result is very bad

[lianNice]

test mpv datasets, but result is very bad?

[ghost]

Hello,

I'll take a look at your codebase if you'll send it along. I suspect something has gone awry with your parser.

-Alex

[xiezhy6]

test mpv datasets, but result is very bad?

Hi,

Can your show the image id in your test cases? I will try to analyze the reason for the bad results.

-Zhenyu

[zzzolo]

Hi. I generated similar results and I found that the wrong pre-processing causes them. Could you please release some samples to understand the pre-processing better?

[miibotree]

Here is my result on MPV test set. The result is also not good. During the preprocess, the official openpose model obtains 25 keypoints, while the authors only use 18 keypoints? Is this preprocess step error that causes the bad results?

[xiezhy6]

obtains 25 keypoints, while the authors only use 18 keypoints? Is this preprocess step error that causes the bad results?

I guess it is the parsing error of the first person that causes such a bad result. Please check whether the inputs of the network are all right.

[BIGJUN777]

I encountered similar issue. any updates? Thanks.

[Zheng-Chong]

This is due to the fact that openpose has two different versions of 18 and 25 keypoints, the code uses the 18 version, just rearrange the 25 points.