试设置一个适合您的方案。

If neutral is the largest number, then the API has determined that the face is mostly likely to be neutral. If you exclude neutral (or exclude neutral if it is below a certain threshold,) then you'll be selecting the image that the API identifies as second most likely to be the correct emotion. While we don't have any particular guidance on thresholds, if you choose to experiment with ignoring "neutral" results, you should also experiment to set one which fits your scenario.

这篇关于情感应用程序参考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！