about demo accuracy

[thinkingIsMagic]

hello, i was impressed with your sota achivement, but when i tried the demo code, the result seems like not fit the person, am i got mistakes somewhere?

[AngioC]

I got the same error. Did you manage to solve it? @jiandandian2

[ZhengyiLuo]

Sorry this is a known issue: https://github.com/ZhengyiLuo/MEVA#known-issues

I think it has something to do with the scaling with the bounding box, or I am made a mistake during the data preprocessing. Feel free to play around with values like here to see if it can resolve it, but do keep in mind changing that scale might lead to worse performance.

The scale does make visualization a bit weird, but the extracted pose looks good so that's good for just-pose use cases.

I am currently a little busy with my Ph.D. applications, will take a crack at it once I am freer :)

[thinkingIsMagic]

抱歉，这是一个已知问题：https : //github.com/ZhengyiLuo/MEVA#known-issues

我认为这与边界框的缩放有关，或者在数据预处理期间犯了一个错误。请随意使用此处的值来查看它是否可以解决它，但是请记住，更改比例可能会导致性能下降。

比例尺确实使可视化有些怪异，但是提取的姿势看起来不错，因此对于刚刚使用的情况非常有用。

我目前在忙于博士学位。应用程序，一旦我变得更自由，就会遇到麻烦:)

thank u, i'll try

[thinkingIsMagic]

我遇到了同样的错误。您设法解决了吗？@ jiandandian2

Writter has replied.

[beyond0310]

Sorry this is a known issue: https://github.com/ZhengyiLuo/MEVA#known-issues

I think it has something to do with the scaling with the bounding box, or I am made a mistake during the data preprocessing. Feel free to play around with values like here to see if it can resolve it, but do keep in mind changing that scale might lead to worse performance.

The scale does make visualization a bit weird, but the extracted pose looks good so that's good for just-pose use cases.

I am currently a little busy with my Ph.D. applications, will take a crack at it once I am freer :)

Have you found a solution to this problem？

[beyond0310]

我遇到了同样的错误。您设法解决了吗？@ jiandandian2

Writter has replied.

Sorry, how much is appropriate? My effect is not very good。

[davrempe]

I am experiencing the same issue running the demo. @ZhengyiLuo is there any update here? I tried adjusting bbox_scale as suggested, but with no success. Do you think this is just a visualization issue or is the camera predicted by the network off somehow?

[wang-zm18]

Same question!