twisted smpl occuring

[skcskc7]

hi

I have 30 cameras as input datasets and face twited smpl results in some frames.

Actually, I am using easymocap 0.1 version, and in your last issue, I confirmed that it was solved in 0.2 version.

Can you point me to the part of the code where the twisted smpl problem is resolved?

https://user-images.githubusercontent.com/22520046/191878954-9ab57eb8-f446-4b30-baa6-42b6abdcd00f.mp4

[chingswy]

Hello, have you tried the latest version? We keep the same data structure. You can just pull and run it with the same dataset.

[qzane]

Hello, have you tried the latest version? We keep the same data structure. You can just pull and run it with the same dataset.

I have the same issue with the newest code (apps/demo/mv1p.py, 4/10/2023, commit: 26d07c2), any update on this?

I guess this issue comes from that: when you take numerical average between two rotations (177 and -179 degree), you will get -1, while the correct answer should be 179. Here is a solution to this kind of problem: https://stackoverflow.com/a/30129248

Here is a fix. I'm testing the code and will submit a pull request if it solves the issue. Hopefully this will also fix issue #109 #110 Seems that only modifying the interp function doesn't fix the issue.

[chingswy]

Run newest code as this. We don't use interp now.

[qzane]

Run newest code as this. We don't use interp now.

It's not working, this is what I got: Here is how I run the code, let me know if there is anything wrong I did:

1. copy-paste the images folder, intri.yml and extri.yml from my old dataset
2. run openpose: python3 apps/preprocess/extract_keypoints.py ${data} --mode openpose --openpose ${openpose} --hand --face
3. run python3 apps/demo/mocap.py ${data} --work lightstage-dense-smpl --subs_vis 0 1 2 --ranges 0 800 1

I can share my dataset if that helps